Search

This paper presents a novel deep learning-based model for forecasting water demand Specifically, a transformer network architecture-based iTransformer model is introduced to forecast total water demand at both country and province levels over the medium term. Comparative evaluations with Transformer, PatchTST, and LSTM models are conducted across various forecasting lengths, with hyperparameter optimization performed through grid search. The optimal model and parameters are then applied to historical water demand data from 2000 to 2023, yielding forecasts for subsequent years. Results demonstrate that the iTransformer model achieves the lowest RMSE (92.72/1.39/22.71/21.69/9.16), MAE (68.65/1.11/17.42/13.38/5.85), and MAPE (0.01/0.28/0.03/0.08/0.01) in forecasting water demand for China, Beijing, Jiangsu, Zhejiang, and Guangdong respectively. The study emphasizes the importance of considering population size and economic activity in managing socio-economic water demand in China, advocating for a balanced approach to water resource utilization. While the research offers valuable insights for water management authorities, challenges remain in quantifying future water allocations and refining prediction methodologies for enhanced accuracy. Nonetheless, the study paves the way for future research in advancing water demand forecasting methodologies.

In order to improve the light extraction efficiency (LEE) of a plane AlGaAs-based LED (Device A), three different surface periodic micron-scale arrays, micron pillar (Device B), micron truncated cone (Device C) and micron cone (Device D), are prepared by the conventional ultra-violet (UV) lithography. The morphology of the three devices B–D is characterized by a scanning electron microscopy (SEM). The periodic array structures improve the photon escape probabilities. In comparison to the Device A, the light output powers (LOPs) of the Devices B–D increase 54.0%, 145.5% and 157.2% at an injection current of 500 mA, respectively. Moreover, we use the Monte Carlo ray-tracing method to simulate the LEE of the four devices. The energy distribution of the Devices B-D is more concentrated in the middle region in contrast with the Device A.

Organic shales usually present significant heterogeneities in rock textures and reservoir properties due to differing kerogen contents and morphologies, subsequently impacting shale elastic properties and acoustic responses. Numerical upscaling of digital organic shales to evaluate effective elastic properties and acoustic responses has important implications for source rocks and unconventional reservoir characterization. We propose a modeling framework that includes the multiscale reconstruction of kerogen distributions, the numerical modeling of effective elastic properties, and the acoustic response to evaluate the contribution of organic matter. Based on digitized images of the microstructure of Longmaxi black shale samples obtained by X-ray CT, the kerogen components are identified and decomposed into different-level slices in terms of organic matter sizes and morphologies. Multiscale random media reconstruction is applied to these kerogen slices, with synthetic kerogen distributions validated by original counterparts. A finite-element method is used to model the effective elastic properties of digital organic shales, by which we investigate the effect of different kerogen contents and organic matter morphologies. We use a rotated staggered-grid finite-difference method to simulate ultrasonic wave propagation in digital organic shales to evaluate the response of different kerogen contents and organic matter morphologies. Numerical examples show that the multiscale random media method can be applicable to natural organic shales for the reconstruction of kerogen distributions. The elastic properties mainly depend on kerogen contents, with less influence by organic matter morphologies. The ultrasonic scattering effects become stronger for higher kerogen contents with smaller rounding coefficients. Our results confirm the applicability of the proposed modeling framework to support unconventional reservoir characterization. The purpose of this study is to provide the possibility of indicating the sweet point of shale.

The human microbiome is comprised of thousands of microbial species. These species will substantially influence the normal physiology of humans and cause numerous diseases. Microbiome data can be measured by sequencing, microarray, or other technologies. With the fast development of these technologies, downstream analysis methods should also be designed to effectively and accurately discover the valuable information that is hidden in the data. Many methods have been designed for the count data of microbiome. However, to our knowledge, there are only a few methods developed for the continuous microbiome data. Many microbiome data have an over-dispersed and zero-inflated data structure. Traditional methods rarely characterize this data structure and only focus on the differences in the abundance between different samples. In this study, we introduce a novel method, the zero-inflated gamma (ZIG) omnibus test, to specifically test the continuous and zero-inflated microbiome data. In this test, abundance will be tested along with zero prevalence and dispersion. We compared this method with five other popular methods. We found that ZIG omnibus test has significantly higher power and a similar or lower false-positive rate than the competing methods in the tests of simulated data. It also found more proved microbiomes in the real data with tonsil cancer. So, we conclude that ZIG omnibus test is a robust method across various biological conditions in the differential expression test of microbiome data.

