This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that th...This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.展开更多
In data-driven fault diagnosis for turbo-generator sets,the fault samples are usually expensive to obtain,and inevitably with noise,which will both lead to an unsatisfying identification performance of diagnosis model...In data-driven fault diagnosis for turbo-generator sets,the fault samples are usually expensive to obtain,and inevitably with noise,which will both lead to an unsatisfying identification performance of diagnosis models.To address these issues,this paper proposes a fault diagnosis model for turbo-generator sets based on Weighted Extension Neural Network(W-ENN).WENN is a novel neural network which has three types of connection weights and an improved correlation function.The performance of the proposed model is validated against Extension Neural Network(ENN),Support Vector Machine(SVM),Relevance Vector Machine(RVM)and Extreme Learning Machine(ELM)based models.The results indicate that,on noisy small sample sets,the proposed model is superior to the other models in terms of higher identification accuracy with fewer samples and strong noise-tolerant ability.The findings of this study may serve as a powerful fault diagnosis model for turbo-generator sets on noisy small sample sets.展开更多
The limited labeled sample data in the field of advanced security threats detection seriously restricts the effective development of research work.Learning the sample labels from the labeled and unlabeled data has rec...The limited labeled sample data in the field of advanced security threats detection seriously restricts the effective development of research work.Learning the sample labels from the labeled and unlabeled data has received a lot of research attention and various universal labeling methods have been proposed.However,the labeling task of malicious communication samples targeted at advanced threats has to face the two practical challenges:the difficulty of extracting effective features in advance and the complexity of the actual sample types.To address these problems,we proposed a sample labeling method for malicious communication based on semi-supervised deep neural network.This method supports continuous learning and optimization feature representation while labeling sample,and can handle uncertain samples that are outside the concerned sample types.According to the experimental results,our proposed deep neural network can automatically learn effective feature representation,and the validity of features is close to or even higher than that of features which extracted based on expert knowledge.Furthermore,our proposed method can achieve the labeling accuracy of 97.64%~98.50%,which is more accurate than the train-then-detect,kNN and LPA methodsin any labeled-sample proportion condition.The problem of insufficient labeled samples in many network attack detecting scenarios,and our proposed work can function as a reference for the sample labeling tasks in the similar real-world scenarios.展开更多
The difficulty of build-up area extraction is due to complexity of remote sensing data in terms of heterogeneous appearance with large intra-class variations and lower inter-class variations. In order to extract the b...The difficulty of build-up area extraction is due to complexity of remote sensing data in terms of heterogeneous appearance with large intra-class variations and lower inter-class variations. In order to extract the built-up area from Landsat 8-OLI images provided by Google earth engine(GEE), we propose a convolutional neural networks(CNN) utilizing spatial and spectral information synchronously, which is built in Google drive using Colaboratory-Keras. To train a CNN model with good generalization ability, we choose Beijing, Lanzhou, Chongqing, Suzhou and Guangzhou of China as the training sites, which are very different in term of natural environments. The Arc GIS-Model Builder is employed to automatically select 99 332 samples from the 38-m global built-up production of the European Space Agency(ESA) in 2014. The validate accuracy of the five experimental sites is higher than 90%. We compare the results with other existing building data products. The classification results of CNN can be very good for the details of the built-up areas, and greatly reduce the classification error and leakage error. We applied the well-trained CNN model to extract built-up areas of Chengdu, Xi’an, Zhengzhou, Harbin, Hefei, Wuhan, Kunming and Fuzhou, for the sake of evaluating the generalization ability of the CNN. The fine classification results of the eight sites indicate that the generalization ability of the well-trained CNN is pretty good. However, the extraction results of Xi’an, Zhengzhou and Hefei are poor. As for the training data, only Lanzhou is located in the northwest region, so the trained CNN has poor image classification ability in the northwest region of China.展开更多
In airborne gamma ray spectrum processing,different analysis methods,technical requirements,analysis models,and calculation methods need to be established.To meet the engineering practice requirements of airborne gamm...In airborne gamma ray spectrum processing,different analysis methods,technical requirements,analysis models,and calculation methods need to be established.To meet the engineering practice requirements of airborne gamma-ray measurements and improve computational efficiency,an improved shuffled frog leaping algorithm-particle swarm optimization convolutional neural network(SFLA-PSO CNN)for large-sample quantitative analysis of airborne gamma-ray spectra is proposed herein.This method was used to train the weight of the neural network,optimize the structure of the network,delete redundant connections,and enable the neural network to acquire the capability of quantitative spectrum processing.In full-spectrum data processing,this method can perform the functions of energy spectrum peak searching and peak area calculations.After network training,the mean SNR and RMSE of the spectral lines were 31.27 and 2.75,respectively,satisfying the demand for noise reduction.To test the processing ability of the algorithm in large samples of airborne gamma spectra,this study considered the measured data from the Saihangaobi survey area as an example to conduct data spectral analysis.The results show that calculation of the single-peak area takes only 0.13~0.15 ms,and the average relative errors of the peak area in the U,Th,and K spectra are 3.11,9.50,and 6.18%,indicating the high processing efficiency and accuracy of this algorithm.The performance of the model can be further improved by optimizing related parameters,but it can already meet the requirements of practical engineering measurement.This study provides a new idea for the full-spectrum processing of airborne gamma rays.展开更多
