Search

A coverage prediction model helps network operators find coverage gaps, plan base station locations, evaluate quality of service, and build radio maps for spectrum sharing, interference management, localization, etc. Existing coverage prediction models rely on the height and transmission power of the base station, or the assistance of a path loss model. All of these increase the complexity of large-scale coverage predictions. In this paper, we propose a multi-modal model, DNN-SS, which combines a DNN (deep neural network) and SS (semantic segmentation) to perform coverage prediction for mobile networks. Firstly, DNN-SS filters the samples with a geospatial-temporal moving average filter algorithm, and then uses a DNN to extract numerical features. Secondly, a pre-trained model is used to perform semantic segmentation of satellite images of the measurement area. Thirdly, a DNN is used to extract features from the results after semantic segmentation to form environmental features. Finally, the prediction model is trained on the dataset consisting of numerical features and environmental features. The experimental results on campus show that for random location prediction, the model achieves a RMSE (Root Mean Square Error) of 1.97 dB and a MAE (Mean Absolute Error) of 1.41 dB, which is an improvement of 10.86% and 10.2%, respectively, compared with existing models. For the prediction of a test area, the RMSE and MAE of the model are 4.32 dB and 3.45 dB, respectively, and the RMSE is only 0.22 dB lower than that of existing models. However, the DNN-SS model does not need the height, transmission power, and antenna gain of the base station, or a path loss model, which makes it more suitable for large-scale coverage prediction. [ABSTRACT FROM AUTHOR]

Among the numerous gaze-estimation methods currently available, appearance-based methods predominantly use RGB images as input and employ convolutional neural networks (CNNs) to detect facial images to regressively obtain gaze angles or gaze points. Model-based methods require high-resolution images to obtain a clear eyeball geometric model. These methods face significant challenges in outdoor environments and practical application scenarios. This paper proposes a model-based gaze-estimation algorithm using a low-resolution 3D TOF camera. This study uses infrared images instead of RGB images as input to overcome the impact of varying illumination intensity in the environment on gaze estimation. We utilized a trained YOLOv8 neural network model to detect eye landmarks in captured facial images. Combined with the depth map from a time-of-flight (TOF) camera, we calculated the 3D coordinates of the canthus points of a single eye of the subject. Based on this, we fitted a 3D geometric model of the eyeball to determine the subject's gaze angle. Experimental validation showed that our method achieved a root mean square error of 6.03° and 4.83° in the horizontal and vertical directions, respectively, for the detection of the subject's gaze angle. We also tested the proposed method in a real car driving environment, achieving stable driver gaze detection at various locations inside the car, such as the dashboard, driver mirror, and the in-vehicle screen. [ABSTRACT FROM AUTHOR]

This paper investigates spiking neural networks (SNN) for novel robotic controllers with the aim of improving accuracy in trajectory tracking. By emulating the operation of the human brain through the incorporation of temporal coding mechanisms, SNN offer greater adaptability and efficiency in information processing, providing significant advantages in the representation of temporal information in robotic arm control compared to conventional neural networks. Exploring specific implementations of SNN in robot control, this study analyzes neuron models and learning mechanisms inherent to SNN. Based on the principles of the Neural Engineering Framework (NEF), a novel spiking PID controller is designed and simulated for a 3-DoF robotic arm using Nengo and MATLAB R2022b. The controller demonstrated good accuracy and efficiency in following designated trajectories, showing minimal deviations, overshoots, or oscillations. A thorough quantitative assessment, utilizing performance metrics like root mean square error (RMSE) and the integral of the absolute value of the time-weighted error (ITAE), provides additional validation for the efficacy of the SNN-based controller. Competitive performance was observed, surpassing a fuzzy controller by 5% in terms of the ITAE index and a conventional PID controller by 6% in the ITAE index and 30% in RMSE performance. This work highlights the utility of NEF and SNN in developing effective robotic controllers, laying the groundwork for future research focused on SNN adaptability in dynamic environments and advanced robotic applications. [ABSTRACT FROM AUTHOR]

Soil organic matter (SOM) is an important source of nutrients required during crop growth and is an important component of cultivated soil. In this paper, we studied the possibility of using deep learning methods to establish a multi-feature model to predict SOM content. Moreover, using Nong'an County of Changchun City as the study area, Sentinel-2A remote sensing images were taken as the data source to construct the dataset by using field sampling and image processing. The LeNet-5 convolutional neural network model was chosen as the deep learning model, which was improved based on the basic model. The evaluation metrics were selected as the root mean square error (RMSE) and the coefficient of determination R2. Through comparison, the R2 of the improved model was found to be higher than that of the linear regression method, Support Vector Machines (SVM) (RMSE = 2.471, R2 = 0.4035), and Random Forest (RF) (RMSE = 2.577, R2 = 0.4913). The result shows that: (1) It is feasible to use the multispectral data extracted from remote sensing images for soil organic matter content inversion based on the deep learning model with a minimum RMSE of 2.979 and with the R2 reaching 0.89. (2) The choice of features has an impact on the prediction of the model to a certain extent. After ranking the importance of features, selecting the appropriate number of features for inversion provides better results than full feature inversion, and the computational speed is improved. [ABSTRACT FROM AUTHOR]

The 5G deployment brings forth the usage of Unmanned Aerial Vehicles (UAV) to assist wireless networks by providing improved signal coverage, acting as relays or base-stations. Another technology that could help achieve 5G massive machine-type communications (mMtc) goals is the Long Range Wide-Area Network (LoRaWAN) communication protocol. This paper studied these complementary technologies, LoRa and UAV, through measurement campaigns in suburban, densely forested environments. Downlink and uplink communication at different heights and spreading factors (SF) demonstrate distinct behavior through our analysis. Moreover, a neural network was trained to predict the measured signal-to-noise ratio (SNR) behavior and results compared with SNR regression models. For the downlink scenario, the neural network results show a root mean square error (RMSE) variation between 1.2322–1.6623 dB, with an error standard deviation (SD) less than 1.6730 dB. For the uplink, the RMSE variation was between 0.8714–1.3891 dB, with an error SD less than 1.1706 dB. [ABSTRACT FROM AUTHOR]

