Search

Over the years, farmers have continuously faced the issue of planting various crops at locations that are not suitable for the growth of certain crops and that in turn affects the crop yield per season. This is because they still use the traditional method of making these inaccurate predictions. Several literatures have proposed different machine learning techniques for crop yield prediction. However, not many literatures have compared the performance of different machine learning technique in the prediction of crop yield. Hence, in this study examines the performance of three machine learning techniques in predicting crop yield. For this study, soil and climate factors were considered. The system started with a general data cleaning. The cleaning was aimed at preparing the data for classification while the correlation was aimed at determining the relevance of factor to the yield of the farm. The three methods considered include Multi-layer Perceptron, Artificial Neural Network and Long Short-Term Memory. Amongst the three techniques considered, LSTM had the best accuracy of 97%, ANN had 86% and MLP had 87% accuracy respectively. [ABSTRACT FROM AUTHOR]

All man-made flying objects in the sky, ships in the ocean can be regarded as small infrared targets, and the method of tracking them has been received widespread attention in recent years. In search of a further efficient method for infrared small target recognition, we propose a hierarchical attention-guided multiscale aggregation network (HAMANet) in this thesis. The proposed HAMANet mainly consists of a compound guide multilayer perceptron (CG-MLP) block embedded in the backbone net, a spatial-interactive attention module (SiAM), a pixel-interactive attention module (PiAM) and a contextual fusion module (CFM). The CG-MLP marked the width-axis, height-axis, and channel-axis, which can result in a better segmentation effect while reducing computational complexity. SiAM improves global semantic information exchange by increasing the connections between different channels, while PiAM changes the extraction of local key information features by enhancing information exchange at the pixel level. CFM fuses low-level positional information and high-level channel information of the target through coding to improve network stability and target feature utilization. Compared with other state-of-the-art methods on public infrared small target datasets, the results show that our proposed HAMANet has high detection accuracy and a low false-alarm rate. [ABSTRACT FROM AUTHOR]

This paper explores into stock market analysis using computer models to predict stock price changes, particularly focusing on the AAPL dataset from Apple Inc. Accurate stock forecasts are vital for informed investment decisions. By comparing different models—Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)—the study offers a comprehensive performance analysis of deep learning approaches. The use of different deep learning architectures allowed us to capture different aspects of the stock price data and assess their effectiveness in making predictions. The project's findings would enable us to make informed decisions about which model to choose for future stock price prediction tasks, benefiting traders, investors, and other stakeholders in the financial domain. LSTM exhibited the best predictive capability, capturing intricate patterns due to its ability to handle long-term dependencies GRU also performed well. However, the MLP model didn't do as well because it couldn't grasp the complex patterns in the data as effectively as the other two models. So, for forecasting stock prices, the LSTM model proved to be the most accurate and reliable choice. These findings offer valuable insights for financial practitioners aiming to enhance stock price prediction using deep learning techniques, aiding in smarter investment strategies. [ABSTRACT FROM AUTHOR]

To build machine learning-based radiomics models to discriminate between high- (HGGs) and low-grade gliomas (LGGs) and to compare the effectiveness of three-dimensional arterial spin labelling (3D-ASL) to evaluate which is a better method. We retrospectively analysed the magnetic resonance imaging T1WI-enhanced images of 105 patients with gliomas that were pathologically confirmed in our hospital. We divided the patients into a training group and a verification group at a ratio of 8:2; 200 patients from the Brain Tumour Segmentation Challenge 2020 were selected as the test group for image segmentation, feature extraction and screening. We constructed models using multilayer perceptron (MLP), support vector machine, random forest and logistic regression and evaluated their predictive performance. We obtained the mean maximum relative cerebral blood flow (rCBFmax) value from 3D-ASL of 105 patients from the hospital to evaluate its efficacy in discriminating between HGGs and LGGs. In machine learning, the MLP classifier model exhibited the best performance in discriminating between HGGs and LGGs; the areas under the curve obtained by MLP and rCBFmax were 0.968 versus 0.815 (verification group) and 0.981 versus 0.815 (test group), respectively. The machine learning-based MLP classifier model performed better in discriminating between HGGs and LGGs than 3D-ASL. In our study, we found that machine learning-based radiomics models and 3D-ASL were valuable in discriminating between HGGs and LGGs and between them, the machine learning-based MLP model had better diagnostic performance. • We analysed the two methods that have been shown to have good diagnostic performance in previous studies. • We compared their ability to discriminate between HGGs and LGGs. • The aim was to identify a better imaging method for preoperative diagnosis and treatment of patients. • We found that machine learning-based radiomics models and 3D-ASL were valuable in discriminating between HGGs and LGGs. • Between them, the machine learning-based MLP model had better diagnostic performance. [ABSTRACT FROM AUTHOR]

This paper presents a decentralized guidance and target assignment (DGTA) strategy for reconfiguring swarms made up of hundreds of modules that have limited communication capabilities. This strategy provides a decentralized onboard computation method to address both near-optimal target assignment and collision-free trajectory planning for swarms of modules. To effectively estimate the transfer cost between modules and targets for target assignment, a multilayer perceptron (MLP) is developed via the backward generation method. The decentralized target assignment problem is solved by greedy algorithm according to the result of the efficient estimating solver. A decentralized model predictive control approach is proposed to guide swarm modules towards their assigned targets, while avoiding collisions with other modules. The feasibility of this guidance framework has been demonstrated through rigorous theoretical analysis. Numerical experiment confirms that the MLP-based method shows an accurate and real-time advantage for estimating transfer cost. Furthermore, the DGTA strategy has been shown to possess robustness and generalization ability, as evidenced by its successful application in a pinpoint assembly task involving 125 and 182 modules. The simulation results confirm the feasibility of this approach for in-orbit assembly application. [ABSTRACT FROM AUTHOR]

High angular resolution diffusion imaging (HARDI) is a promising method for advanced analysis of brain microstructure. However, comprehensive HARDI analysis requires multiple acquisitions of diffusion images (multi-shell HARDI), which is time consuming and often impractical in clinical settings. This study aimed to establish neural network models that can predict new diffusion datasets from clinically feasible brain diffusion MRI for multi-shell HARDI. The development included 2 algorithms: multi-layer perceptron (MLP) and convolutional neural network (CNN). Both followed a voxel-based approach for model training (70%), validation (15%), and testing (15%). The investigations involved 2 multi-shell HARDI datasets: 1) 11 healthy subjects from the Human Connectome Project (HCP); and 2) 10 local subjects with multiple sclerosis (MS). To assess outcomes, we conducted neurite orientation dispersion and density imaging using both predicted and original data and compared their orientation dispersion index (ODI) and neurite density index (NDI) in different brain tissues with 2 measures: peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). Results showed that both models achieved robust predictions, which provided competitive ODI and NDI, especially in brain white matter. The CNN outperformed MLP with the HCP data on both PSNR (p < 0.001) and SSIM (p < 0.01). With the MS data, the models performed similarly. Overall, the optimized neural networks can help generate non-acquired brain diffusion MRI, which will make advanced HARDI analysis possible in clinical practice following further validation. Enabling detailed characterization of brain microstructure will allow enhanced understanding of brain function in both health and disease. [ABSTRACT FROM AUTHOR]

Two-line element sets (TLE) released by the 18th Space Control Squadron are the most complete source of information for space situational awareness in the public domain. They are used in a wide range of contexts that include navigation of nanosatellites and ground communication. However, the TLE data are known for having significant bias and errors. Predicting the bias of TLE can be a way to correct them automatically and hence optimize the systems that rely on them at low cost, for instance for reentry calculations and mission analysis. This paper shows the performance of a neural network trained on more than 4500 inactive resident space objects (RSO) on 10 months of orbital data, on all types of orbits. It is shown that the state vector errors are corrected by at least 40% for 70% of the TLE and that the residual error distributions are well described by a Student's t-distribution for which covariance elements are defined consistently. The performance of the trained neural network are shown to be similar for multiple active satellites. [ABSTRACT FROM AUTHOR]