Physics-informed neural networks(PINNs)have shown remarkable prospects in solving the forward and inverse problems involving partial differential equations(PDEs).The method embeds PDEs into the neural network by calcu...Physics-informed neural networks(PINNs)have shown remarkable prospects in solving the forward and inverse problems involving partial differential equations(PDEs).The method embeds PDEs into the neural network by calculating the PDE loss at a set of collocation points,providing advantages such as meshfree and more convenient adaptive sampling.However,when solving PDEs using nonuniform collocation points,PINNs still face challenge regarding inefficient convergence of PDE residuals or even failure.In this work,we first analyze the ill-conditioning of the PDE loss in PINNs under nonuniform collocation points.To address the issue,we define volume weighting residual and propose volume weighting physics-informed neural networks(VW-PINNs).Through weighting the PDE residuals by the volume that the collocation points occupy within the computational domain,we embed explicitly the distribution characteristics of collocation points in the loss evaluation.The fast and sufficient convergence of the PDE residuals for the problems involving nonuniform collocation points is guaranteed.Considering the meshfree characteristics of VW-PINNs,we also develop a volume approximation algorithm based on kernel density estimation to calculate the volume of the collocation points.We validate the universality of VW-PINNs by solving the forward problems involving flow over a circular cylinder and flow over the NACA0012 airfoil under different inflow conditions,where conventional PINNs fail.By solving the Burgers’equation,we verify that VW-PINNs can enhance the efficiency of existing the adaptive sampling method in solving the forward problem by three times,and can reduce the relative L 2 error of conventional PINNs in solving the inverse problem by more than one order of magnitude.展开更多
Optical solitons,as self-sustaining waveforms in a nonlinear medium where dispersion and nonlinear effects are balanced,have key applications in ultrafast laser systems and optical communications.Physics-informed neur...Optical solitons,as self-sustaining waveforms in a nonlinear medium where dispersion and nonlinear effects are balanced,have key applications in ultrafast laser systems and optical communications.Physics-informed neural networks(PINN)provide a new way to solve the nonlinear Schrodinger equation describing the soliton evolution by fusing data-driven and physical constraints.However,the grid point sampling strategy of traditional PINN suffers from high computational complexity and unstable gradient flow,which makes it difficult to capture the physical details efficiently.In this paper,we propose a residual-based adaptive multi-distribution(RAMD)sampling method to optimize the PINN training process by dynamically constructing a multi-modal loss distribution.With a 50%reduction in the number of grid points,RAMD significantly reduces the relative error of PINN and,in particular,optimizes the solution error of the(2+1)Ginzburg–Landau equation from 4.55%to 1.98%.RAMD breaks through the lack of physical constraints in the purely data-driven model by the innovative combination of multi-modal distribution modeling and autonomous sampling control for the design of all-optical communication devices.RAMD provides a high-precision numerical simulation tool for the design of all-optical communication devices,optimization of nonlinear laser devices,and other studies.展开更多
We investigate the stochastic asymptotical synchronization of chaotic Markovian jumping fuzzy cellular neural networks （MJFCNNs） with discrete, unbounded distributed delays, and the Wiener process based on sampled-d...We investigate the stochastic asymptotical synchronization of chaotic Markovian jumping fuzzy cellular neural networks （MJFCNNs） with discrete, unbounded distributed delays, and the Wiener process based on sampled-data control using the linear matrix inequality （LMI） approach. The Lyapunov–Krasovskii functional combined with the input delay approach as well as the free-weighting matrix approach is employed to derive several sufficient criteria in terms of LMIs to ensure that the delayed MJFCNNs with the Wiener process is stochastic asymptotical synchronous. Restrictions （e.g., time derivative is smaller than one） are removed to obtain a proposed sampled-data controller. Finally, a numerical example is provided to demonstrate the reliability of the derived results.展开更多
This paper is concerned with the synchronization of delayed neural networks via sampled-data control. A new technique, namely, the free-matrix-based time-dependent discontinuous Lyapunov functional approach, is adopte...This paper is concerned with the synchronization of delayed neural networks via sampled-data control. A new technique, namely, the free-matrix-based time-dependent discontinuous Lyapunov functional approach, is adopted in constructing the Lyapunov functional, which takes advantage of the sampling characteristic of sawtooth input delay. Based on this discontinuous Lyapunov functional, some less conservative synchronization criteria are established to ensure that the slave system is synchronous with the master system. The desired sampled-data controller can be obtained through the use of the linear matrix inequality（LMI） technique. Finally, two numerical examples are provided to demonstrate the effectiveness and the improvements of the proposed methods.展开更多
Explores the generalization error of fuzzy neural network, analyzes the reason for occurrence and presents the equation of calculating error by the confidence interval approach. In addition, a generalization error tra...Explores the generalization error of fuzzy neural network, analyzes the reason for occurrence and presents the equation of calculating error by the confidence interval approach. In addition, a generalization error transfering(GET) method of improving the generalization error is proposed. The simulation experimental results of heating furnance show that the GET scheme is efficient.展开更多
Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new meth...Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new method based on virtual model and invariant moments was proposed to generate training samples.The method was composed of the following steps：use computer and simulation software to build target object＇s virtual model and then simulate the environment,light condition,camera parameter,etc.;rotate the model by spin and nutation of inclination to get the image sequence by virtual camera;preprocess each image and transfer them into binary image;calculate the invariant moments for each image and get a vectors＇ sequence.The vectors＇ sequence which was proved to be complete became the training samples together with the target outputs.The simulated results showed that the proposed method could be used to recognize the real targets and improve the accuracy of target recognition effectively when the sampling interval was short enough and the circumstance simulation was close enough.展开更多
Neural Networks (NN) are the functional unit of Deep Learning and are known to mimic the behavior of the human brain to solve complex data-driven problems. Whenever we train our own neural networks, we need to take ca...Neural Networks (NN) are the functional unit of Deep Learning and are known to mimic the behavior of the human brain to solve complex data-driven problems. Whenever we train our own neural networks, we need to take care of something called the generalization of the neural network. The performance of Artificial Neural Networks (ANN) mostly depends upon its generalization capability. In this paper, we propose an innovative approach to enhance the generalization capability of artificial neural networks (ANN) using structural redundancy. A novel perspective on handling input data prototypes and their impact on the development of generalization, which could improve to ANN architectures accuracy and reliability is described.展开更多
In order to carry out an integrated assessment of sea water quality objectively, this paper based on the concept and principle of artificial neural network, generated appropriate training samples for BP artificial neu...In order to carry out an integrated assessment of sea water quality objectively, this paper based on the concept and principle of artificial neural network, generated appropriate training samples for BP artificial neural network model through the method of producing samples to the concentration of various pollution index of sea water quality from the viewpoint of threshold, established the BP artificial neural network model of sea water quality assessment using multi-layer neural network with error back-propagation algorithm. This model was used to assess water environment and obtain sea water quality categories of offshore area in Bohai Bay through calculating. The calculations shown that pollution index in river＇s wet season was higher than that in dry season from 2004 to 2007, and the pollution was particularly serious in 2005 and 2006, but a little better in 2007. The assessed results of cases shown that the model was reasonable in design and higher in generalization, meanwhile, it was common, objective and practical to sea water quality assessment.展开更多
To study the capacity of artificial neural network (ANN) applying to battlefield target classification and result of classification, according to the characteristics of battlefield target acoustic and seismic sign...To study the capacity of artificial neural network (ANN) applying to battlefield target classification and result of classification, according to the characteristics of battlefield target acoustic and seismic signals, an on the spot experiment was carried out to derive acoustic and seismic signals of a tank and jeep by special experiment system. Experiment data processed by fast Fourier transform(FFT) were used to train the ANN to distinguish the two battlefield targets. The ANN classifier was performed by the special program based on the modified back propagation (BP) algorithm. The ANN classifier has high correct identification rates for acoustic and seismic signals of battlefield targets, and is suitable for the classification of battlefield targets. The modified BP algorithm eliminates oscillations and local minimum of the standard BP algorithm, and enhances the convergence rate of the ANN.展开更多
Concerning the issue of high-dimensions and low-failure probabilities including implicit and highly nonlinear limit state function, reliability analysis based on the directional importance sampling in combination with...Concerning the issue of high-dimensions and low-failure probabilities including implicit and highly nonlinear limit state function, reliability analysis based on the directional importance sampling in combination with the radial basis function （RBF） neural network is used, and the RBF neural network based on first-order reliability method （FORM） is to approximate the unknown implicit limit state functions and calculate the most probable point （MPP） with iterative algorithm. For good efficiency, based on the ideas that directional sampling reduces dimensionality and importance sampling focuses on the domain contributing to failure probability, the joint probability density function of importance sampling is constructed, and the sampling center is moved to MPP to ensure that more random sample points draw belong to the failure domain and the simulation efficiency is improved. Then the numerical example of initiating explosive devices for rocket booster explosive bolts demonstrates the applicability, versatility and accuracy of the approach compared with other reliability simulation algorithm.展开更多
Hand gestures are a natural way for human-robot interaction.Vision based dynamic hand gesture recognition has become a hot research topic due to its various applications.This paper presents a novel deep learning netwo...Hand gestures are a natural way for human-robot interaction.Vision based dynamic hand gesture recognition has become a hot research topic due to its various applications.This paper presents a novel deep learning network for hand gesture recognition.The network integrates several well-proved modules together to learn both short-term and long-term features from video inputs and meanwhile avoid intensive computation.To learn short-term features,each video input is segmented into a fixed number of frame groups.A frame is randomly selected from each group and represented as an RGB image as well as an optical flow snapshot.These two entities are fused and fed into a convolutional neural network(Conv Net)for feature extraction.The Conv Nets for all groups share parameters.To learn longterm features,outputs from all Conv Nets are fed into a long short-term memory(LSTM)network,by which a final classification result is predicted.The new model has been tested with two popular hand gesture datasets,namely the Jester dataset and Nvidia dataset.Comparing with other models,our model produced very competitive results.The robustness of the new model has also been proved with an augmented dataset with enhanced diversity of hand gestures.展开更多