This paper proposes a novel inverse method based on the deep convolutional neural network (ConvNet) to extract snow's layer thickness and temperature via passive microwave remote sensing (PMRS). The proposed ConvNet is trained using simulated data obtained through conventional computational electromagnetic methods. Compared with the traditional inverse method, the trained ConvNet can predict the result with higher accuracy. Besides, the proposed method has a strong tolerance for noise. The proposed ConvNet composes three pairs of convolutional and activation layers with one additional fully connected layer to realize regression, i.e., the inversion of snow parameters. The feasibility of the proposed method in learning the inversion of snow parameters is validated by numerical examples. The inversion results indicate that the correlation coefficient ( R 2 ) ratio between the proposed ConvNet and conventional methods reaches 4.8, while the ratio for the root mean square error (RMSE) is only 0.18. Hence, the proposed method experiments with a novel path to improve the inversion of passive microwave remote sensing through deep learning approaches. [ABSTRACT FROM AUTHOR]

The brain–computer interface (BCI) has many applications in various fields. In EEG-based research, an essential step is signal denoising. In this paper, a generative adversarial network (GAN)-based denoising method is proposed to denoise the multichannel EEG signal automatically. A new loss function is defined to ensure that the filtered signal can retain as much effective original information and energy as possible. This model can imitate and integrate artificial denoising methods, which reduces processing time; hence it can be used for a large amount of data processing. Compared to other neural network denoising models, the proposed model has one more discriminator, which always judges whether the noise is filtered out. The generator is constantly changing the denoising way. To ensure the GAN model generates EEG signals stably, a new normalization method called sample entropy threshold and energy threshold-based (SETET) normalization is proposed to check the abnormal signals and limit the range of EEG signals. After the denoising system is established, although the denoising model uses the different subjects' data for training, it can still apply to the new subjects' data denoising. The experiments discussed in this paper employ the HaLT public dataset. Correlation and root mean square error (RMSE) are used as evaluation criteria. Results reveal that the proposed automatic GAN denoising network achieves the same performance as the manual hybrid artificial denoising method. Moreover, the GAN network makes the denoising process automatic, representing a significant reduction in time. [ABSTRACT FROM AUTHOR]

In recent years, greenhouse-based precision agriculture (PA) has been strengthened by utilization of Internet of Things applications and low-power wide area network communication. The advancements in multidisciplinary technologies such as artificial intelligence (AI) have created opportunities to assist farmers further in detecting disease and poor nutrition of plants. Neural networks and other AI techniques need an initial set of measurement campaigns along with extensive datasets as a training set to baseline and evolve different applications. This paper presents LoRaWAN-based greenhouse monitoring datasets over a period of nine months. The dataset has both the network and sensing information from multiple sensor nodes for tomato crops in two different greenhouse environments. The goal is to provide the research community with a dataset to evaluate performance of LoRaWAN inside a greenhouse and develop more efficient PA monitoring techniques. In this paper, we carried out an exploratory data analysis to infer crop growth by analyzing just the LoRaWAN signals and without inclusion of any extra hardware. This work uses a multilayer perceptron artificial neural network to predict the weekly plant growth, trained using RSSI value from sensor data and manual measurement of plant height from the greenhouse. We developed this proof of concept of joint communication and sensing by using generated dataset from the "Proefcentrum Hoogstraten" greenhouse in Belgium. Results for the proposed method yield a root mean square error of 10% in detecting the average plant height inside a greenhouse. In future, we can use this concept of landscape sensing for different supplementary use-cases and to develop optimized methods. [ABSTRACT FROM AUTHOR]

With the automotive industry moving towards automated driving, sensing is increasingly important in enabling technology. The virtual sensors allow data fusion from various vehicle sensors and provide a prediction for measurement that is hard or too expensive to measure in another way or in the case of demand on continuous detection. In this paper, virtual sensing is discussed for the case of vehicle suspension control, where information about the relative velocity of the unsprung mass for each vehicle corner is required. The corresponding goal can be identified as a regression task with multi-input sequence input. The hypothesis is that the state-of-art method of Bidirectional Long–Short Term Memory (BiLSTM) can solve it. In this paper, a virtual sensor has been proposed and developed by training a neural network model. The simulations have been performed using an experimentally validated full vehicle model in IPG Carmaker. Simulations provided the reference data which were used for Neural Network (NN) training. The extensive dataset covering 26 scenarios has been used to obtain training, validation and testing data. The Bayesian Search was used to select the best neural network structure using root mean square error as a metric. The best network is made of 167 BiLSTM, 256 fully connected hidden units and 4 output units. Error histograms and spectral analysis of the predicted signal compared to the reference signal are presented. The results demonstrate the good applicability of neural network-based virtual sensors to estimate vehicle unsprung mass relative velocity. [ABSTRACT FROM AUTHOR]

The phase recovery module is dedicated to acquiring phase distribution information within imaging systems, enabling the monitoring and adjustment of a system's performance. Traditional phase inversion techniques exhibit limitations, such as the speed of the sensor and complexity of the system. Therefore, we propose an indirect phase retrieval approach based on a diffraction neural network. By utilizing non-source diffraction through multiple layers of diffraction units, this approach reconstructs coefficients based on Zernike polynomials from incident beams with distorted phases, thereby indirectly synthesizing interference phases. Through network training and simulation testing, we validate the effectiveness of this approach, showcasing the trained network's capacity for single-order phase recognition and multi-order composite phase inversion. We conduct an analysis of the network's generalization and evaluate the impact of the network depth on the restoration accuracy. The test results reveal an average root mean square error of 0.086 λ for phase inversion. This research provides new insights and methodologies for the development of the phase recovery component in adaptive optics systems. [ABSTRACT FROM AUTHOR]

Gestational diabetes mellitus (GDM) is a subtype of diabetes that develops during pregnancy. Managing blood glucose (BG) within the healthy physiological range can reduce clinical complications for women with gestational diabetes. The objectives of this study are to (1) develop benchmark glucose prediction models with long short-term memory (LSTM) recurrent neural network models using time-series data collected from the GDm-Health platform, (2) compare the prediction accuracy with published results, and (3) suggest an optimized clinical review schedule with the potential to reduce the overall number of blood tests for mothers with stable and within-range glucose measurements. A total of 190,396 BG readings from 1110 patients were used for model development, validation and testing under three different prediction schemes: 7 days of BG readings to predict the next 7 or 14 days and 14 days to predict 14 days. Our results show that the optimized BG schedule based on a 7-day observational window to predict the BG of the next 14 days achieved the accuracies of the root mean square error (RMSE) = 0.958 ± 0.007, 0.876 ± 0.003, 0.898 ± 0.003, 0.622 ± 0.003, 0.814 ± 0.009 and 0.845 ± 0.005 for the after-breakfast, after-lunch, after-dinner, before-breakfast, before-lunch and before-dinner predictions, respectively. This is the first machine learning study that suggested an optimized blood glucose monitoring frequency, which is 7 days to monitor the next 14 days based on the accuracy of blood glucose prediction. Moreover, the accuracy of our proposed model based on the fingerstick blood glucose test is on par with the prediction accuracies compared with the benchmark performance of one-hour prediction models using continuous glucose monitoring (CGM) readings. In conclusion, the stacked LSTM model is a promising approach for capturing the patterns in time-series data, resulting in accurate predictions of BG levels. Using a deep learning model with routine fingerstick glucose collection is a promising, predictable and low-cost solution for BG monitoring for women with gestational diabetes. [ABSTRACT FROM AUTHOR]