In the initial phase of this study, Principal Component Analysis (PCA) was employed to meticulously select noncollinear critical defects associated with construction quality from a comprehensive dataset comprising inspections conducted on 1,015 construction projects. The identified critical components were subsequently categorized into three distinct aspects: Inspection Records and Occupational Safety and Health (Aspect A), Concrete Quality (Aspect B), and Construction Team Quality Management (Aspect C). Subsequently, in the second stage of the study, a Multilayer Perceptron (MLP) network was meticulously trained, utilizing the 23 input variables corresponding to critical defects that had been identified through the PCA process. The training regimen persisted until convergence was achieved, resulting in a well-trained model capable of transforming input data into a probability value indicative of a project's quality level. The MLP model exhibited a remarkable performance, achieving an impressive accuracy rate of 91.3%. These robust evaluation metrics affirm the model's exceptional proficiency in predicting construction quality. The hybrid machine learning approach proposed in this study has proven to be highly effective in extracting valuable insights from inspection data and identifying defects intricately linked to construction quality. This model, with its capacity to accurately forecast construction quality, holds substantial potential for practical application in the construction industry. [ABSTRACT FROM AUTHOR]

Stock price prediction is one of the processes of analyzing and determining stock prices in the future. With technical analysis, future stock price predictions can be predicted through the pattern of fluctuations in the stock price in the past. In this study, the researcher predicts the stock price for the next week using the Deep Learning method, namely the Multilayer Perceptron, and combined with the day-shifting method. To expect the results of this stock, the author also observes the model's usefulness and proposes a Mean Error to Mean Price Ratio (MEMPR) to increase the insights processed by the model. Then to find out the accuracy of stock price predictions for each algorithm, testing is carried out using stock data which consists of new data which is then carried out by a training process to get an absolute error value. The experimental results show that the model can predict stock prices with an R2 metric of 0.995. [ABSTRACT FROM AUTHOR]

Uploading research articles to a database can be a complex process with multiple HTML fields. The complexity of this process makes users less efficient and productive. This study explored the use of the Multilayer Perceptron (MLP) algorithm using language-independent features (count and Boolean features) to create a multi-class text classifier that can classify 10 research article components (e.g., title, authors, abstract, etc.). The text classifier was developed to pave the way for a streamlined manual data entry process for uploading research articles to a database by implementing a single textarea in favor of using multiple HTML fields. The text samples were obtained from multiple sources using web scraping technology, consolidated, cleaned, and standardized. The 12 language-independent features were generated based on the textual formats of sample texts (e.g., count of capitalized letters, digits, punctuation, checking the existence of a URL pattern, etc.). Recursive feature elimination with cross validation (RFECV) was used to determine the optimal number of input features. The hyperparameter values of the model were determined through the grid search technique. A trial-and-error process was conducted to determine the number of hidden layers. The model achieved 95%, 94%, and 95% scores for micro, macro, and weighted average f1-scores, respectively. The model performed well in classifying research article components. However, it is sensitive to textual format (e.g., lower or upper case, punctuation used, etc.). For subsequent research in this area, this study recommends investigating the use of both language-dependent and language-independent features to address the limitations of the current model. [ABSTRACT FROM AUTHOR]

One of JST and JICA's technical cooperation projects, "Smart Transportation Strategy to Realize Thailand 4.0," aims to reduce traffic congestion to simultaneously achieve low carbon emissions and improve citizens' quality of Life (QoL). To achieve this goal, it is necessary to develop an AI-based system that allows users to select a combination of transportation modes that meets the various needs of people of all ages and genders, and to realize this goal, evaluation of QoL from images is required. The project suggests how to improve QoL when people moves from some point to another point and some application software to recommend the time scheduling about how to optimize the QoL in a day. It recommends the highest QoL route when we have several path candidate of roads and further some application is necessary to show how QoL dynamically changes in the route. That is scenario of the QoL project. In the previous research, a method for QoL estimation from images was proposed based on semantic segmentation and MLP (Multi Layer Perceptron) by constructing a unique questionnaire form and collecting data in a car view scene. However, the previous method estimates QoL without considering differences in the attributes of the questionnaire data, so the constructed model does not satisfy the attributes of all people. In the proposed method, multiple subjects are asked to subjectively evaluate QoL in 5 levels for the same driving image data set, and the subjects are grouped by clustering based on the evaluation results. The effectiveness of the proposed method was demonstrated by comparing the MAE (mean absolute error) of the model constructed by the proposed method with that constructed by other subjects in a 5-step cross-validation. [ABSTRACT FROM AUTHOR]

Distributed Denial of Service (DDoS) is a type of attack that leverages many compromised systems or computers, as well as multiple Internet connections, to flood targeted resources simultaneously. A DDoS attack's main purpose is to disrupt website traffic and cause it to crash. As traffic grows over time, detecting a Distributed Denial of Service (DDoS) assault is a challenging task. Furthermore, a dataset containing a large number of features may degrade machine learning's detection performance. Therefore, in machine learning, it is necessary to prepare a relevant list of features for the training phase in order to obtain good accuracy performance. With far too many possibilities, choosing the relevant feature is complicated. This study proposes the Boruta algorithm as a suitable approach to achieve accuracy in identifying the relevant features. To evaluate the Boruta algorithm, multiple classifiers (J48, random forest, naïve bayes, and multilayer perceptron) were used so as to determine the effectiveness of the features selected by the the Boruta algorithm. The outcomes obtained showed that the random forest classifier had a higher value, with a 100% true positive rate, and 99.993% in the performance measure of accuracy, when compared to other classifiers. [ABSTRACT FROM AUTHOR]

Students face problems that might hinder their academic pursuit toward success, problems ranging from trivial matters such as class condition, feeling of the student to severe matters such as family breakdown, economic reasons, and many more. This is a major problem because students shape the future of a nation – which will affect many things in the future. Teachers are looking for an effective way to find what might generally be the best solution for solving certain problems, as each student may face different problems, solving one at a time is not possible with the number of students each year. In this paper, we will try to find factors that might hinder or improve student performance using Pearson correlation between each feature toward the student G3 result. Based on the result, past failures will negatively impact student grades with -0.360415 correlation, and then Mother's Education will positively impact student grades with 0.217147. After finding out which factor affects student grade, we try to predict student grade using ML models to prove whether that factor actually affects student grade. Our MLP 12-Neuron model performs the best with RMSE value of 4.32, followed by Random Forest with RMSE value of 4.52, and finally Decision Tree with RMSE value of 5.69. [ABSTRACT FROM AUTHOR]