In this study, the application of artificial intelligence to monthly and seasonal rainfall forecasting in Queensland, Australia, was assessed by inputting recognized climate indices, monthly historical rainfall data, ...In this study, the application of artificial intelligence to monthly and seasonal rainfall forecasting in Queensland, Australia, was assessed by inputting recognized climate indices, monthly historical rainfall data, and atmospheric temperatures into a prototype stand-alone, dynamic, recurrent, time-delay, artificial neural network. Outputs, as monthly rainfall forecasts 3 months in advance for the period 1993 to 2009, were compared with observed rainfall data using time-series plots, root mean squared error （RMSE）, and Pearson correlation coefficients. A comparison of RMSE values with forecasts generated by the Australian Bureau of Meteorology＇s Predictive Ocean Atmosphere Model for Australia （POAMA）-I.5 general circulation model （GCM） indicated that the prototype achieved a lower RMSE for 16 of the 17 sites compared. The application of artificial neural networks to rainfall forecasting was reviewed. The prototype design is considered preliminary, with potential for significant improvement such as inclusion of output from GCMs and experimentation with other input attributes.展开更多
We consider solving the forward and inverse partial differential equations(PDEs)which have sharp solutions with physics-informed neural networks(PINNs)in this work.In particular,to better capture the sharpness of the ...We consider solving the forward and inverse partial differential equations(PDEs)which have sharp solutions with physics-informed neural networks(PINNs)in this work.In particular,to better capture the sharpness of the solution,we propose the adaptive sampling methods(ASMs)based on the residual and the gradient of the solution.We first present a residual only-based ASM denoted by ASMⅠ.In this approach,we first train the neural network using a small number of residual points and divide the computational domain into a certain number of sub-domains,then we add new residual points in the sub-domain which has the largest mean absolute value of the residual,and those points which have the largest absolute values of the residual in this sub-domain as new residual points.We further develop a second type of ASM(denoted by ASMⅡ)based on both the residual and the gradient of the solution due to the fact that only the residual may not be able to efficiently capture the sharpness of the solution.The procedure of ASMⅡis almost the same as that of ASMⅠ,and we add new residual points which have not only large residuals but also large gradients.To demonstrate the effectiveness of the present methods,we use both ASMⅠand ASMⅡto solve a number of PDEs,including the Burger equation,the compressible Euler equation,the Poisson equation over an Lshape domain as well as the high-dimensional Poisson equation.It has been shown from the numerical results that the sharp solutions can be well approximated by using either ASMⅠor ASMⅡ,and both methods deliver much more accurate solutions than the original PINNs with the same number of residual points.Moreover,the ASMⅡalgorithm has better performance in terms of accuracy,efficiency,and stability compared with the ASMⅠalgorithm.This means that the gradient of the solution improves the stability and efficiency of the adaptive sampling procedure as well as the accuracy of the solution.Furthermore,we also employ the similar adaptive sampling technique for the data points of boundary conditions(BCs)if the sharpness of the solution is near the boundary.The result of the L-shape Poisson problem indicates that the present method can significantly improve the efficiency,stability,and accuracy.展开更多
Monte Carlo Simulations(MCS),commonly used for reliability analysis,require a large amount of data points to obtain acceptable accuracy,even if the Subset Simulation with Importance Sampling(SS/IS)methods are used.The...Monte Carlo Simulations(MCS),commonly used for reliability analysis,require a large amount of data points to obtain acceptable accuracy,even if the Subset Simulation with Importance Sampling(SS/IS)methods are used.The Second Order Reliability Method(SORM)has proved to be an excellent rapid tool in the stochastic analysis of laminated composite structures,when compared to the slower MCS techniques.However,SORM requires differentiating the performance function with respect to each of the random variables involved in the simulation.The most suitable approach to do this is to use a symbolic solver,which renders the simulations very slow,although still faster than MCS.Moreover,the inability to obtain the derivative of the performance function with respect to some parameters,such as ply thickness,limits the capabilities of the classical SORM.In this work,a Neural Network-Based Second Order Reliability Method(NNBSORM)is developed to replace the finite element algorithm in the stochastic analysis of laminated composite plates in free vibration.Because of the ability to obtain expressions for the first and second derivatives of the NN system outputs with respect to any of its inputs,such as material properties,ply thicknesses and orientation angles,the need for using a symbolic solver to calculate the derivatives of the performance function no longer exists.The proposed approach is accordingly much faster,and easily allows for the consideration of ply thickness uncertainty.The present analysis showed that dealing with ply thicknesses as random variables results in 37%increase in the laminate’s probability of failure.展开更多
Firstly,general regression neural network(GRNN) was used for variable selection of key influencing factors of residential load(RL) forecasting.Secondly,the key influencing factors chosen by GRNN were used as the input...Firstly,general regression neural network(GRNN) was used for variable selection of key influencing factors of residential load(RL) forecasting.Secondly,the key influencing factors chosen by GRNN were used as the input and output terminals of urban and rural RL for simulating and learning.In addition,the suitable parameters of final model were obtained through applying the evidence theory to combine the optimization results which were calculated with the PSO method and the Bayes theory.Then,the model of PSO-Bayes least squares support vector machine(PSO-Bayes-LS-SVM) was established.A case study was then provided for the learning and testing.The empirical analysis results show that the mean square errors of urban and rural RL forecast are 0.02% and 0.04%,respectively.At last,taking a specific province RL in China as an example,the forecast results of RL from 2011 to 2015 were obtained.展开更多
文摘This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.