There is an increasing trend in the industry of knowing in real-time the condition of their assets. In particular, tool wear is a critical aspect, which requires real-time monitoring to reduce costs and scrap in machining processes. Traditionally, for the purpose of predicting tool wear conditions in machining, mathematical models have been developed to extract the information from the signal of sensors attached to the machines. To reduce the complexity of developing physical models, where an in-depth knowledge of the system being modelled is required, the current trend is to use machine-learning (ML) models based on data from the tool wear. The acoustic emission (AE) technique has been widely used to capture data from and understand the real-time condition of industrial assets such as cutting tools. However, AE signal interpretation and processing is rather complex. One of the most common features extracted from AE signals to predict the tool wear is the counts parameter, defined as the number of times that the amplitude of the signal exceeds a predefined threshold. A recurrent problem of this feature is to define the adequate threshold to obtain consistent wear prediction. Additionally, AE signal bandwidth is rather wide, and the selection of the optimum frequencies band for feature extraction has been pointed out as critical and complex by many authors. To overcome these problems, this paper proposes a methodology that applies multi-threshold count feature extraction at multiresolution level using wavelet packet transform, which extracts a redundant and non-optimal feature map from the AE signal. Next, recursive feature elimination is performed to reduce and optimize the vast number of predicting features generated in the previous step, and random forests regression provides the estimated tool wear. The methodology presented was tested using data captured when turning 19NiMoCr6 steel under pre-established cutting conditions. The results obtained were compared with several ML algorithms such as k-nearest neighbors, support vector machines, artificial neural networks and decision trees. Experimental results show that the proposed method can reduce the predicted root mean squared error by 36.53%. [ABSTRACT FROM AUTHOR]

An ultra-wideband (UWB) localization system is an alternative in a GPS-denied environment. However, a distance measurement with UWB modules using a two-way communication protocol induces an orientation-dependent error. Previous research studied this error by looking at parameters such as the received power and the channel response signal. In this paper, the neural network (NN) method for correcting the orientation-induced distance error without the need to calculate the signal strength, obtain the channel response or know any parameters of the antenna and the UWB modules is presented. The NN method utilizes only the measured distance and the tag orientation, and implements an NN model obtained by machine learning, using measurements at different distances and orientations of the two UWB modules. The verification of the experimental setup with 12 anchors and a tag shows that with the proposed NN method, 5 cm better root mean square error values (RMSEs) are obtained for the measured distance between the anchors and the tag compared to the calibration method that did not include orientation information. With the least-square estimator, 14 cm RMSE in 3D is obtained with the NN model corrected distances, with a 9 cm improvement compared to when raw distances are used. The method produces better results without the need to obtain the UWB module's diagnostics parameters that are required to calculate the received signal strength or channel response, and in this way maintain the minimum packet size for the ranging protocol. [ABSTRACT FROM AUTHOR]

In this paper, asynchronous complex histogram (ACH)-based multi-task artificial neural networks (MT-ANNs), are proposed to realize modulation format identification (MFI), optical signal-to-noise ratio (OSNR) estimation and fiber nonlinear (NL) noise power estimation simultaneously for coherent optical communication. Optical performance monitoring (OPM) is demonstrated with polarization mode multiplexing (PDM), 16 quadrature amplitude modulation (QAM), PDM-32QAM, as well as PDM-star 16QAM (S-16QAM) for the first time. The range of launched power is −3 to −2 dBm with a fiber link of 160–1600 km. Then, the accuracy of MFI reaches 100%. The average root mean square error (RMSE) of OSNR estimation can reach 0.37 dB. The average RMSE of NL noise power estimation can reach 0.25 dB. The results show that the monitoring scheme is robust to the increase of fiber length, and the solution can monitor more optical network parameters with better performance and fewer training data, simultaneously. The proposed ACH MT-ANN has certain reference significance for the future long-haul coherent OPM system. [ABSTRACT FROM AUTHOR]

The normalized differential vegetation index (NDVI) for Landsat is not continuous on the time scale due to the long revisit period and the influence of clouds and cloud shadows, such that the Landsat NDVI needs to be filled in and reconstructed. This study proposed a method based on the genetic algorithm–artificial neural network (GA-ANN) algorithm to reconstruct the Landsat NDVI when it has been affected by clouds, cloud shadows, and uncovered areas by relying on the MODIS characteristics for a wide coverage area. According to the self-validating results of the model test, the RMSE, MAE, and R were 0.0508, 0.0557, and 0.8971, respectively. Compared with the existing research, the reconstruction model based on the GA-ANN algorithm achieved a higher precision than the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) and the flexible space–time data fusion algorithm (FSDAF) for complex land use types. The reconstructed method based on the GA-ANN algorithm had a higher root mean square error (RMSE) and mean absolute error (MAE). Then, the Sentinel NDVI data were used to verify the accuracy of the results. The validation results showed that the reconstruction method was superior to other methods in the sample plots with complex land use types. Especially on the time scale, the obtained NDVI results had a strong correlation with the Sentinel NDVI data. The correlation coefficient (R) of the GA-ANN algorithm reconstruction's NDVI and the Sentinel NDVI data was more than 0.97 for the land use types of cropland, forest, and grassland. Therefore, the reconstruction model based on the GA-ANN algorithm could effectively fill in the clouds, cloud shadows, and uncovered areas, and produce NDVI long-series data with a high spatial resolution. [ABSTRACT FROM AUTHOR]