With the simulation calibration for negative capacitor considering Landau model and multi-domain (MD) effect, MD MFIS negative capacitance field-effect transistor (NCFET) was thoroughly established for performing the simultaneous analysis of multi-variables (ferroelectric layer thickness ( ${T} _{\mathrm{ FE}}$ ), oxide layer thickness ( ${T} _{\mathrm{ OX}}$ ) and gate length ( ${L} _{\mathrm{ g}}$ )) effect on the device performance. In this study, subthreshold swing (SS) and hysteresis properties of MD-MFIS-NCFET were demonstrated by employing TCAD simulation tool. Compared with the previous reported study on single variable effect based on single-domain (SD) NCFET, the simultaneous analysis of multi-variables effect on MD-NCFET enabled to obtain better device performance and generate more comprehensive results. Convincing models were established based on the experimental data by calibration. Demonstration on the basic simulated results including the lowering SS mechanism and the multi-variables effect on MD-NCFET performance was completely presented based on the capacitance matching theory and short channel effect. With the optimal ${T} _{\mathrm{ FE}}$ and ${T} _{\mathrm{ OX}}$ , a trade-off mechanism between the SS and ${L} _{\mathrm{ g}}$ was shown with the consideration of ${L} _{\mathrm{ g}}$ scaling. Noticeable in-depth study in association with the simultaneous analysis of the multi-variables effect was carried out, indicating that the hysteresis-free SS obtained by simultaneous analysis of multi-variables was lower than that obtained by single-variable analysis. Final validation results demonstrate that the optimization proposed in this work by considering the multi-variable effect shows high compatibility with other NCFET devices, providing an instructive strategy for the high-performance NCFET optimization.

In this work, we demonstrated a HfO2-based ferroelectric field-effect transistor (FeFET) in series with a HfAlO ferroelectric capacitor for memory application and further investigated the impact of the ferroelectric capacitor with different thicknesses and areas. It was revealed that the memory window of the FeFET has a significant correlation with the ferroelectric capacitor from the transfer curves of the transistor after the series connection. By decreasing the thickness and area of the ferroelectric capacitor, the memory window of the FeFET was improved, which could be verified through the TCAD simulation tool and capacitance matching model. With the series-connected ferroelectric capacitor, in addition to optimizing the characteristics of ferroelectric memory, we can effectively avoid the interface-caused undesirable effects in conventional ferroelectric memory during film stacking. These results provide a solution for future low-power embedded memory application.

The traditional pulsar period estimation method using a single distorted profile averts multiple epoch folding processes, but still has space for accuracy to improve. To give consideration to both accuracy and computational load, we combine the fast butterfly epoch folding (FBEF) with the single distorted profile (SDP)-based pulsar period estimation, and propose the FBEF-based pulsar period estimation with a few distorted profiles (FBEF-FDP), which merely uses a few epoch folding processes to enhance accuracy sharply. Firstly, the fast butterfly epoch folding is used to produce a few distorted pulsar profiles with different periods. And then the single distorted profile-based pulsar period estimation explores each distorted prolife to provide a corresponding distortion degree. Finally, using these distortion degree values, the nonlinear search method for multiple minimums can get an optimal distortion degree estimation. Besides, we prove that distortion degree estimation errors of different accumulative distorted profiles from the same pulsar photon sequence are incompletely correlated. This is the theorical basis that the distortion degree estimation values from the same pulsar photon sequence can be fitted to obtain more highly accurate distortion degree estimation. The computational complexity demonstrates that the computational load increases with the number of groups. Thus, to ensure the real-time of the FBEF-FDP, the number of groups should not be too large. Simulation results demonstrate that the accuracy of the fast butterfly epoch folding-based pulsar period estimation using a few distorted profiles enhances at the cost of a small computational load.