基金the National Natural Science Foundation of China(No.51775272,No.51005114)The Fundamental Research Funds for the Central Universities,China(No.NS2014050)。
文摘In data-driven fault diagnosis for turbo-generator sets,the fault samples are usually expensive to obtain,and inevitably with noise,which will both lead to an unsatisfying identification performance of diagnosis models.To address these issues,this paper proposes a fault diagnosis model for turbo-generator sets based on Weighted Extension Neural Network(W-ENN).WENN is a novel neural network which has three types of connection weights and an improved correlation function.The performance of the proposed model is validated against Extension Neural Network(ENN),Support Vector Machine(SVM),Relevance Vector Machine(RVM)and Extreme Learning Machine(ELM)based models.The results indicate that,on noisy small sample sets,the proposed model is superior to the other models in terms of higher identification accuracy with fewer samples and strong noise-tolerant ability.The findings of this study may serve as a powerful fault diagnosis model for turbo-generator sets on noisy small sample sets.
基金partially funded by the National Natural Science Foundation of China (Grant No. 61272447)National Entrepreneurship & Innovation Demonstration Base of China (Grant No. C700011)Key Research & Development Project of Sichuan Province of China (Grant No. 2018G20100)
文摘The limited labeled sample data in the field of advanced security threats detection seriously restricts the effective development of research work.Learning the sample labels from the labeled and unlabeled data has received a lot of research attention and various universal labeling methods have been proposed.However,the labeling task of malicious communication samples targeted at advanced threats has to face the two practical challenges:the difficulty of extracting effective features in advance and the complexity of the actual sample types.To address these problems,we proposed a sample labeling method for malicious communication based on semi-supervised deep neural network.This method supports continuous learning and optimization feature representation while labeling sample,and can handle uncertain samples that are outside the concerned sample types.According to the experimental results,our proposed deep neural network can automatically learn effective feature representation,and the validity of features is close to or even higher than that of features which extracted based on expert knowledge.Furthermore,our proposed method can achieve the labeling accuracy of 97.64%~98.50%,which is more accurate than the train-then-detect,kNN and LPA methodsin any labeled-sample proportion condition.The problem of insufficient labeled samples in many network attack detecting scenarios,and our proposed work can function as a reference for the sample labeling tasks in the similar real-world scenarios.
基金supported by the National Key R&D Program of China(No.2017YFB0504104)the National Natural Science Foundation of China(No.41571334)。
文摘The difficulty of build-up area extraction is due to complexity of remote sensing data in terms of heterogeneous appearance with large intra-class variations and lower inter-class variations. In order to extract the built-up area from Landsat 8-OLI images provided by Google earth engine(GEE), we propose a convolutional neural networks(CNN) utilizing spatial and spectral information synchronously, which is built in Google drive using Colaboratory-Keras. To train a CNN model with good generalization ability, we choose Beijing, Lanzhou, Chongqing, Suzhou and Guangzhou of China as the training sites, which are very different in term of natural environments. The Arc GIS-Model Builder is employed to automatically select 99 332 samples from the 38-m global built-up production of the European Space Agency(ESA) in 2014. The validate accuracy of the five experimental sites is higher than 90%. We compare the results with other existing building data products. The classification results of CNN can be very good for the details of the built-up areas, and greatly reduce the classification error and leakage error. We applied the well-trained CNN model to extract built-up areas of Chengdu, Xi’an, Zhengzhou, Harbin, Hefei, Wuhan, Kunming and Fuzhou, for the sake of evaluating the generalization ability of the CNN. The fine classification results of the eight sites indicate that the generalization ability of the well-trained CNN is pretty good. However, the extraction results of Xi’an, Zhengzhou and Hefei are poor. As for the training data, only Lanzhou is located in the northwest region, so the trained CNN has poor image classification ability in the northwest region of China.
基金the National Natural Science Foundation of China(No.42127807)Natural Science Foundation of Sichuan Province(Nos.23NSFSCC0116 and 2022NSFSC12333)the Nuclear Energy Development Project(No.[2021]-88).