Due to the growing public awareness of cardiovascular disease (CVD), blood pressure (BP) estimation models have been developed based on physiological parameters extracted from both electrocardiograms (ECGs) and photoplethysmograms (PPGs). Still, in order to enhance the usability as well as reduce the sensor cost, researchers endeavor to establish a generalized BP estimation model using only PPG signals. In this paper, we propose a deep neural network model capable of extracting 32 features exclusively from PPG signals for BP estimation. The effectiveness and accuracy of our proposed model was evaluated by the root mean square error (RMSE), mean absolute error (MAE), the Association for the Advancement of Medical Instrumentation (AAMI) standard and the British Hypertension Society (BHS) standard. Experimental results showed that the RMSEs in systolic blood pressure (SBP) and diastolic blood pressure (DBP) are 4.643 mmHg and 3.307 mmHg, respectively, across 9000 subjects, with 80.63% of absolute errors among estimated SBP records lower than 5 mmHg and 90.19% of absolute errors among estimated DBP records lower than 5 mmHg. We demonstrated that our proposed model has remarkably high accuracy on the largest BP database found in the literature, which shows its effectiveness compared to some prior works. [ABSTRACT FROM AUTHOR]

The objective of this paper is to investigate the potential of sentinel-1 SAR sensor products and the contribution of soil roughness parameters to estimate volumetric residual soil moisture (RSM) in the Upper Blue Nile (UBN) basin, Ethiopia. The backscatter contribution of crop residue water content was estimated using Landsat sensor product and the water cloud model (WCM). The surface roughness parameters were estimated from the Oh and Baghdadi models. A feed-forward artificial neural network (ANN) method was tested for its potential to translate SAR backscattering and surface roughness input variables to RSM values. The model was trained for three inversion configurations: (i) SAR backscattering from vertical transmit and vertical receive (SAR VV) polarization only; (ii) using SAR VV and the standard deviation of surface heights ( h r m s ), and (iii) SAR VV, h r m s , and optimal surface correlation length ( l e f f ). Field-measured volumetric RSM data were used to train and validate the method. The results showed that the ANN soil moisture estimation model performed reasonably well for the estimation of RSM using the single input variable of SAR VV data only. The ANN prediction accuracy was slightly improved when SAR VV and the surface roughness parameters ( h r m s and l e f f ) were incorporated into the prediction model. Consequently, the ANN's prediction accuracy with root mean square error (RMSE) = 0.035 cm3/cm3, mean absolute error (MAE) = 0.026 cm3/cm3, and r = 0.73 was achieved using the third inversion configuration. The result implies the potential of Sentinel-1 SAR data to accurately retrieve RSM content over an agricultural site covered by stubbles. The soil roughness parameters are also potentially an important variable to soil moisture estimation using SAR data although their contribution to the accuracy of RSM prediction is slight in this study. In addition, the result highlights the importance of combining Sentinel-1 SAR and Landsat images based on an ANN approach for improving RSM content estimations over crop residue areas. [ABSTRACT FROM AUTHOR]

In this paper, a novel machine learning based algorithm for water supply pollution source identification is presented built specifically for high performance parallel systems. The algorithm utilizes the combination of Artificial Neural Networks for classification of the pollution source with Random Forests for regression analysis to determine significant variables of a contamination event such as start time, end time and contaminant chemical concentration. The algorithm is based on performing Monte Carlo water quality and hydraulic simulations in parallel, recording data with sensors placed within a water supply network and selecting a most probable pollution source based on a tournament style selection between suspect nodes in a network with mentioned machine learning methods. The novel algorithmic framework is tested on a small (92 nodes) and medium sized (865 nodes) water supply sensor network benchmarks with a set contamination event start time, end time and chemical concentration. Out of the 30 runs, the true source node was the finalist of the algorithm's tournament style selection for 30/30 runs for the small network, and 29/30 runs for the medium sized network. For all the 30 runs on the small sensor network, the true contamination event scenario start time, end time and chemical concentration was set as 14:20, 20:20 and 813.7 mg/L, respectively. The root mean square errors for all 30 algorithm runs for the three variables were 48 min, 4.38 min and 18.06 mg/L. For the 29 successful medium sized network runs the start time was 06:50, end time 07:40 and chemical concentration of 837 mg/L and the root mean square errors were 6.06 min, 12.36 min and 299.84 mg/L. The algorithmic framework successfully narrows down the potential sources of contamination leading to a pollution source identification, start and ending time of the event and the contaminant chemical concentration. [ABSTRACT FROM AUTHOR]

Forecasting vessel flows is important to the development of intelligent transportation systems in the maritime field, as real-time and accurate traffic information has favorable potential in helping a maritime authority to alleviate congestion, mitigate emission of GHG (greenhouse gases) and enhance public safety, as well as assisting individual vessel users to plan better routes and reduce additional costs due to delays. In this paper, we propose three deep learning-based solutions to forecast the inflow and outflow of vessels within a given region, including a convolutional neural network (CNN), a long short-term memory (LSTM) network, and the integration of a bidirectional LSTM network with a CNN (BDLSTM-CNN). To apply those solutions, we first divide the given maritime region into M × N grids, then we forecast the inflow and outflow for all the grids. Experimental results based on the real AIS (Automatic Identification System) data of marine vessels in Singapore demonstrate that the three deep learning-based solutions significantly outperform the conventional method in terms of mean absolute error and root mean square error, with the performance of the BDLSTM-CNN-based hybrid solution being the best. [ABSTRACT FROM AUTHOR]

Due to industrialization and the rising demand for energy, global energy consumption has been rapidly increasing. Recent studies show that the biggest portion of energy is consumed in residential buildings, i.e., in European Union countries up to 40% of the total energy is consumed by households. Most residential buildings and industrial zones are equipped with smart sensors such as metering electric sensors, that are inadequately utilized for better energy management. In this paper, we develop a hybrid convolutional neural network (CNN) with an long short-term memory autoencoder (LSTM-AE) model for future energy prediction in residential and commercial buildings. The central focus of this research work is to utilize the smart meters' data for energy forecasting in order to enable appropriate energy management in buildings. We performed extensive research using several deep learning-based forecasting models and proposed an optimal hybrid CNN with the LSTM-AE model. To the best of our knowledge, we are the first to incorporate the aforementioned models under the umbrella of a unified framework with some utility preprocessing. Initially, the CNN model extracts features from the input data, which are then fed to the LSTM-encoder to generate encoded sequences. The encoded sequences are decoded by another following LSTM-decoder to advance it to the final dense layer for energy prediction. The experimental results using different evaluation metrics show that the proposed hybrid model works well. Also, it records the smallest value for mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) when compared to other state-of-the-art forecasting methods over the UCI residential building dataset. Furthermore, we conducted experiments on Korean commercial building data and the results indicate that our proposed hybrid model is a worthy contribution to energy forecasting. [ABSTRACT FROM AUTHOR]