Stock return forecasting typically requires a large number of factors and these factors usually exhibit nonlinear relations with each other. Conventional methods of stock return forecasting mainly fall into two categories: Technical Analysis and Fundamental Analysis. Technical Analysis focuses on time-series data, while Fundamental Analysis explores low-frequency fundamental variables. Although there are substantial works on either time-series analysis or fundamental analysis, few studies have enriched the time-series forecasting with fundamental variables, as the features are characterized by different frequencies, scales and types. In this paper, we propose a Long Short-Term Memory and Deep Neural Network (LSTM-DNN) hybrid model to integrate the fundamental information into time-series forecasting tasks. We demonstrate how investors can benefit from the superior performance of LSTM-DNN by constructing a long-short portfolio that takes long positions in stocks with the highest forecasting returns and short positions in stocks that are expected to decline. Extensive experimental results on real data show that our novel framework could improve the profitability of long-short portfolio strategies compared to the state-of-the-art approaches. We also find evidence indicating that the outperformance of LSTM-DNN model comes from its enhanced ability to extract information from the nonlinear relations among various features, rather than bearing more market risks. Besides the novel framework, we propose a cross-section normalization method, which benefits the framework by providing enriched cross-section signals.

The spatial-temporal baseline threshold method is commonly used to select interferograms for the Small BAseline Subset interferometric synthetic aperture radar (SBAS-InSAR) technique. However, this selection strategy is rather empirical and prone to including highly contaminated interferograms or excluding those with high quality. To overcome these limitations, this study first derives the relationship between the measurement accuracy of unknown parameters and the number of selected interferograms with their corresponding qualities. Subsequently, an adaptive interferogram selection method is proposed on the basis of Graph Theory (GT) and the turbulence atmospheric effects of interferogram. This proposed method first identifies and deletes the SAR image that is severely polluted by atmospheric phase. Second, high-quality interferograms are selected for SBAS-InSAR based on their corresponding turbulence atmospheric variance. Compared with the traditional selection method, this approach can significantly reduce the effect of turbulence atmosphere on SBAS-InSAR. A set of simulated experiments and real Sentinel-1A data in Hawaii, United States, validate the good performance of the proposed method.

We propose reconfiguration of the conventional feedback scheme to suppress the external-cavity time-delay signature (TDS) by adding noise phase modulation in the feedback of semiconductor laser. Noise-phase-modulated feedback introduces broad band noise frequencies into the feedback light and enables the suppression of the TDS. In this work, by means of simulations, the effect on TDS suppression of phase modulation (PM) index is explored. In addition, the influence of noise bandwidth variation is investigated. Using the auto-correlation function to quantity the TDS, we find that, for a large range of operating parameters, the TDS is significantly suppressed to the noise level and even submerged into the base noise. It is shown that suppression TDS is achievable over a wide operating parameter provided the noise generator bandwidth is of order 10 GHz and the PM index is greater than about 3. The proposed configuration will have widespread applications in contexts where suppression of the TDS is required.

With the rapid evolution of technology, growing accessibility, and environmental appeal of wind and solar electric systems, distributed generation (DG) has been pushed from the fringe to a mainstream factor in the grid. However, due to the randomness and uncertainty of environmental and operational conditions, DG also brings many risks and may adversely affect the reliability and safety of the power grid when connected to the distribution network. Therefore, it is necessary to introduce the risk theory into the allocation and placement of DGs. This paper establishes a comprehensive set of risk and economic indexes by modeling the randomness and uncertainty of DG outputs. In addition, islanded operation, which is a promising development direction of microgrids, is explicitly studied and the related indexes are modeled. Putting them together, we propose a risk-based multi-objective optimal allocation model to optimize the placement and configuration of DGs and provide a reliable and cost-effective system. We solve the formulated multi-objective optimization problem by combining the gradient particle swarm optimization algorithm and the bacterial foraging algorithm. We demonstrate the validity and rationality of the proposed method through the analyses of the American PG&E 69-node system.