文摘In airborne gamma ray spectrum processing,different analysis methods,technical requirements,analysis models,and calculation methods need to be established.To meet the engineering practice requirements of airborne gamma-ray measurements and improve computational efficiency,an improved shuffled frog leaping algorithm-particle swarm optimization convolutional neural network(SFLA-PSO CNN)for large-sample quantitative analysis of airborne gamma-ray spectra is proposed herein.This method was used to train the weight of the neural network,optimize the structure of the network,delete redundant connections,and enable the neural network to acquire the capability of quantitative spectrum processing.In full-spectrum data processing,this method can perform the functions of energy spectrum peak searching and peak area calculations.After network training,the mean SNR and RMSE of the spectral lines were 31.27 and 2.75,respectively,satisfying the demand for noise reduction.To test the processing ability of the algorithm in large samples of airborne gamma spectra,this study considered the measured data from the Saihangaobi survey area as an example to conduct data spectral analysis.The results show that calculation of the single-peak area takes only 0.13~0.15 ms,and the average relative errors of the peak area in the U,Th,and K spectra are 3.11,9.50,and 6.18%,indicating the high processing efficiency and accuracy of this algorithm.The performance of the model can be further improved by optimizing related parameters,but it can already meet the requirements of practical engineering measurement.This study provides a new idea for the full-spectrum processing of airborne gamma rays.
基金supported by the National Natural Science Foundation of China(Grant No.92152301)the National Key Research and Development Program of China(Grant No.2022YFB4300200)the Shaanxi Provincial Key Research and Development Program(Grant No.2023-ZDLGY-27).
文摘Physics-informed neural networks(PINNs)have shown remarkable prospects in solving the forward and inverse problems involving partial differential equations(PDEs).The method embeds PDEs into the neural network by calculating the PDE loss at a set of collocation points,providing advantages such as meshfree and more convenient adaptive sampling.However,when solving PDEs using nonuniform collocation points,PINNs still face challenge regarding inefficient convergence of PDE residuals or even failure.In this work,we first analyze the ill-conditioning of the PDE loss in PINNs under nonuniform collocation points.To address the issue,we define volume weighting residual and propose volume weighting physics-informed neural networks(VW-PINNs).Through weighting the PDE residuals by the volume that the collocation points occupy within the computational domain,we embed explicitly the distribution characteristics of collocation points in the loss evaluation.The fast and sufficient convergence of the PDE residuals for the problems involving nonuniform collocation points is guaranteed.Considering the meshfree characteristics of VW-PINNs,we also develop a volume approximation algorithm based on kernel density estimation to calculate the volume of the collocation points.We validate the universality of VW-PINNs by solving the forward problems involving flow over a circular cylinder and flow over the NACA0012 airfoil under different inflow conditions,where conventional PINNs fail.By solving the Burgers’equation,we verify that VW-PINNs can enhance the efficiency of existing the adaptive sampling method in solving the forward problem by three times,and can reduce the relative L 2 error of conventional PINNs in solving the inverse problem by more than one order of magnitude.
基金supported by the National Key R&D Program of China(Grant No.2022YFA1604200)National Natural Science Foundation of China(Grant No.12261131495)+1 种基金Beijing Municipal Science and Technology Commission,Adminitrative Commission of Zhongguancun Science Park(Grant No.Z231100006623006)Institute of Systems Science,Beijing Wuzi University(Grant No.BWUISS21)。
文摘Optical solitons,as self-sustaining waveforms in a nonlinear medium where dispersion and nonlinear effects are balanced,have key applications in ultrafast laser systems and optical communications.Physics-informed neural networks(PINN)provide a new way to solve the nonlinear Schrodinger equation describing the soliton evolution by fusing data-driven and physical constraints.However,the grid point sampling strategy of traditional PINN suffers from high computational complexity and unstable gradient flow,which makes it difficult to capture the physical details efficiently.In this paper,we propose a residual-based adaptive multi-distribution(RAMD)sampling method to optimize the PINN training process by dynamically constructing a multi-modal loss distribution.With a 50%reduction in the number of grid points,RAMD significantly reduces the relative error of PINN and,in particular,optimizes the solution error of the(2+1)Ginzburg–Landau equation from 4.55%to 1.98%.RAMD breaks through the lack of physical constraints in the purely data-driven model by the innovative combination of multi-modal distribution modeling and autonomous sampling control for the design of all-optical communication devices.RAMD provides a high-precision numerical simulation tool for the design of all-optical communication devices,optimization of nonlinear laser devices,and other studies.
基金the Ministry of Science and Technology of India(Grant No.DST/Inspire Fellowship/2010/[293]/dt.18/03/2011)
文摘We investigate the stochastic asymptotical synchronization of chaotic Markovian jumping fuzzy cellular neural networks （MJFCNNs） with discrete, unbounded distributed delays, and the Wiener process based on sampled-data control using the linear matrix inequality （LMI） approach. The Lyapunov–Krasovskii functional combined with the input delay approach as well as the free-weighting matrix approach is employed to derive several sufficient criteria in terms of LMIs to ensure that the delayed MJFCNNs with the Wiener process is stochastic asymptotical synchronous. Restrictions （e.g., time derivative is smaller than one） are removed to obtain a proposed sampled-data controller. Finally, a numerical example is provided to demonstrate the reliability of the derived results.
基金Project supported by the National Natural Science Foundation of China(Grant No.61304064)the Scientific Research Fund of Hunan Provincial Education Department,China(Grant Nos.15B067 and 16C0475)a Discovering Grant from Australian Research Council
文摘This paper is concerned with the synchronization of delayed neural networks via sampled-data control. A new technique, namely, the free-matrix-based time-dependent discontinuous Lyapunov functional approach, is adopted in constructing the Lyapunov functional, which takes advantage of the sampling characteristic of sawtooth input delay. Based on this discontinuous Lyapunov functional, some less conservative synchronization criteria are established to ensure that the slave system is synchronous with the master system. The desired sampled-data controller can be obtained through the use of the linear matrix inequality（LMI） technique. Finally, two numerical examples are provided to demonstrate the effectiveness and the improvements of the proposed methods.