This paper proposes two deep learning methods for remaining useful life (RUL) prediction of bearings. The methods have the advantageous end-to-end property that they take raw data as input and generate the predicted RUL directly. They are TSMC-CNN, which stands for the time series multiple channel convolutional neural network, and TSMC-CNN-ALSTM, which stands for the TSMC-CNN integrated with the attention-based long short-term memory (ALSTM) network. The proposed methods divide a time series into multiple channels and take advantage of the convolutional neural network (CNN), the long short-term memory (LSTM) network, and the attention-based mechanism for boosting performance. The CNN performs well for extracting features from data with multiple channels; dividing a time series into multiple channels helps the CNN extract relationship among far-apart data points. The LSTM network is excellent for processing temporal data; the attention-based mechanism allows the LSTM network to focus on different features at different time steps for better prediction accuracy. PRONOSTIA bearing operation datasets are applied to the proposed methods for the purpose of performance evaluation and comparison. The comparison results show that the proposed methods outperform the others in terms of the mean absolute error (MAE) and the root mean squared error (RMSE) of RUL prediction. [ABSTRACT FROM AUTHOR]

The gas sweetening process removes hydrogen sulfide (H2S) in an acid gas removal unit (AGRU) to meet the gas sales' specification, known as sweet gas. Monitoring the concentration of H2S in sweet gas is crucial to avoid operational and environmental issues. This study shows the capability of artificial neural networks (ANN) to predict the concentration of H2S in sweet gas. The concentration of N-methyldiethanolamine (MDEA) and Piperazine (PZ), temperature and pressure as inputs, and the concentration of H2S in sweet gas as outputs have been used to create the ANN network. Two distinct backpropagation techniques with various transfer functions and numbers of neurons were used to train the ANN models. Multiple linear regression (MLR) was used to compare the outcomes of the ANN models. The models' performance was assessed using the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2). The findings demonstrate that ANN trained by the Levenberg–Marquardt technique, equipped with a logistic sigmoid (logsig) transfer function with three neurons achieved the highest R2 (0.966) and the lowest MAE (0.066) and RMSE (0.122) values. The findings suggested that ANN can be a reliable and accurate prediction method in predicting the concentration of H2S in sweet gas. [ABSTRACT FROM AUTHOR]

In this article, the interpolation of daily data of global solar irradiation, and the maximum, average, and minimum temperatures were measured. These measurements were carried out in the agrometeorological stations belonging to the Agro-climatic Information System for Irrigation (SIAR, in Spanish) of the Region of Castilla and León, in Spain, through the concept of Virtual Weather Station (VWS), which is implemented with Artificial Neural Networks (ANNs). This is serving to estimate data in every point of the territory, according to their geographic coordinates (i.e., longitude and latitude). The ANNs of the Multilayer Feed-Forward Perceptron (MLP) used are daily trained, along with data recorded in 53 agro-meteorological stations, and where the validation of the results is conducted in the station of Tordesillas (Valladolid). The ANN models for daily interpolation were tested with one, two, three, and four neurons in the hidden layer, over a period of 15 days (from 1 to 15 June 2020), with a root mean square error (RMSE, MJ/m2) of 1.23, 1.38, 1.31, and 1.04, respectively, regarding the daily global solar irradiation. The interpolation of ambient temperature also performed well when applying the VWS concept, with an RMSE (°C) of 0.68 for the maximum temperature with an ANN of four hidden neurons, 0.58 for the average temperature with three hidden neurons, and 0.83 for the minimum temperature with four hidden neurons. [ABSTRACT FROM AUTHOR]

Measuring temperature changes at the body-seat interface has been drawing increased attention from both industrial and scientific fields, due to the increasingly sedentary nature from daily leisure activity to routine work. Although contact measurement is considered the gold standard, it can affect the local micro-environment and the perception of sitting comfort. A non-contact temperature measurement system was developed to determine the interface temperature using data gathered unobtrusively and continuously from an infrared sensor (IRs). System performance was evaluated regarding linearity, hysteresis, reliability and accuracy. Then a healthy participant sat for an hour on low/intermediate density foams with thickness varying from 0.5–8 cm while body-seat interface temperature was measured simultaneously using a temperature sensor (contact) and an IRs (non-contact). IRs data were filtered with empirical mode decomposition and fractal scaling indices before a data-driven artificial neural network was utilized to estimate the contact surface temperature. A strong correlation existed between non-contact and contact temperature measurement (ρ > 0.85) and the estimation results showed a low root mean square error (RMSE) (<0.07 for low density foam and <0.16 for intermediate density foam) and high Nash-Sutcliff efficiency (NSE) values (≈1 for both types of foam materials). [ABSTRACT FROM AUTHOR]

Saybolt color is a standard measurement scale used to determine the quality of petroleum products and the appropriate refinement process. However, the current color measurement methods are mostly laboratory-based, thereby consuming much time and being costly. Hence, we designed an automated model based on an artificial neural network to predict Saybolt color. The network has been built with five input variables, density, kinematic viscosity, sulfur content, cetane index, and total acid number; and one output, i.e., Saybolt color. Two backpropagation algorithms with different transfer functions and neurons number were tested. Mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2) were used to assess the performance of the developed model. Additionally, the results of the ANN model are compared with the multiple linear regression (MLR). The results demonstrate that the ANN with the Levenberg–Marquart algorithm, tangent sigmoid transfer function, and three neurons achieved the highest performance (R2 = 0.995, MAE = 1.000, and RMSE = 1.658) in predicting the Saybolt color. The ANN model appeared to be superior to MLR (R2 = 0.830). Hence, this shows the potential of the ANN model as an effective method with which to predict Saybolt color in real time. [ABSTRACT FROM AUTHOR]