A convolutional neural network (CNN)-based method for predicting the mechanical properties of hot rolled steel using chemical composition and process parameters is proposed. The novel contribution of this research is to introduce the prediction method of CNN into the steel properties prediction field by converting the production data into two-dimensional data images. Compared with the traditional artificial neural network method, the CNN adopts the idea of local connection and weight sharing, which reduces the complexity of the network model, and uses convolution and pooling operations to extract local features for better prediction precision. The experiments in this paper show that the proposed CNN model with the optimal structure provides higher prediction accuracy and higher robustness when compared with other prediction model reported in the literature. Finally, the metallurgical phenomena in the steel rolling processes are verified by sensitivity analysis using the proposed CNN model. The results are consistent with the metallurgical properties of the steel materials used in the experiments. Therefore, the proposed CNN model has a guiding significance in predicting the mechanical properties of hot rolled steel products in practical applications.

Researches on computer games for Go, Chess, and Japanese Chess stand out as one of the notable landmarks in the progress of artificial intelligence. AlphaGo, AlphaGo Zero, and AlphaZero algorithms, which are called AlphaZero style (AZ-style) algorithms in some literature [1], have achieved superhuman performance by using deep reinforcement learning (DRL). However, the unavailability of training details, expensive equipment used for model training, and the low evaluation accuracy resulted by slow self-play training without expensive computing equipment in practical applications have been the defects of AZ-style algorithms. To solve the problems to a certain extent, the paper proposes an improved online sequential extreme learning machine (IOS-ELM), a new evaluation method, to evaluate chess board positions for AZ-style algortihm. Firstly, the theoretical principles of IOS-ELM is given. Secondly, the study considers Gomoku as the application object and uses IOS-ELM as the evaluation method for AZ-style's board positions to discuss the loss in the training process and hyperparameters affecting performance in detail. Under the same experimental conditions, the proposed method reduces the training parameters by 14 times, training time to 15%, and error of evaluation by 13% compared with the board evaluation network used in original AZ-style algorithms.

Go Artificial Intellects(AIs) using deep reinforcement learning and neural networks have achieved superhuman performance, but they rely on powerful computing resources. They are not applicable to low-cost personal computer(PC). In our life, most entertainment programs of Go run on the general PC. A human Go master consider different strategies for different stages, especially for the middle stage that has a significant impact on winning or losing. To study arguably a more humanlike approach that is applicable to low-cost PC while not reducing chess power, this paper proposes a new search algorithm based on hypothesis testing and dynamic randomization for the middle stage of the game Go. Firstly, a new method to decide the intervals of different playing stages more reasonable based on hypothesis testing is proposed. Secondly, a new search algorithm including a layered pruning branch method, a comprehensive evaluation function and a new selecting node method is proposed. The pruning method based on domain knowledge and upper confidence bound formula(UCB) are all applied to subtract the branches from the lower evaluation score, which was ranked behind 20%. The comprehensive evaluation function with adjustable parameters is proposed to evaluate the tree nodes after pruning. The new selecting node method based on dynamic randomization is used to expand the tree by selecting a node randomly from the high-quality node interval. Finally, the experimental results show that the designed algorithm outperforms Gnugo3.6 and Gnugo3.8 in chess power while reducing average search time and average RAM cost for one move effectively on a 19×19 board.

In some sense, computer game can be used as a test bed of artificial intelligence to develop intelligent algorithms. The paper proposed a kind of intelligent method: a reinforcement learning model based on temporal difference (TD) algorithm. And then the method is used to improve the playing power of the computer game of a special kind of chess. JIU chess, also called Tibetan Go chess, is mainly played in places where Tibetan tribes gather. Its play process is divided two sequential stages: preparation and battle. The layout at preparation is vital for the successive battle, even for the final winning. Studies on Tibetan JIU chess have focused on Bayesian network based pattern extraction and chess shape based strategy, which do not perform well. To address the low chess power of JIU chess from the view of artificial intelligence, we developed a reinforcement learning model based on temporal difference (TD) algorithm for the preparation stage of JIU. First, the search range was limited within a 6 × 6 area at the center of the chessboard, and the TD learning architecture was combined with chess shapes to construct an intelligent environmental feedback system. Second, optimal state transition strategies were obtained by self-play. In addition, the results of the reinforcement learning model were output as SGF files, which act as a pattern library for the battle stage. The experimental results demonstrate that this reinforcement learning model can effectively improve the playing strength of JIU program and outperform the other methods.