Dissection of lymph nodes at the roots of the inferior mesenteric artery (IMAN) should be offered only to selected patients at a major risk of developing IMAN involvement. The aim of this study is to present the first artificial intelligence (AI) models to predict IMAN metastasis risk in the left colon and rectal cancer patients. A total of 2891 patients with descending colon including splenic flexure, sigmoid colon and rectal cancer undergoing major primary surgery and IMAN dissection were included as a study cohort, which was then split into a training set (67%) and a testing set (33%). Feature selection was conducted using the least absolute shrinkage and selection operator (LASSO) regression model. Seven AI algorithms, namely Support Vector Machine (SVM), Logistic Regression (LR), Extreme Gradient Boosting (XGB), Light Gradient Boosting (LGB), Decision Tree Classifier (DTC), Random Forest (RF) classifier, and Multilayer Perceptron (MLP), as well as traditional multivariate LR model were employed to construct predictive models. The optimal hyperparameters were determined with 5 fold cross-validation. The predictive performance of models and the expert surgeon was assessed and compared in the testing set independently. The IMAN involvement incidence was 4.6%. The optimal set of features selected by LASSO included 10 characteristics: neoadjuvant treatment, age, synchronous liver metastasis, synchronous lung metastasis, signet ring adenocarcinoma, neural invasion, lymphovascular invasion, CA199, endoscopic obstruction, T stage evaluated by MRI. The most accurate model derived from MLP showed excellent prediction power with area under the receiver operating characteristic curve (AUROC) of 0.873 and produced 81.0% recognition sensitivity and 82.5% specificity in the testing set independently. In contrast, the judgment of IMAN metastasis by expert surgeon yield rather imprecise and unreliable results with a significantly lower AUROC of 0.509. Additionally, the proposed MLP had the highest net benefits and the largest reduction of unnecessary IMAN dissection without the cost of additional involved IMAN missed. MLP model was able to maintain its prediction accuracy in the testing set better than other models and expert surgeons. Our MLP model could be used to help identify IMA nodal metastasis and to select candidates for individual IMAN dissection. [ABSTRACT FROM AUTHOR]

A method for the detection of plastic anti-personnel mines called " leg breakers " is presented from images of the ground acquired remotely by a drone at heights between 1 and 5 m. The proposal achieved an accuracy of 92.13 % and a sensitivity of 82.4 % when evaluating the performance on 260 test images, a number higher than that used in previous works, and with mines buried in real terrain with the presence of vegetation. The above on images acquired on different dates with different environmental conditions and soil moisture. In addition, it was found that the method was adapted to three different thermographic cameras, being necessary only to adjust the gamma parameter associated with the thermal contrast enhancement technique used. It was corroborated that mines are easier to detect to the extent that they are buried more superficially, and when the flight height of the drone is between 1 and 2 m. Finally, it is highlighted that the proposal has the potential to be executed in processing systems onboard the drone, thanks to its algorithmic simplicity. [ABSTRACT FROM AUTHOR]

Harmonic power flow simulation is the main method to deduce and evaluate the steady-state harmonic content of urban rail flexible traction power supply system (FTPSS). The AC power quality of FTPSS can be optimized by conducting harmonic power flow simulation. However, traditional modeling and simulation methods are inefficient, so that it is difficult to achieve the real-time and accurate deduction of the harmonic power flow. Therefore, a high precision real-time simulation method of harmonic power flow based on multi-layer perceptron (MLP) is proposed in this paper. Firstly, the harmonic model of fully controlled converter is established by adopting MLP, which reduces the complexity of the model and improves the efficiency of harmonic calculation. Subsequently, the high precision data samples are used for the regression training of MLP model to ensure the accuracy of harmonic calculation. Moreover, the high precision real-time simulation of harmonic power flow is realized through the synchronous and interactive solution of MLP model and power flow model. Finally, theoretical analysis and simulation are conducted. The results demonstrate that the steady-state harmonic calculation time of the MLP model is significantly reduced to merely 1.02% of the traditional mechanism model's duration, which enables the harmonic power flow simulation to be effectively integrated into real-time system applications. In addition, the RMSE of the MLP model is only 0.0147, indicating the accuracy of the real-time simulation. The method proposed in this paper can quickly and precisely calculate the harmonic power flow, which is of practical value for harmonic evaluation and optimization of FTPSS. [ABSTRACT FROM AUTHOR]

The problem of the Remaining Useful Life (RUL) prediction, aiming at providing an accurate estimate of the remaining time from the current predicting moment to the complete failure of the device, has gained significant attention from researchers in recent years. In this paper, to overcome the shortcomings of rigid combination for temporal and spatial features in most existing RUL prediction approaches, a spatial–temporal homogeneous feature extractor, named Dual-Mixer model, is firstly proposed. Flexible layer-wise progressive feature fusion is employed to ensure the homogeneity of spatial–temporal features and enhance the prediction accuracy. Secondly, the Feature Space Global Relationship Invariance (FSGRI) training method is introduced based on supervised contrastive learning. This method maintains the consistency of relationships between sample features with their degradation process during model training, simplifying the subsequently regression task in the output layer and improving the model's performance in RUL prediction. Finally, the effectiveness of the proposed method is validated through comparisons with other latest research works on the C-MAPSS dataset. The Dual-Mixer model demonstrates superiority across most metrics, while the FSGRI training method shows an average improvement of 7.00% and 2.41% in RMSE and MAPE, respectively, for all baseline models. Our experiments and model code are publicly available at https://github.com/fuen1590/PhmDeepLearningProjects. [ABSTRACT FROM AUTHOR]

The operational well-being of aircraft-engine turbine components is paramount for engine safety. Monitoring exhaust gas temperature (EGT) serves as a key indicator of their condition. Real-time and precise EGT prediction in aircraft engines plays a pivotal role in ensuring flight safety and effective engine health management. A deep learning model based on the long short-term memory (LSTM), integrated with the physical topology of the aircraft engine, served as the basic prediction model for EGT. Based on Taylor expansion, error compensation was performed on model errors arising from sensor error and performance degradation, three compensation models were developed: a Base model consistent with the basic prediction model, an LSTM model, and a multilayer perceptron model. Their influence on prediction accuracy was examined. Each compensation model was trained using two distinct fusion methods: a global compensation method (GCM) and a real-time compensation method (RtCM). The research also explored how different fusion methods contributed to enhancing prediction accuracy. The results showed that the basic model used in this study achieved high prediction precision. The addition of the compensation model further improved the prediction precision. The GCM effectively reduced the mean absolute relative error (MARE), while the RtCM effectively reduced the maximum absolute relative error (Emax) without increasing the prediction time. The best model evaluated on the four engine test datasets had a MARE value of 0.166%, Emax value of 2.745% and mean absolute error of 1. 13 ° C , indicating high prediction precision. [ABSTRACT FROM AUTHOR]

Recent studies attempted to utilize multilayer perceptrons (MLPs) to solve semi-supervised node classification on graphs, by training a student MLP by knowledge distillation (KD) from a teacher graph neural network (GNN). While previous studies have focused mostly on training the student MLP by matching the output probability distributions between the teacher and student models during KD, it has not been systematically studied how to inject the structural information in an explicit and interpretable manner. Inspired by GNNs that separate feature transformation T and propagation Π , we re-frame the KD process as enabling the student MLP to explicitly learn both T and Π. Although this can be achieved by applying the inverse propagation Π − 1 before distillation from the teacher GNN, it still comes with a high computational cost from large matrix multiplications during training. To solve this problem, we propose Propagate & Distill (P&D) , which propagates the output of the teacher GNN before KD and can be interpreted as an approximate process of the inverse propagation Π − 1. Through comprehensive evaluations using real-world benchmark datasets, we demonstrate the effectiveness of P&D by showing further performance boost of the student MLP. • Knowledge distillation (KD) from GNNs to MLPs has received attention as a new graph learning framework. • In GNN-to-MLP KD approaches, depending solely on the teacher GNN's output disable the student MLP to capture high-order connectivity. • Explicitly injecting additional structural information during the KD process can further improve the performance. [ABSTRACT FROM AUTHOR]