Software sensors are playing an increasingly important role in current vehicle development. Such soft sensors can be based on both physical modeling and data-based modeling. Data-driven modeling is based on building a model purely on captured data which means that no system knowledge is required for the application. At the same time, hyperparameters have a particularly large influence on the quality of the model. These parameters influence the architecture and the training process of the machine learning algorithm. This paper deals with the comparison of different hyperparameter optimization methods for the design of a roll angle estimator based on an artificial neural network. The comparison is drawn based on a pre-generated simulation data set created with ISO standard driving maneuvers. Four different optimization methods are used for the comparison. Random Search and Hyperband are two similar methods based purely on randomness, whereas Bayesian Optimization and the genetic algorithm are knowledge-based methods, i.e., they process information from previous iterations. The objective function for all optimization methods consists of the root mean square error of the training process and the reference data generated in the simulation. To guarantee a meaningful result, k-fold cross-validation is integrated for the training process. Finally, all methods are applied to the predefined parameter space. It is shown that the knowledge-based methods lead to better results. In particular, the Genetic Algorithm leads to promising solutions in this application. [ABSTRACT FROM AUTHOR]

Water stress, which affects yield and wine quality, is often evaluated using the midday stem water potential (Ψstem). However, this measurement is acquired on a per plant basis and does not account for the assessment of vine water status spatial variability. The use of multispectral cameras mounted on unmanned aerial vehicle (UAV) is capable to capture the variability of vine water stress in a whole field scenario. It has been reported that conventional multispectral indices (CMI) that use information between 500-800 nm, do not accurately predict plant water status since they are not sensitive to water content. The objective of this study was to develop artificial neural network (ANN) models derived from multispectral images to predict the Ψstem spatial variability of a drip-irrigated Carménère vineyard in Talca, Maule Region, Chile. The coefficient of determination (R2) obtained between ANN outputs and ground-truth measurements of Ψstem were between 0.56-0.87, with the best performance observed for the model that included the bands 550, 570, 670, 700 and 800 nm. Validation analysis indicated that the ANN model could estimate Ψstem with a mean absolute error (MAE) of 0.1 MPa, root mean square error (RMSE) of 0.12 MPa, and relative error (RE) of -9.1%. For the validation of the CMI, the MAE, RMSE and RE values were between 0.26-0.27 MPa, 0.32-0.34 MPa and -24.2-25.6%, respectively. [ABSTRACT FROM AUTHOR]

In order to guarantee the stable operation of shearers and promote construction of an automatic coal mining working face, an online cutting pattern recognition method with high accuracy and speed based on Improved Ensemble Empirical Mode Decomposition (IEEMD) and Probabilistic Neural Network (PNN) is proposed. An industrial microphone is installed on the shearer and the cutting sound is collected as the recognition criterion to overcome the disadvantages of giant size, contact measurement and low identification rate of traditional detectors. To avoid end-point effects and get rid of undesirable intrinsic mode function (IMF) components in the initial signal, IEEMD is conducted on the sound. The end-point continuation based on the practical storage data is performed first to overcome the end-point effect. Next the average correlation coefficient, which is calculated by the correlation of the first IMF with others, is introduced to select essential IMFs. Then the energy and standard deviation of the reminder IMFs are extracted as features and PNN is applied to classify the cutting patterns. Finally, a simulation example, with an accuracy of 92.67%, and an industrial application prove the efficiency and correctness of the proposed method. [ABSTRACT FROM AUTHOR]

The dramatic increase in the computational facilities integrated with the explainable machine learning algorithms allows us to do fast intrusion detection and prevention at border areas using Wireless Sensor Networks (WSNs). This study proposed a novel approach to accurately predict the number of barriers required for fast intrusion detection and prevention. To do so, we extracted four features through Monte Carlo simulation: area of the Region of Interest (RoI), sensing range of the sensors, transmission range of the sensor, and the number of sensors. We evaluated feature importance and feature sensitivity to measure the relevancy and riskiness of the selected features. We applied log transformation and feature scaling on the feature set and trained the tuned Support Vector Regression (SVR) model (i.e., LT-FS-SVR model). We found that the model accurately predicts the number of barriers with a correlation coefficient (R) = 0.98, Root Mean Square Error (RMSE) = 6.47, and bias = 12.35. For a fair evaluation, we compared the performance of the proposed approach with the benchmark algorithms, namely, Gaussian Process Regression (GPR), Generalised Regression Neural Network (GRNN), Artificial Neural Network (ANN), and Random Forest (RF). We found that the proposed model outperforms all the benchmark algorithms. [ABSTRACT FROM AUTHOR]

Though current remote sensing technologies, especially synthetic aperture radars (SARs), exhibit huge potential for soil moisture content (SMC) retrievals, such technologies also present several performance disadvantages. This study explored the merits of proposing a novel data fusion methodology (partly decision level and partly feature level) for SMC estimation. Initially, individual estimations were derived from three distinct methods: the inversion of an Empirically Adapted Integral Equation Model (EA-IEM) applied to SAR data, the Perpendicular Drought Index (PDI), and the Temperature Vegetation Dryness Index (TVDI) determined from Landsat-8 data. Subsequently, three feature level fusions were performed to produce three different novel salient feature combinations where said features were extracted from each of the previously mentioned methods to be the input of an artificial neural network (ANN). The latter underwent a modification of its performance function, more specifically from absolute error to root mean square error (RMSE). Eventually, all SMC estimations, including the feature level fusion estimation, were fused at the decision level through a novel weight-based estimation. The performance of the proposed system was analysed and validated by measurements collected from three study areas, an agricultural field in Blackwell farms, Guildford, United Kingdom, and two different agricultural fields in Sidi Rached, Tipasa, Algeria. Those measurements contained SMC levels and surface roughness profiles. The proposed SMC estimation system yielded stronger correlations and lower RMSE values than any of the considered SMC estimation methods in the order of 0.38%, 1.4%, and 1.09% for the Blackwell farms, Sidi Rached 1, and Sidi Rached 2 datasets, respectively. [ABSTRACT FROM AUTHOR]

Industrial environments are characterised by the non-lineal and highly complex processes they perform. Different control strategies are considered to assure that these processes are correctly performed. Nevertheless, these strategies are sensible to noise-corrupted and delayed measurements. For that reason, denoising techniques and delay correction methodologies should be considered but, most of these techniques require a complex design and optimisation process as a function of the scenario where they are applied. To alleviate this, a complete data-based approach devoted to denoising and correcting the delay of measurements is proposed here with a two-fold objective: simplify the solution design process and achieve its decoupling from the considered control strategy as well as from the scenario. Here it corresponds to a Wastewater Treatment Plant (WWTP). However, the proposed solution can be adopted at any industrial environment since neither an optimization nor a design focused on the scenario is required, only pairs of input and output data. Results show that a minimum Root Mean Squared Error (RMSE) improvement of a 63.87% is achieved when the new proposed data-based denoising approach is considered. In addition, the whole system performance show that similar and even better results are obtained when compared to scenario-optimised methodologies. [ABSTRACT FROM AUTHOR]