文摘Explores the generalization error of fuzzy neural network, analyzes the reason for occurrence and presents the equation of calculating error by the confidence interval approach. In addition, a generalization error transfering(GET) method of improving the generalization error is proposed. The simulation experimental results of heating furnance show that the GET scheme is efficient.
基金Supported by the Ministerial Level Research Foundation(404040401)
文摘Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new method based on virtual model and invariant moments was proposed to generate training samples.The method was composed of the following steps：use computer and simulation software to build target object＇s virtual model and then simulate the environment,light condition,camera parameter,etc.;rotate the model by spin and nutation of inclination to get the image sequence by virtual camera;preprocess each image and transfer them into binary image;calculate the invariant moments for each image and get a vectors＇ sequence.The vectors＇ sequence which was proved to be complete became the training samples together with the target outputs.The simulated results showed that the proposed method could be used to recognize the real targets and improve the accuracy of target recognition effectively when the sampling interval was short enough and the circumstance simulation was close enough.
文摘Neural Networks (NN) are the functional unit of Deep Learning and are known to mimic the behavior of the human brain to solve complex data-driven problems. Whenever we train our own neural networks, we need to take care of something called the generalization of the neural network. The performance of Artificial Neural Networks (ANN) mostly depends upon its generalization capability. In this paper, we propose an innovative approach to enhance the generalization capability of artificial neural networks (ANN) using structural redundancy. A novel perspective on handling input data prototypes and their impact on the development of generalization, which could improve to ANN architectures accuracy and reliability is described.
文摘In order to carry out an integrated assessment of sea water quality objectively, this paper based on the concept and principle of artificial neural network, generated appropriate training samples for BP artificial neural network model through the method of producing samples to the concentration of various pollution index of sea water quality from the viewpoint of threshold, established the BP artificial neural network model of sea water quality assessment using multi-layer neural network with error back-propagation algorithm. This model was used to assess water environment and obtain sea water quality categories of offshore area in Bohai Bay through calculating. The calculations shown that pollution index in river＇s wet season was higher than that in dry season from 2004 to 2007, and the pollution was particularly serious in 2005 and 2006, but a little better in 2007. The assessed results of cases shown that the model was reasonable in design and higher in generalization, meanwhile, it was common, objective and practical to sea water quality assessment.
文摘To study the capacity of artificial neural network (ANN) applying to battlefield target classification and result of classification, according to the characteristics of battlefield target acoustic and seismic signals, an on the spot experiment was carried out to derive acoustic and seismic signals of a tank and jeep by special experiment system. Experiment data processed by fast Fourier transform(FFT) were used to train the ANN to distinguish the two battlefield targets. The ANN classifier was performed by the special program based on the modified back propagation (BP) algorithm. The ANN classifier has high correct identification rates for acoustic and seismic signals of battlefield targets, and is suitable for the classification of battlefield targets. The modified BP algorithm eliminates oscillations and local minimum of the standard BP algorithm, and enhances the convergence rate of the ANN.
文摘Concerning the issue of high-dimensions and low-failure probabilities including implicit and highly nonlinear limit state function, reliability analysis based on the directional importance sampling in combination with the radial basis function （RBF） neural network is used, and the RBF neural network based on first-order reliability method （FORM） is to approximate the unknown implicit limit state functions and calculate the most probable point （MPP） with iterative algorithm. For good efficiency, based on the ideas that directional sampling reduces dimensionality and importance sampling focuses on the domain contributing to failure probability, the joint probability density function of importance sampling is constructed, and the sampling center is moved to MPP to ensure that more random sample points draw belong to the failure domain and the simulation efficiency is improved. Then the numerical example of initiating explosive devices for rocket booster explosive bolts demonstrates the applicability, versatility and accuracy of the approach compared with other reliability simulation algorithm.
文摘Hand gestures are a natural way for human-robot interaction.Vision based dynamic hand gesture recognition has become a hot research topic due to its various applications.This paper presents a novel deep learning network for hand gesture recognition.The network integrates several well-proved modules together to learn both short-term and long-term features from video inputs and meanwhile avoid intensive computation.To learn short-term features,each video input is segmented into a fixed number of frame groups.A frame is randomly selected from each group and represented as an RGB image as well as an optical flow snapshot.These two entities are fused and fed into a convolutional neural network(Conv Net)for feature extraction.The Conv Nets for all groups share parameters.To learn longterm features,outputs from all Conv Nets are fed into a long short-term memory(LSTM)network,by which a final classification result is predicted.The new model has been tested with two popular hand gesture datasets,namely the Jester dataset and Nvidia dataset.Comparing with other models,our model produced very competitive results.The robustness of the new model has also been proved with an augmented dataset with enhanced diversity of hand gestures.