The identification of individuals based on their walking patterns, also known as gait recognition, has garnered considerable interest as a biometric trait. The use of gait patterns for gender classification has emerged as a significant research domain with diverse applications across multiple fields. The present investigation centers on the classification of gender based on gait utilizing data from Ultra-wide band radar. A total of 181 participants were included in the study, and data was gathered using Ultra-wide band radar technology. This study investigates various preprocessing techniques, feature extraction methods, and dimensionality reduction approaches to efficiently process Ultra-wide band radar data. The data quality is improved through the utilization of a two-pulse canceller and discrete wavelet transform. The hybrid feature dataset is generated through the creation of gray-level co-occurrence matrices and subsequent extraction of statistical features. Principal Component Analysis is utilized for dimensionality reduction, and prediction probabilities are incorporated as features for classification optimization. The present study employs k-fold cross-validation to train and assess machine learning classifiers, Decision Tree, Random Forest, Support Vector Machine, Logistic Regression, Multi-Layer Perceptron, K-Nearest Neighbors, and Extra Tree Classifier. The Multilayer Perceptron exhibits superior performance, achieving an accuracy of 0.936. The Support Vector Machine and k-Nearest Neighbors classifiers closely trail behind, both achieving an accuracy of 0.934. This research is of the utmost importance due to its capacity to offer solutions to crucial problems in multiple domains. The findings indicate that the utilization of UWB radar data for gait-based gender classification holds promise in diverse domains, including biometrics, surveillance, and healthcare. The present study makes a valuable contribution to the progress of gender classification systems that rely on gait patterns. [ABSTRACT FROM AUTHOR]

Green roofs have recently been popular in buildings for saving energy. A reliable model is required to include green roofs' role in the thermal performance of buildings. This study applies artificial neural networks (ANN) to estimate the buildings' heat gain through green roofs. By comparing different ANN architectures (feedforward, recurrent, and cascade), the best model to estimate heat flux from four significant design factors (plant height, leaf area index of plant, soil depth, and overall heat transfer coefficient of support layer) is determined. The relevance test clarifies that the overall heat transfer coefficient of the support layer and the leaf area index of the plant have the highest correlation with the target variable. The modeling results prove that the multilayer perceptron (MLP) neural network with 4-5-1 topology and the logarithm sigmoid and liner activation functions in hidden and output layers has the best accuracy for the given task. This model predicts 2700 literature data with the regression coefficient = 0.9999, mean absolute errors = 0.025, and root mean squared errors = 0.035. Our topology-engineered MLP can analyze the effect of design factors on the heat flux through green roofs and help quickly determine its impact on building thermal performance. • Buildings' heat gain through green roofs is simulated by artificial neural networks. • Feedforward topology of the neural network shows the highest accurate predictions. • This model simulates 2700 literature data with the regression coefficient = 0.9999. • Heat gain mostly governs by overall heat transfer coefficient of the support layer. [ABSTRACT FROM AUTHOR]

The real-time on-board component-level model (CLM) of gas turbine engines (GTEs) is an important foundation for fault-tolerant control and health management. However, traditional iterative solving algorithms typically rely on the multiple computations of gas-path thermodynamic parameters, which seriously limits their practical application. In this paper, aiming to improve the real-time performance of CLM, an enhanced non-iterative real-time solver (ENRS) method is proposed. It has combined an input selection strategy, an under-sampling supplemental enhancement training (USET) technique, and a multilayer perceptron (MLP) network. Compared with the existing CLM solving methods, its innovations are as follows: (1) a novel framework for directly solving CLM is proposed, which helps get rid of the dilemma of increased time consumption due to multiple iterations. (2) an enhanced training process is designed for the MLP network, which helps improve the adaptivity to the full flight envelope and all engine states, especially to reduce the maximum error. Simulation tests eventually show that the ENRS effectively improves the real-time performance of CLM in practical applications, and maintains solution accuracy and robustness. • The proposed method can improve the component-level model's real-time performance. • An enhanced training module has improved the adaptivity to various conditions. • Both microprocessor test and full envelope simulation verification are conducted. [ABSTRACT FROM AUTHOR]

• An MLP model is proposed to predict particle size distribution (PSD) in grinding. • The model is initially trained with data generated from empirical equations. • It is then trained with experimental data to improve the prediction accuracy. • The model shows high PSD prediction accuracy, even with less data volume. • The model and related findings have significance for mineral processing. In mineral processing, traditional methods for determining the grinding characteristics of ores, such as single-size-fraction grinding tests, are time-consuming and costly. This study investigated an efficient alternative that uses an artificial neural network (ANN) to predict the particle size distributions (PSDs) of breakage products. A multilayer perceptron was used for this purpose, and its modified form was introduced to improve the prediction accuracy. By using an empirical equation based on a kinetic grinding model, the PSDs of the breakage products were generated for various grinding characteristics and the data were used to train the ANN models. When the PSDs of the breakage product obtained from the experiment were predicted using the trained model, the prediction accuracy was relatively low. Therefore, the model pre-trained with the data generated using the empirical equation was fine-tuned based on the experimental data to improve the prediction accuracy. Consequently, the coefficient of determination between the predicted and true labels for the PSD improved to approximately 0.99. In particular, we were able to successfully predict the experimental data that showed a different trend from that of the pre-training data. This ANN model can be used to flexibly predict the PSD of the breakage product with respect to the top size or distribution of the ore and grinding time. [ABSTRACT FROM AUTHOR]

The February 6, 2023 Kahramanmaraş-Türkiye earthquakes with moment magnitudes 7.7 and 7.6 resulted in substantial casualties, injuries and extensive infrastructure devastation. Soil liquefaction was identified as one of the contributing factors to the damages. In this study, a data-driven approach to assess liquefaction-prone areas within an artificial neural network (MultiLayer Perceptron - MLP) was proposed. The study area, selected to cover a region with the size of 11,500 km2 containing Amik and Kahramanmaraş Plains, is governed mainly by active tectonism of the East Anatolian Fault Zone. The earthquakes were considered to be responsible for numerous liquefaction occurrences in both plains. Here, a comprehensive inventory of liquefied regions was compiled from aerial photogrammetric images taken in the days following the earthquakes. Considering the availability of suitable geospatial datasets, the key factors for liquefaction modeling were selected as distance to streams, land use and land cover, slope, and topographic wetness index, and normalized difference water index (NDWI) and normalized difference vegetation index (NDVI) derived from satellite images taken a few days before the earthquakes. The Holocene unit was used as a mask to perform modeling and prediction within this lithological type. The F1-score and overall accuracy values obtained from the MLP on a test dataset were 80% and 82%, respectively. The study showed that geospatial databases including airborne and satellite image products have great potential for assessing liquefaction hazard at regional scale, which can be used as base data for planning and conducting further field and laboratory studies to improve the accuracy in predictions. • 6 February 2023 Kahramanmaraş Earthquakes (M w 7.7- and 7.6) caused extensive damage. • Liquefaction in loose cohesionless soils contributed to resulting damage. • An extensive liquefaction inventory with 803 areas was compiled. • Regional liquefaction susceptibility map was developed for an area of 11,500 km2. • An overall accuracy of 82% was achieved using an Artificial Neural Network. [ABSTRACT FROM AUTHOR]

In-situ remediation of total petroleum hydrocarbon (TPH) contaminated soils via Fenton oxidation is a promising approach. However, determining the proper injection amount of H 2 O 2 and Fe source over the Fenton reaction in the complex geological conditions for in-situ TPH soil remediation remains a daunting challenge. Herein, we introduced a practical and novel approach using soft computational models, a multilayer perception artificial neural network (MPLNN), for predicting the TPH removal performance. In this study, we conducted 48 sets of TPH removal experiments using Fenton oxidation to determine the TPH removal performance of a wide range of different ground conditions and generated 336 data points. As a result, a negative Pearson correlation coefficient was obtained in the Fe injection mass and the natural presence of Fe mineral in the soil, indicating that the excess of Fe could significantly retarded the TPH removal performance in the Fenton reaction. In addition, the MPLNN model with 6-6-1 training using Scaled conjugate gradient backpropagation (SCG) with tangent sigmoid as the transfer function demonstrated a high accuracy for TPH removal prediction with the correlation determination of 0.974 and mean square error value of 0.0259. The optimized MPLNN model achieved less than 20% error for predicting TPH removal performance in actual TPH-contaminated soil via Fenton oxidation. Hence, the proposed MPLNN can be useful in improving the Fenton oxidation of TPH removal performance in-situ soil remediation. [Display omitted] • 48 sets of Fenton oxidation experiments were conducted for TPH removal, obtaining 336 data points. • TPH removal performance decreases as Fe concentration in soil increases. • A negative Pearson correlation was obtained for Fe mass and TPH removal performance. • Optimized MPLNN model achieved <20% error in predicting actual TPH removal performance. [ABSTRACT FROM AUTHOR]