Weather is affected by a complex interplay of factors, including topography, location, and time. For the prediction of temperature in Korea, it is necessary to use data from multiple regions. To this end, we investigate the use of deep neural-network-based temperature prediction model time-series weather data obtained from an automatic weather station and image data from a regional data assimilation and prediction system (RDAPS). To accommodate such different types of data into a single model, a bidirectional long short-term memory (BLSTM) model and a convolutional neural network (CNN) model are chosen to represent the features from the time-series observed data and the RDAPS image data. The two types of features are combined to produce temperature predictions for up to 14 days in the future. The performance of the proposed temperature prediction model is evaluated by objective measures, including the root mean squared error and mean bias error. The experiments demonstrated that the proposed model combining both the observed and RDAPS image data is better in all performance measures for all prediction periods compared with the BLSTM-based model using observed data and the CNN-BLSTM-based model using RDAPS image data alone. [ABSTRACT FROM AUTHOR]

Conventional biomechanical modelling approaches involve the solution of large systems of equations that encode the complex mathematical representation of human motion and skeletal structure. To improve stability and computational speed, being a common bottleneck in current approaches, we apply machine learning to train surrogate models and to predict in near real-time, previously calculated medial and lateral knee contact forces (KCFs) of 54 young and elderly participants during treadmill walking in a speed range of 3 to 7 km/h. Predictions are obtained by fusing optical motion capture and musculoskeletal modeling-derived kinematic and force variables, into regression models using artificial neural networks (ANNs) and support vector regression (SVR). Training schemes included either data from all subjects (LeaveTrialsOut) or only from a portion of them (LeaveSubjectsOut), in combination with inclusion of ground reaction forces (GRFs) in the dataset or not. Results identify ANNs as the best-performing predictor of KCFs, both in terms of Pearson R (0.89–0.98 for LeaveTrialsOut and 0.45–0.85 for LeaveSubjectsOut) and percentage normalized root mean square error (0.67–2.35 for LeaveTrialsOut and 1.6–5.39 for LeaveSubjectsOut). When GRFs were omitted from the dataset, no substantial decrease in prediction power of both models was observed. Our findings showcase the strength of ANNs to predict simultaneously multi-component KCF during walking at different speeds—even in the absence of GRFs—particularly applicable in real-time applications that make use of knee loading conditions to guide and treat patients. [ABSTRACT FROM AUTHOR]

Anomaly detection is of great significance in condition-based maintenance of power plant equipment. The conventional fixed threshold detection method is not able to perform early detection of equipment abnormalities. In this study, a general anomaly detection framework based on a long short-term memory-based autoencoder (LSTM-AE) network is proposed. A normal behavior model (NBM) is established to learn the normal behavior patterns of the operating variables of the equipment in space and time. Based on the similarity analysis between the NBM output distribution and the corresponding measurement distribution, the Mahalanobis distance (MD) is used to describe the overall residual (OR) of the model. The reasonable range is obtained using kernel density estimation (KDE) with a 99% confidence interval, and the OR is monitored to detect abnormalities in real-time. An induced draft fan is chosen as a case study. Results show that the established NBM has excellent accuracy and generalizability, with average root mean square errors of 0.026 and 0.035 for the training and test data, respectively, and average mean absolute percentage errors of 0.027%. Moreover, the abnormal operation case shows that the proposed framework can be effectively used for the early detection of abnormalities. [ABSTRACT FROM AUTHOR]

Internet of Things is advancing, and the augmented role of smart navigation in automating processes is at its vanguard. Smart navigation and location tracking systems are finding increasing use in the area of the mission-critical indoor scenario, logistics, medicine, and security. A demanding emerging area is an Indoor Localization due to the increased fascination towards location-based services. Numerous inertial assessments unit-based indoor localization mechanisms have been suggested in this regard. However, these methods have many shortcomings pertaining to accuracy and consistency. In this study, we propose a novel position estimation system based on learning to the prediction model to address the above challenges. The designed system consists of two modules; learning to prediction module and position estimation using sensor fusion in an indoor environment. The prediction algorithm is attached to the learning module. Moreover, the learning module continuously controls, observes, and enhances the efficiency of the prediction algorithm by evaluating the output and taking into account the exogenous factors that may have an impact on its outcome. On top of that, we reckon a situation where the prediction algorithm can be applied to anticipate the accurate gyroscope and accelerometer reading from the noisy sensor readings. In the designed system, we consider a scenario where the learning module, based on Artificial Neural Network, and Kalman filter are used as a prediction algorithm to predict the actual accelerometer and gyroscope reading from the noisy sensor reading. Moreover, to acquire data, we use the next-generation inertial measurement unit, which contains a 3-axis accelerometer and gyroscope data. Finally, for the performance and accuracy of the proposed system, we carried out numbers of experiments, and we observed that the proposed Kalman filter with learning module performed better than the traditional Kalman filter algorithm in terms of root mean square error metric. [ABSTRACT FROM AUTHOR]

Predicting the results of soccer competitions and the contributions of match attributes, in particular, has gained popularity in recent years. Big data processing obtained from different sensors, cameras and analysis systems needs modern tools that can provide a deep understanding of the relationship between this huge amount of data produced by sensors and cameras, both linear and non-linear data. Using data mining tools does not appear sufficient to provide a deep understanding of the relationship between the match attributes and results and how to predict or optimize the results based upon performance variables. This study aimed to suggest a different approach to predict wins, losses and attributes' sensitivities which enables the prediction of match results based on the most sensitive attributes that affect it as a second step. A radial basis function neural network model has successfully weighted the effectiveness of all match attributes and classified the team results into the target groups as a win or loss. The neural network model's output demonstrated a correct percentage of win and loss of 83.3% and 72.7% respectively, with a low Root Mean Square training error of 2.9% and testing error of 0.37%. Out of 75 match attributes, 19 were identified as powerful predictors of success. The most powerful respectively were: the Total Team Medium Pass Attempted (MBA) 100%; the Distance Covered Team Average in zone 3 (15–20 km/h; Zone3_TA) 99%; the Team Average ball delivery into the attacking third of the field (TA_DAT) 80.9%; the Total Team Covered Distance without Ball Possession (Not in_Poss_TT) 76.8%; and the Average Distance Covered by Team (Game TA) 75.1%. Therefore, the novel radial based function neural network model can be employed by sports scientists to adapt training, tactics and opposition analysis to improve performance. [ABSTRACT FROM AUTHOR]