文摘In this study, the application of artificial intelligence to monthly and seasonal rainfall forecasting in Queensland, Australia, was assessed by inputting recognized climate indices, monthly historical rainfall data, and atmospheric temperatures into a prototype stand-alone, dynamic, recurrent, time-delay, artificial neural network. Outputs, as monthly rainfall forecasts 3 months in advance for the period 1993 to 2009, were compared with observed rainfall data using time-series plots, root mean squared error （RMSE）, and Pearson correlation coefficients. A comparison of RMSE values with forecasts generated by the Australian Bureau of Meteorology＇s Predictive Ocean Atmosphere Model for Australia （POAMA）-I.5 general circulation model （GCM） indicated that the prototype achieved a lower RMSE for 16 of the 17 sites compared. The application of artificial neural networks to rainfall forecasting was reviewed. The prototype design is considered preliminary, with potential for significant improvement such as inclusion of output from GCMs and experimentation with other input attributes.
基金Project supported by the National Key R&D Program of China(No.2022YFA1004504)the National Natural Science Foundation of China(Nos.12171404 and 12201229)the Fundamental Research Funds for Central Universities of China(No.20720210037)。
文摘We consider solving the forward and inverse partial differential equations(PDEs)which have sharp solutions with physics-informed neural networks(PINNs)in this work.In particular,to better capture the sharpness of the solution,we propose the adaptive sampling methods(ASMs)based on the residual and the gradient of the solution.We first present a residual only-based ASM denoted by ASMⅠ.In this approach,we first train the neural network using a small number of residual points and divide the computational domain into a certain number of sub-domains,then we add new residual points in the sub-domain which has the largest mean absolute value of the residual,and those points which have the largest absolute values of the residual in this sub-domain as new residual points.We further develop a second type of ASM(denoted by ASMⅡ)based on both the residual and the gradient of the solution due to the fact that only the residual may not be able to efficiently capture the sharpness of the solution.The procedure of ASMⅡis almost the same as that of ASMⅠ,and we add new residual points which have not only large residuals but also large gradients.To demonstrate the effectiveness of the present methods,we use both ASMⅠand ASMⅡto solve a number of PDEs,including the Burger equation,the compressible Euler equation,the Poisson equation over an Lshape domain as well as the high-dimensional Poisson equation.It has been shown from the numerical results that the sharp solutions can be well approximated by using either ASMⅠor ASMⅡ,and both methods deliver much more accurate solutions than the original PINNs with the same number of residual points.Moreover,the ASMⅡalgorithm has better performance in terms of accuracy,efficiency,and stability compared with the ASMⅠalgorithm.This means that the gradient of the solution improves the stability and efficiency of the adaptive sampling procedure as well as the accuracy of the solution.Furthermore,we also employ the similar adaptive sampling technique for the data points of boundary conditions(BCs)if the sharpness of the solution is near the boundary.The result of the L-shape Poisson problem indicates that the present method can significantly improve the efficiency,stability,and accuracy.
文摘Monte Carlo Simulations(MCS),commonly used for reliability analysis,require a large amount of data points to obtain acceptable accuracy,even if the Subset Simulation with Importance Sampling(SS/IS)methods are used.The Second Order Reliability Method(SORM)has proved to be an excellent rapid tool in the stochastic analysis of laminated composite structures,when compared to the slower MCS techniques.However,SORM requires differentiating the performance function with respect to each of the random variables involved in the simulation.The most suitable approach to do this is to use a symbolic solver,which renders the simulations very slow,although still faster than MCS.Moreover,the inability to obtain the derivative of the performance function with respect to some parameters,such as ply thickness,limits the capabilities of the classical SORM.In this work,a Neural Network-Based Second Order Reliability Method(NNBSORM)is developed to replace the finite element algorithm in the stochastic analysis of laminated composite plates in free vibration.Because of the ability to obtain expressions for the first and second derivatives of the NN system outputs with respect to any of its inputs,such as material properties,ply thicknesses and orientation angles,the need for using a symbolic solver to calculate the derivatives of the performance function no longer exists.The proposed approach is accordingly much faster,and easily allows for the consideration of ply thickness uncertainty.The present analysis showed that dealing with ply thicknesses as random variables results in 37%increase in the laminate’s probability of failure.
基金Project(07JA790092) supported by the Research Grants from Humanities and Social Science Program of Ministry of Education of ChinaProject(10MR44) supported by the Fundamental Research Funds for the Central Universities in China
文摘Firstly,general regression neural network(GRNN) was used for variable selection of key influencing factors of residential load(RL) forecasting.Secondly,the key influencing factors chosen by GRNN were used as the input and output terminals of urban and rural RL for simulating and learning.In addition,the suitable parameters of final model were obtained through applying the evidence theory to combine the optimization results which were calculated with the PSO method and the Bayes theory.Then,the model of PSO-Bayes least squares support vector machine(PSO-Bayes-LS-SVM) was established.A case study was then provided for the learning and testing.The empirical analysis results show that the mean square errors of urban and rural RL forecast are 0.02% and 0.04%,respectively.At last,taking a specific province RL in China as an example,the forecast results of RL from 2011 to 2015 were obtained.