• Deep learning models are used to predict the ionospheric GPS-VTEC. • Gated recurrent neural network performs better than Long Short-Term Memory and Multilayer perceptron during intense storms. • Deep learning models perform better than that of IRI-Plas 2017 model. This paper provides the application of deep learning models such as Long Short-Term Memory (LSTM) and a recently proposed Gated Recurrent Unit (GRU) in forecasting the ionospheric GPS_VTEC, and compare the performance of the results with that of Multilayer Perceptron (MLP) neural networks, GIM_TEC and the IRI-Plas 2017 models. GPS_VTEC time series data estimated from GPS measurement over low latitude equatorial station MAL2 with (geo.lat - 2.7 0 N a n d L o n g 40.19 0 E) located in Kenya is used in this study. The data span from January 1, 2010 to December 31, 2018 which covers 9 years of solar cycle 24. The data from the year 2010 to 2016 is used for the training, while the year 2017 data is used for validation and finally the data in the year 2018 is used to examine the performance of the models during the testing period. The performance of the models is based on statistical parameters such as root-mean-square error (RMSE) and correlation coefficient (R). The GRU unit shows a correlation coefficient of 0.971 with GPS_VTEC and a prediction error of 2.004 TECU while that of LSTM, MLP, GIM_TEC, and IRI-Plas 2017 models are 0.967, 0.951, 0.832 and 0.774 with a prediction error of 2.055 TECU, 2.336 TECU, 5.913 and 16.183 TECU, respectively. This shows that the predictions of the gated system models are better than the MLP, GIM_TEC and IRI-Plas models. Considering the prediction ability of these models to forecast the GPS-VTEC values under an intense Geomagnetic Storm event, it is observed that the GRU unit can achieve the best prediction accuracy and shows a strong performance for the trend prediction of this event more than the LSTM, MLP, GIM_TEC, and IRI-Plas models. In all, the three deep learning models perform better than the GIM_TEC and the IRI-Plas 2017 model. [ABSTRACT FROM AUTHOR]

• The combination of SAR and optical data is critical to improving LULC mapping in a tropical region with high cloud cover frequency. • Sentinel-1 was the main source of data during the rainy season. • Annual and perennial crops can be better mapped using both optical and SAR data combined. Remote sensing has an important role in the Land Use and Land Cover (LULC) mapping process worldwide. Combining spaceborne optical and microwave data is essential for accurate classification in areas with frequent cloud cover, such as tropical regions. In this study, we investigate the possible improvements, when SAR data is incorporated into the classification process along with optical data. We used MSI/Sentinel-2 and SAR/Sentinel-1 to provide LULC mapping in the Roraima State, Brazil, in 2019. This State is located in a tropical area, where the cloud cover is frequent over the year. Cloud cover becomes substantial, especially during the May-August period when crops are grown. Twenty-nine scenarios involving a combination of optical- and SAR-based features, as well as times of data acquisition, were considered in this study. Our results showed that optical or SAR data used individually are not enough to provide accurate LULC mapping. The best results in terms of overall accuracy (OA) were achieved using metrics of multi-temporal surface reflectance and vegetation index (VI) for optical imagery, and values of backscatter coefficient in different polarizations and their ratios yielding an OA of 86.41 ± 1.74%. Analysis of three periods of data (January to April, May to August, and September to December) used for classification allowed us to identify the optimal period for distinguishing specific classes. When comparing our LULC map with a LULC product derived within the MapBiomas project we observed that our method performed better to map annual and perennial crops and water classes. Our methodology provides a more accurate LULC for the Roraima State, and the proposed technique can be applied to benefit other regions that are affected by persistent cloud cover. [ABSTRACT FROM AUTHOR]

• Investigated various ML algorithms to model PEM electrolyzer performance characteristics. • Employed MLP, SVM, and RF algorithms for predictions of current density and hydrogen flowrate. • The SVM model is the best technique for simulating performance evaluation with high accuracy. • The SVM algorithm produces an MAE of 0.0317 for current density and 0.0671 for hydrogen flowrate. Proton exchange membrane water electrolyzer (PEMWE) is a sustainable energy conversion device that uses electrical energy to oxidize water and convert it into chemical energy (hydrogen and oxygen) and heat. Despite their numerous advantages, such as high current density, efficiency, high-purity hydrogen production, compatibility with renewable energy sources, and compact design, they have not yet matured due to durability, cost, and electrochemical performance deficiencies. Effective assessment of performance and durability is crucial for electrolyzer design and optimization. This study analyzes the electrochemical performance of PEMWE with artificial intelligence approaches using experimental data. Three different machine learning (ML) techniques − Multilayer Perceptron (MLP), Support Vector Machine (SVM), and Random Forest (RF) were trained and tested for predicting the hydrogen flowrate and current density for PEMWE using different input parameters. These input parameters include cell voltage, temperature, torque, and water flowrate. SVM was detected to be the best technique in predicting all output parameters with a Mean Absolute Error (MAE) of 0.0317 and 0.0671 for current density and hydrogen flowrate predictions in the test set, respectively. The study results show that machine-learning algorithms can optimize operational parameters in electrolyzer control systems. [ABSTRACT FROM AUTHOR]

[Display omitted] • The XGBoost preprocessing method is used to improve prediction performance. • The proportion map is used to select the hyperparameters in X-LR and X-MLP models. • The XGBoost preprocessing method has higher accuracy and generalization capability. To improve prediction performance and reduce artifacts in Raman spectra, we developed an eXtreme Gradient Boosting (XGBoost) preprocessing method to preprocess the Raman spectra of glucose, glycerol and ethanol mixtures. To ensure the robustness and reliability of the XGBoost preprocessing method, an X-LR model (which combined XGBoost preprocessing and a linear regression (LR) model) and a X-MLP model (which combined XGBoost preprocessing and a multilayer perceptron (MLP) model) were developed. These two models were used to quantitatively analyze the concentrations of glucose, glycerol and ethanol in the Raman spectra of mixed solutions. The proportion map of hyperparameters was firstly used to narrow down the search range of hyperparameters in the X-LR and the X-MLP models. Then the correlation coefficients (R2), root mean square of calibration (RMSEC), and root mean square error of prediction (RMSEP) were used to evaluate the models' performance. Experimental results indicated that the XGBoost preprocessing method achieved higher accuracy and generalization capability, with better performance than those of other preprocessing methods for both LR and MLP models. [ABSTRACT FROM AUTHOR]

The cardiotoxic effects of various pollutants have been a growing concern in environmental and material science. These effects encompass arrhythmias, myocardial injury, cardiac insufficiency, and pericardial inflammation. Compounds such as organic solvents and air pollutants disrupt the potassium, sodium, and calcium ion channels cardiac cell membranes, leading to the dysregulation of cardiac function. However, current cardiotoxicity models have disadvantages of incomplete data, ion channels, interpretability issues, and inability of toxic structure visualization. Herein, an interpretable deep-learning model known as CardioDPi was developed, which is capable of discriminating cardiotoxicity induced by the human Ether-à-go-go-related gene (hERG) channel, sodium channel (Na_v1.5), and calcium channel (Ca_v1.5) blockade. External validation yielded promising area under the ROC curve (AUC) values of 0.89, 0.89, and 0.94 for the hERG, Na_v1.5, and Ca_v1.5 channels, respectively. The CardioDPi can be freely accessed on the web server CardioDPipredictor (http://cardiodpi.sapredictor.cn/). Furthermore, the structural characteristics of cardiotoxic compounds were analyzed and structural alerts (SAs) can be extracted using the user-friendly CardioDPi-SAdetector web service (http://cardiosa.sapredictor.cn/). CardioDPi is a valuable tool for identifying cardiotoxic chemicals that are environmental and health risks. Moreover, the SA system provides essential insights for mode-of-action studies concerning cardiotoxic compounds. [Display omitted] • Cardiotoxicity of chemical materials pose to public health and the environment. • Addressing cardiotoxic effects of pollutants via explainable deep learning. • Discriminates hERG, Na_v1.5, and Ca_v1.5 channel blockade with high accuracy. • Structural insights via CardioDPi-SAdetector for mode-of-action studies [ABSTRACT FROM AUTHOR]