Online laboratories are offering new experimental potential for research and education purposes. This paper investigates the design and implementation of a web-based hybrid laboratory framework for research and education. Based on the previous work of the Networked Control System Laboratory (NCSLab), a hybrid laboratory with a highly modular design providing plug-in free online experiments is discussed. The proposed modular design of the NCSLab is provided in four aspects as hardware, software, control algorithms, and deployment, which covers all of the phases that compose an online experimental platform. The experiments that are integrated into the NCSLab for research and education are also explored in detail. To verify the effectiveness of the proposed modular design, a virtual wiring fan speed control system and a physical wireless power transfer system are deployed as examples to illustrate the virtual and remote experiments, respectively. The modular hybrid laboratory has been applied to research and engineering education at Wuhan University, and the corresponding pedagogical evaluation verified the effectiveness of the NCSLab.

Polyadenylation (Poly(A)) plays crucial roles in gene regulation, especially in messenger RNA metabolism, protein diversification, and protein localization. Accurate prediction of polyadenylation sites and identification of motifs that controlling polyadenylation are fundamental for interpreting the patterns of gene expression, improving the accuracy of genome annotation and comprehending the mechanisms that governing gene regulation. Despite considerable advances in using machine learning techniques for this problem, its efficiency is still limited by the lack of experiences and domain knowledge to carefully design and generate useful features, especially for plants. With the increasing availability of extensive genomic data sets and leading computational techniques, deep learning methods, especially convolutional neural networks, have been applied to automatically identify and understand gene regulation directly from gene sequences and predict unknown sequence profiles. Here, we present DeepPolyA, a new deep convolutional neural network-based approach, to predict polyadenylation sites from the plant Arabidopsis thaliana gene sequences. We investigate various deep neural network architectures and evaluate their performance against classical machine learning algorithms and several popular deep learning models. Experimental results demonstrate that DeepPolyA is substantially better than competing methods regarding various performance metrics. We further visualize the learned motifs of DeepPolyA to provide insights of our model and learned polyadenylation signals.

The optimal power flow (OPF) model is a central optimization problem in power system network. In this paper, we propose a novel approach to solve the OPF problem that has a convex objective function and non-convex feasible domain due to the constraints. Based on the concept of abstract convex analysis, we construct the dual M-convex subsets family of original variables by using the variable separation method, followed by the analysis of the extremum according to the infimum base of the M-convex subsets. It is challenging to obtain an explicit mapping function among the separated variables due to the nonlinear equality constraints. We, therefore, use the theorem of implicit function and the differential function to do duality analysis on separated variables. Based on the min-max principle of the primal-dual problem of OPF, we derive the condition of the complementary factor which leads the Lagrange dual problem to maximum, and make the variable separation that results in a Minkowski-type dual M-convex subset. We then can obtain a local minimum using the principle of abstract convex optimization, which will be the global optimal solution under the Karush-Kuhn-Tucker conditions. We evaluate the proposed approach on several IEEE systems. The simulation results indicate that the approach is feasible and effective to deal with the non-convex OPF problem with a non-convex feasible domain.

Traditional machine learning approaches to drug sensitivity prediction assume that training data and test data must be in the same feature space and have the same underlying distribution. However, in real-world applications, this assumption does not hold. For example, we sometimes have limited training data for the task of drug sensitivity prediction in multiple myeloma patients (target task), but we have sufficient auxiliary data for the task of drug sensitivity prediction in patients with another cancer type (related task), where the auxiliary data for the related task are in a different feature space or have a different distribution. In such cases, transfer learning, if applied correctly, would improve the performance of prediction algorithms on the test data of the target task via leveraging the auxiliary data from the related task. In this paper, we present two transfer learning approaches that combine the auxiliary data from the related task with the training data of the target task to improve the prediction performance on the test data of the target task. We evaluate the performance of our transfer learning approaches exploiting three auxiliary data sets and compare them against baseline approaches using the area under the receiver operating characteristic curve on the test data of the target task. Experimental results demonstrate the good performance of our approaches and their superiority over the baseline approaches when auxiliary data are incorporated.