It is important to correctly predict the microclimate of a greenhouse for control and crop management purposes. Accurately forecasting temperatures in greenhouses has been a focus of research because internal temperature is one of the most important factors influencing crop growth. Artificial Neural Networks (ANNs) are a powerful tool for making forecasts. The purpose of our research was elaboration of a model that would allow to forecast changes in temperatures inside the heated foil tunnel using ANNs. Experimental research has been carried out in a heated foil tunnel situated on the property of the Agricultural University of Krakow. Obtained results have served as data for ANNs. Conducted research confirmed the usefulness of ANNs as tools for making internal temperature forecasts. From all tested networks, the best is the three-layer Perceptron type network with 10 neurons in the hidden layer. This network has 40 inputs and one output (the forecasted internal temperature). As the networks input previous historical internal temperature, external temperature, sun radiation intensity, wind speed and the hour of making a forecast were used. These ANNs had the lowest Root Mean Square Error (RMSE) value for the testing data set (RMSE value = 3.7 °C). [ABSTRACT FROM AUTHOR]

Compressive sensing (CS) spectroscopy is well known for developing a compact spectrometer which consists of two parts: compressively measuring an input spectrum and recovering the spectrum using reconstruction techniques. Our goal here is to propose a novel residual convolutional neural network (ResCNN) for reconstructing the spectrum from the compressed measurements. The proposed ResCNN comprises learnable layers and a residual connection between the input and the output of these learnable layers. The ResCNN is trained using both synthetic and measured spectral datasets. The results demonstrate that ResCNN shows better spectral recovery performance in terms of average root mean squared errors (RMSEs) and peak signal to noise ratios (PSNRs) than existing approaches such as the sparse recovery methods and the spectral recovery using CNN. Unlike sparse recovery methods, ResCNN does not require a priori knowledge of a sparsifying basis nor prior information on the spectral features of the dataset. Moreover, ResCNN produces stable reconstructions under noisy conditions. Finally, ResCNN is converged faster than CNN. [ABSTRACT FROM AUTHOR]

The presented manuscript reports the simultaneous detection of a ternary mixture of the benzodiazepines diazepam, lorazepam, and flunitrazepam using an array of voltammetric sensors and the electronic tongue principle. The electrodes used in the array were selected from a set of differently modified graphite epoxy composite electrodes; specifically, six electrodes were used incorporating metallic nanoparticles of Cu and Pt, oxide nanoparticles of CuO and WO3, plus pristine electrodes of epoxy-graphite and metallic Pt disk. Cyclic voltammetry was the technique used to obtain the voltammetric responses. Multivariate examination using Principal Component Analysis (PCA) justified the choice of sensors in order to get the proper discrimination of the benzodiazepines. Next, a quantitative model to predict the concentrations of mixtures of the three benzodiazepines was built employing the set of voltammograms, and was first processed with the Discrete Wavelet Transform, which fed an artificial neural network response model. The developed model successfully predicted the concentration of the three compounds with a normalized root mean square error (NRMSE) of 0.034 and 0.106 for the training and test subsets, respectively, and coefficient of correlation R ≥ 0.938 in the predicted vs. expected concentrations comparison graph. [ABSTRACT FROM AUTHOR]

The main purpose of this study is to investigate the performance of two radar backscattering models; the calibrated integral equation model (CIEM) and the modified Dubois model (MDB) over an agricultural area in Karaj, Iran. In the first part, the performance of the models is evaluated based on the field measurement and the mentioned backscattering models, CIEM and MDB performed with root mean square error (RMSE) of 0.78 dB and 1.45 dB, respectively. In the second step, based on the neural networks (NNS), soil surface moisture is estimated using the two backscattering models, based on neural networks (NNs), from single polarization Sentinel-1 images over bare soils. The inversion results show the efficiency of the single polarized data for retrieving soil surface moisture, especially for VV polarization. [ABSTRACT FROM AUTHOR]

Phosphate is a key element affecting plant growth. Therefore, the accurate determination of phosphate concentration in hydroponic nutrient solutions is essential for providing a balanced set of nutrients to plants within a suitable range. This study aimed to develop a data fusion approach for determining phosphate concentrations in a paprika nutrient solution. As a conventional multivariate analysis approach using spectral data, partial least squares regression (PLSR) and principal components regression (PCR) models were developed using 56 samples for calibration and 24 samples for evaluation. The R2 values of estimation models using PCR and PLSR ranged from 0.44 to 0.64. Furthermore, an estimation model using raw electromotive force (EMF) data from cobalt electrodes gave R2 values of 0.58–0.71. To improve the model performance, a data fusion method was developed to estimate phosphate concentration using near infrared (NIR) spectral and cobalt electrochemical data. Raw EMF data from cobalt electrodes and principle component values from the spectral data were combined. Results of calibration and evaluation tests using an artificial neural network estimation model showed that R2 = 0.90 and 0.89 and root mean square error (RMSE) = 96.70 and 119.50 mg/L, respectively. These values are sufficiently high for application to measuring phosphate concentration in hydroponic solutions. [ABSTRACT FROM AUTHOR]

Oceanographic remote sensing, which is based on the sensitivity of reflected signals from the Global Navigation Satellite Systems (GNSS), so-called GNSS-Reflectometry (GNSS-R), is very useful for the observation of ocean wind speed. Wind speed estimation over the ocean is the core factor in maritime transportation management and the study of climate change. The main concept of the GNSS-R technique is using the different times between the reflected and the direct signals to measure the wind speed and wind direction. Accordingly, this research proposes a novel technique for wind speed estimation involving the integration of an artificial neural network and the particle filter based on a theoretical model. Moreover, particle swarm optimization was applied to find the optimal weight and bias of the artificial neural network, in order to improve the accuracy of the estimation result. The observation dataset of the reflected signal information from BeiDou Geostationary Earth Orbit (GEO) satellite number 4 was used as an input for the estimation model. The data consisted of two phases with I and Q components. Two periods of BeiDou data were selected, the first period was from 3 to 8 August 2013 and the second period was from 12 to 14 August 2013, which corresponded to events from the typhoon Utor. The in situ wind speed measurement collected from the buoy station was used to validate the results. A coastal experiment was conducted at the Yangjiang site located in the South China Sea. The results show the ability of the proposed technique to estimate wind speed with a root mean square error of approximately 1.9 m/s. [ABSTRACT FROM AUTHOR]