Traditional portfolio optimization strategies often rely on statistical methods and linear programming tools to achieve a balance between return and risk. Despite their usefulness for portfolio optimization, they cannot efficiently capture the specific differences and complexity of real-world financial markets. Several modern approaches attempt to overcome these limitations by using nonlinear models, machine learning, and advanced risk measures. In this study, we propose a novel strategy for optimizing portfolios that incorporates second-order stochastic dominance constraints and solves them with neural networks. we show that portfolios subject to second-order stochastic dominance constraints outperform their traditional counterparts, especially in tail-risk situations. • The study presents a novel paradigm by SSD constraint and leveraging neural networks. • In tail-risk situations, our portfolios outperform their traditional counterparts. • By this approach, we can enhance returns and make portfolio decisions more adaptive. [ABSTRACT FROM AUTHOR]

In order to achieve efficient and low carbon energy systems, attention has been focused on locally distributed energy systems, but a detailed and accurate supply–demand database is essential for simulating and optimizing such local systems. Demand data are generally not maintained at small scales, and it is common to estimate electricity demand by prorating global data from large energy carriers. However, the accuracy of the proration method is questionable. The present study explores the idea of using machine learning as an alternative to the proration method for downscaling electricity demand data based on a neural network model. The input data consist of four categories: weekday/holiday determination data, temperature data, periodic signal data, and demand sector configuration data. The model shows a satisfying generalization capacity for downscaling data from regional to prefectural and municipal scales while capturing local characteristics that are not accounted with the proration method. • A machine learning model is applied to the downscaling of electricity demand data. • The model is able to capture temperature and sectoral dependencies. • The model seems to infer the consumption patterns of different demand sectors. • This approach relies on widely available data such as energy statistics. [ABSTRACT FROM AUTHOR]

Radiative cooling is a groundbreaking technology for sustainable temperature regulation to combat global warming and the urban heat island effect. Traditional optimization algorithms for cooling system structures are often brute-force and even limited to local optimal solutions. Here, we propose a joint simulation method based on a multilayer perceptron (MLP) and a convolutional neural network (CNN) with position encoding (PE), namely the hybrid MLP-CNN model with PE. This method can significantly reduce simulation time, avoid getting trapped in local optima, and accurately predict the optical response and radiative cooling power of the structure. Applying PE enables the MLP model to learn the relationship between structural parameters and optical response more quickly and effectively, reducing loss during training and more minor root mean square error (RMSE) and mean relative error (MRE) during testing. In addition, the residual network can reduce the problem of overfitting in the network. Compared to using only an MLP, the hybrid MLP-CNN with PE model can effectively reduce loss and achieve more accurate prediction outputs. The hybrid model is much faster than traditional electromagnetic simulations, taking only 1% of the time, and the model dramatically reduces the time needed to predict the optical response of structures and effectively decreases the cost of designing optical devices. Utilizing this hybrid model not only avoids the potential drawbacks of being limited to suboptimal solutions and significantly reduces the simulation time, but also effectively predicts the optical response of the device. This approach can demonstrate remarkable adaptability in optimizing geometric parameters of various optical apparatuses, including polarization converters, beam splitters, metalens, and other two-dimensional nanostructures. • The hybrid model can reduce simulation time and accurately predict the optical response. • The application of position encoding accelerates the learning of the network. • The maximum predicted radiative cooling power by the hybrid model is up to 113 W/m2. [ABSTRACT FROM AUTHOR]

• Decision-making processes of GCS projects in saline aquifers require a multitude of forward reservoir simulations. • DL surrogate models ease the computational cost of GCS management and operation optimization. • A DL-based workflow incorporating dimension reduction methods was proposed. • A novel strategy was proposed for effective dimension reduction and reconstruction of 3D CO2 saturation data. • The results suggest the proposed workflow provides sufficient predictive fidelity across temporal and spatial scales. Deep learning (DL) models are extensively used as surrogate models for high-fidelity simulations of multiphase fluid flow in porous media at large scales, enabling fast forecasts of the spatial–temporal evolution of three-dimensional (3D) state variables in geological carbon storage (GCS). However, training these models in high-dimensional space remains computationally demanding and prone to overfitting because of limited training data. This paper presents a novel workflow to address these challenges by integrating dimension reduction (DR) methods. The proposed workflow employed pre-trained DR models to extract the latent variables of geological models and state variables and utilized the multi-layer perceptron (MLP) for constructing mapping functions between the input and output variables in latent spaces. Subsequently, the pre-trained reconstruction models converted the MLP-predicted latent state variables to their original high-dimensional form. Furthermore, we proposed a novel strategy for the DR and reconstruction of 3D saturation fields to account for the unique data characteristics of sparsity, nonuniformity, and discontinuity. The proposed strategy applied PCA and inverse PCA for 2D average saturation fields and developed a DL-based 3D reconstruction model, leveraging three 2D average saturation fields as input to produce a 3D saturation field as output. The pre-training of DR and reconstruction models and training of MLP models were conducted on 84 Gulf of Mexico (GoM) simulations and evaluated on 12 testing simulations. Each simulation contained 720 monthly time steps, with the first 360 months as the injection period and the rest as the post-injection period. The proposed workflow, incorporating DR and DL models, accurately predicts the normalized 3D pressure fields, achieving mean square error (MSE) of 2.92 × 10−7 compared to the ground truth obtained from a full-physics simulator. Furthermore, the proposed strategy outperformed PCA and convolutional autoencoder (CAE) models on 3D saturation fields, resulting in minor workflow prediction errors with an MSE of 2.93 × 10−5. The results suggest the proposed workflow provides sufficient predictive fidelity across temporal and spatial scales, and enables a speedup of 160 times compared to the full-physics simulator, facilitating improved decision-making and risk assessment for large-scale GCS management in real-time scenarios. [ABSTRACT FROM AUTHOR]

This study introduces a novel and refined approach for generating exceptionally efficient S-boxes. The proposed methodology employs a hybrid approach that combines linear fractional transformation (LFT) with a multilayer perceptron (MLP) architecture. This method makes use of a perceptron with three layers: input, hidden, and output. Each layer's neuron count is fine-tuned to conform to the S-box layout. In addition, a threshold nonlinear transformation is utilized to increase nonlinearity, and a novel algorithm for boosting nonlinearity is introduced. The utilization of both LFT and MLP approaches has led to the development of S-boxes that possess nearly ideal average nonlinearity values, surpassing those that have been presented in literature. Notably, one S-box achieved an exceptional nonlinearity score of 114.50. Furthermore, to demonstrate how well the S-box works, this study also employs it in an image encryption application. [ABSTRACT FROM AUTHOR]

The aim of this work is to unveil the relationships between the socio-demographic profile of tourists attracted by flamenco shows and their opinions and perceptions about the importance of music in general and flamenco in particular as a motivation to travel, their interest and preference for flamenco and flamenco shows. The methodology employed is an artificial neural network to estimate a specific socio-demographic profile of tourists interested in attending flamenco shows, taking as the basis the prearranged input values, that is, the responses to Likert questions gathered through a survey. The network allows these responses to be customised, obtaining a "composite sketch" of the tourists' socio-demographic profile determined based on these responses. • Relation among flamenco tourist profile and its preferences and opinions. • An artificial neural network is used to estimate a profile of flamenco tourist. • A specific set of answers brings a certain profile of potential flamenco attendant. • The local tourism sector can adapt its offer to the profile of these tourists. [ABSTRACT FROM AUTHOR]

This paper proposes a methodology for efficient, accurate, and sustainable waste heat recovery, where the energy needs of an industrial plant allow the installation of thermal engines. In this methodology, pinch analysis, mathematical equations, machine learning models, and an optimization algorithm are combined for the first time. To satisfy the industrial requirements, the selected thermal engines are the steam Rankine cycle, the organic Rankine cycle, and the absorption refrigeration cycle, which are modeled by using multilayer perceptron neural networks. The Non-dominated Sorting Genetic Algorithm-III is used to solve the optimization problem. Moreover, multi-objective trade-offs between economic, environmental, and social aspects are studied. A case study is presented to show the applicability of the proposed methodology. The multilayer perceptron models of the thermal engines were created with high accuracy. Furthermore, the results show that with this methodology it is possible to find the optimal operating conditions of thermal engines and solutions that allow the use of different fuels to fulfill the three objective functions. [Display omitted] • A novel approach to waste heat recovery is presented. • Multilayer perceptron neural networks have been incorporated to model thermal engines. • Multi-objective optimization incorporating economic, environmental, and social aspects. • The multilayer perceptron models are created with high accuracy. • The optimal operating conditions of thermal engines are determined. [ABSTRACT FROM AUTHOR]

This paper uses a Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2) re-analysis to identify long-term Mediterranean Sea Offshore Wind (OW) classification possible locations. In particular, an OW classification based on the last 40-years period OW speeds highlighted the best areas for potential Offshore Wind Turbine Generators (OWTG) installations in the Mediterranean basin. Preliminary, long-term OW classification results show that several Mediterranean basin zones in the Aegean Sea, Gulf of Lyon, the Northern Morocco and Tunisia regions have attractive OW potential. Secondly, a combined forecasting model based on the wavelet decomposition method and long-term memory neural network has been developed to predict the short-term wind speed considering the last ten years of hourly data for Mediterranean areas. The results of the proposed model for wind speed prediction have been compared with other single models, Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM), highlighting a higher level of accuracy. Finally, three Weibull fitting algorithms have been provided to analyze the wind energy potential in the Mediterranean basin. • Offshore wind classification using 40 years of re-analysis data and learning models. • Offshore wind speed assessment and mapping of the Mediterranean Sea hot region's. • A prediction model based on wavelet transform and long short-term memory network. • The evaluated model based on re-analysis data for offshore wind hot region's. [ABSTRACT FROM AUTHOR]

Photovoltaic arrays may suffer from a number of temporary and permanent faults. Partial shading and soiling belong to the former group, while cell cracking and delamination fall within the latter one. In these cases, the shape of the current vs voltage curve around the maximum power point shows features that are different from those ones of an array operating in normal conditions. The shape change should allow triggering, in case of a temporary fault, control actions that might improve the electrical power production. Instead, if the fault is permanent, the identification of the modified shape should activate a procedure for a more in-depth analysis of the problem and a maintenance action. In this paper, the conditions leading to a change in the array behavior during its delivering of the maximum power are examined. The change of curvature of the current vs voltage curve around its maximum power point is suitably detected to trigger the fault mitigation action. The feature is caught through an ensemble of artificial neural networks, which analyzes the current vs voltage curve and classifies the module as healthy or faulty. It is demonstrated that few samples around the maximum power point are required, this meaning that the proposed approach is compatible with the operation of any perturbative maximum power point tracking algorithm and its application does not lead to any power production drop. In addition, the approach does not require neither temperature nor irradiance measurements as inputs. The neural networks are trained through synthetic data, so that their application is not limited to arrays including a specific photovoltaic module. The method is also validated through experimental data. [Display omitted] • Some types of PV faults lead to a change of the shape of the I – V curve around the MPP. • An Artificial Neural Network catches this change to classify a curve as healthy or faulty. • Only few samples around the MPP are required without irradiance or temperature data. • The model is trained with synthetic data being not limited to a specific module. • Best trained neural networks can be combined into an ensemble to improve the results. [ABSTRACT FROM AUTHOR]

[Display omitted] • A neural network is used to detect faults in photovoltaic modules. • According to the shape of the current vs voltage curve, the faults are classified. • The training uses synthetic curves and the test is performed also on experimental data. • The points are normalized and resampled radially to improve the results. • With two hidden layers of 100 neurons each, the hit rate reaches more than 98.5%. In this paper a model-based procedure for fault detection and diagnosis of photovoltaic modules is presented. A four-layered feedforward artificial neural network learns the correlation between the features of the current vs. voltage curve and the environmental variables, which are the irradiance and the temperature. This correlation describes the behavior of the module at normal conditions. Moreover, the effect of anomalous variation of some parameters is learnt and correlated to the shape of the same curve, thus associated to a specific failure mechanism and to some assigned ranges quantifying the fault severity. The neural network is trained by using synthetic curves simulated by employing the single diode model and some well assessed and validated translation formulae. The obtained results over the simulated set of curves with different failures allow to achieve a classification error lower than 1.5%. The proposed approach has been also validated for detecting anomalous increases of the series resistance in a large experimental set of curves; in this case, a classification error of 2.7% has been achieved. [ABSTRACT FROM AUTHOR]

Accurately understanding the thermal conductivities of nanorefrigerants is crucial to the development of heat transfer enhancement technology in applications such as cooling, microelectronics, and refrigeration. In this work, three types of nanorefrigerants, Al 2 O 3 −R141b, TiO 2 −R141b, and SiO 2 −R141b, were prepared and characterized. The effective thermal conductivities of those nanorefrigerants at different nanoparticle concentrations (φ v : 0.05–0.6 vol%), temperature (T : 283–298 K), hydrodynamic size (d a : 48.56–115.92 nm) and nanoparticle thermal conductivities (k p : 1.2–36 W/m·K) was experimentally investigated. For the first time, five intelligent models—the radial basis function (RBF), multilayer perceptron (MLP), least-squares support-vector machine (LSSVM), dendrite morphological neural network (DMNN), and spiking neural network (SNN) models—were developed to predict the effective thermal conductivities of nanorefrigerants based on 288 experimental samples. Through statistical and graphical methods, the performances of these intelligent models and seven commonly used theoretical/empirical correlations were compared. In addition, trend and sensitivity analyses were conducted to evaluate the influence degree of each independent input parameter. The results indicated that the effective thermal conductivity increased with increasing nanoparticle concentration, temperature, and nanoparticle thermal conductivity, while it decreased with increasing hydrodynamic size. The RBF model outperformed the other intelligent and theoretical, and empirical models with the highest accuracy (R 2 = 0.985) and the lowest error (RMSE = 0.0178, MAPE = 1.26%) in the complete dataset. The RBF model accurately captured the physical trends of the effective thermal conductivity as the input parameters were varied. Meanwhile, the nanoparticle concentration was the most influential factor of the four input parameters. [Display omitted] • The maximum enhancement in thermal conductivity is achieved for Al 2 O 3 –R141b. • Concentration has the largest impact on thermal conductivity of nanorefrigerant. • The RBF model outperformed all the other intelligent and empirical models. • The RBF model accurate captured the physical trends as the input parameters varied. [ABSTRACT FROM AUTHOR]

Long-term prediction parameters of polar motion (PM) with high precision have an important significance in understanding Earth geophysical processes in the future. The multilayer perceptron (MLP) is introduced into PM prediction and we try for the first time to combine MLP with autoregressive moving average model (ARMA) and singular spectrum analysis (SSA) for the mid-long-term prediction of PM. The principal components of PM were extracted and predicted by SSA first. Then, the combined MLP and ARMA were used to predict the residual term. The PM predictions were obtained by adding the principal components and the residual predictions. Applying this proposed method, 157 500-day and 12 5-year lead time predictions of PM were made based on IERS 14 C04 product. The results showed that the proposed method performed well in the mid-long-term period of PM predictions, which was superior to IERS Bulletin A. [ABSTRACT FROM AUTHOR]

To ensure milk quality and detect cows with signs of mastitis, visual inspection of milk by prestripping quarters before milking is recommended in many countries. An objective method to find milk changed in homogeneity (i.e., with clots) is to use commercially available inline filters to inspect the milk. Due to the required manual labor, this method is not applicable in automatic milking systems (AMS). We investigated the possibility of detecting and predicting changes in milk homogeneity using data generated by AMS. In total, 21,335 quarter-level milk inspections were performed on 5,424 milkings of 624 unique cows on 4 farms by applying visual inspection of inline filters that assembled clots from the separate quarters during milking. Images of the filters with clots were scored for density, resulting in 892 observations with signs of clots for analysis (77% traces or mild cases, 15% moderate cases, and 8% heavy cases). The quarter density scores were combined into 1 score indicating the presence of clots during a single cow milking and into 2 scores summarizing the density scores in cow milkings during a 30-h sampling period. Data generated from the AMS, such as milk yield, milk flow, conductivity, and online somatic cell counts, were used as input to 4 multilayer perceptron models to detect or predict single milkings with clots and to detect milking periods with clots. All models resulted in high specificity (98–100%), showing that the models correctly classified cow milkings or cow milking periods with no clots observed. The ability to successfully classify cow milkings or cow periods with observed clots had a low sensitivity. The highest sensitivity (26%) was obtained by the model that detected clots in a single milking. The prevalence of clots in the data was low (2.4%), which was reflected in the results. The positive predictive value depends on the prevalence and was relatively high, with the highest positive predictive value (72%) reached in the model that detected clots during the 30-h sampling periods. The misclassification rate for cow milkings that included higher-density scores was lower, indicating that the models that detected or predicted clots in a single milking could better distinguish the heavier cases of clots. Using data from AMS to detect and predict changes in milk homogeneity seems to be possible, although the prediction performance for the definitions of clots used in this study was poor. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANNs) are known to be sensitive to the initial setting of parameters and the network architecture, such as the number of hidden nodes in multilayer perceptron (MLP). In this paper, we focus on a network structure which can help to find the proper number of hidden nodes in MLP. In this structure, so called Multilayer Perceptrons with Hidden Multipliers (MLPHM), each of the hidden nodes is associated with a tunable "gate" multiplier. With a specific regularization term, each gate tends to be opened or closed completely at the end of the training, and finally a pruned network is obtained. To study the fault tolerance and to improve the generalization of MLPHM, a noise-injected training scheme is proposed, with both multiplicative noise and additive noise taken into consideration. The objective functions and convergence theorems of the noise-injected training algorithms are obtained, and the latter have been verified by simulations. Applications to several UCI datasets have demonstrated that the proposed algorithms have efficient pruning ability and superior generalization ability. [ABSTRACT FROM AUTHOR]

In regression analysis under artificial neural networks, the prediction performance depends on determining the appropriate weights between layers. As randomly initialized weights are updated during back-propagation using the gradient descent procedure under a given loss function, the loss function structure can affect the performance significantly. In this study, we considered the distribution error, i.e., the inconsistency of two distributions (those of the predicted values and label), as the prediction error, and proposed weighted empirical stretching (WES) as a novel loss function to increase the overlap area of the two distributions. The function depends on the distribution of a given label, thus, it is applicable to any distribution shape. Moreover, it contains a scaling hyperparameter (β) such that the appropriate parameter value maximizes the common section of the two distributions. To test the function capability, we generated ideal distributed curves (unimodal, skewed unimodal, bimodal, and skewed bimodal) as the labels, and used the Fourier-extracted input data from the curves under a feedforward neural network. In general, WES outperformed loss functions in wide use, and the performance was robust to the various noise levels. The improved results in RMSE for the extreme domain (i.e., both tail regions of the distribution) are expected to be utilized for prediction of abnormal events in non-linear complex systems such as natural disaster and financial crisis. [ABSTRACT FROM AUTHOR]

Monitoring sediment transport is essential for managing and maintaining rivers. Estimation of the sediment load in rivers is fundamental for the study of sediment movement, erosion, and flood control. In the current study, three machine learning models—multi-layer perceptron (MLP), multi-layer perceptron-stochastic gradient descent (MLP-SGD), and gradient boosted tree (GBT)—were utilized to estimate the suspended sediment load (SSL) at the St. Louis (SL) and Chester (CH) stations on the Mississippi River, U.S. Four evaluation criteria including the Correlation Coefficient (CC), Nash Sutcliffe Efficiency (NSE), Scatter Index (SI), and Willmott's Index (WI) were utilized to evaluate the performance of the used models. A sensitivity analysis of the models to the input variables revealed that the current day discharge variable had the most effect on the SSL at both stations, but in the absence of current-day discharge data (Q t), a combination of input parameters including S S L t − 3 , S S L t − 2 , S S L t − 1 , Q t − 3 , Q t − 2 , Q t − 1 can be used to estimate the SSL. The comparative outcomes indicated the high accuracy of MLP-SGD-5 model with a CC of 0.983, SI of 0.254, WI of 0.991, and NSE of 0.967 at station CH and the MLP-SGD-6 model with a CC of 0.933, SI of 0.576, WI of 0.961, and NSE of 0.867, respectively, at station SL. The results of MLP models were improved by SGD optimization. Therefore, the MLP-SGD method is recommended as the most accurate model for SSL estimation. [ABSTRACT FROM AUTHOR]

This study presents a dynamic risk modeling strategy for a hydrocarbon sub-surface production system under a gas lift mechanism. A data-driven probabilistic methodology is employed to conduct a risk analysis. The integrated approach comprises a multilayer perceptron (MLP) – artificial neural network (ANN) model and a Bayesian network (BN) technique. The MLP-ANN model performs the production forecast, and the BN model analyzes dynamic risks (the production response) and evaluates the impact of the sand face pressure on risks. The introduced model offers an effective strategy to avoid production failure and to monitor dynamic risks. The dynamic risk analysis yields predictive outcomes at any production time in the well's production life. It offers field operators an early warning system based on the Bayesian model with prognostic capabilities. The proposed strategy effectively manages production risks and assists in production decision-making, especially in complex production systems. [ABSTRACT FROM AUTHOR]

Low temperature performance determines that the application area of biodiesel is one of the most important parameters for biodiesel. Therefore, establishing a model between biodiesel composition and low temperature performance can help manufacturers easily select or deploy raw materials. In this study, different ML methods were evaluated for the first time to establish a prediction model between biodiesel composition and CFPP. The DT-based methods has good performance in predicting CFPP of biodiesel. The stacking method was shown to have the best prediction results and stability via verification of ET, stacking and MLP methods on the test set. Importance analysis shows that palmitic acid has the greatest influence on DT-based methods. This work determind that the coefficient of determination of stacking method (R 2 >0.90) is significantly improved compared to the existing CFPP prediction model (R 2 =0.87). It provides support for the next step in a wider range of verification and promotion. [Display omitted] • Established a new model between biodiesel composition and CFPP. • Evaluated different machine learning methods for biodiesel CFPP prediction. • The stacking model has the best prediction results with the average R 2 above 0.9. • Palmitic acid content has the greatest influence on CFPP in all the factors. [ABSTRACT FROM AUTHOR]