The photonic neural processing unit(PNPU)demonstrates ultrahigh inference speed with low energy consumption,and it has become a promising hardware artificial intelligence(AI)accelerator.However,the nonidealities of th...The photonic neural processing unit(PNPU)demonstrates ultrahigh inference speed with low energy consumption,and it has become a promising hardware artificial intelligence(AI)accelerator.However,the nonidealities of the photonic device and the peripheral circuit make the practical application much more complex.Rather than optimizing the photonic device,the architecture,and the algorithm individually,a joint device-architecture-algorithm codesign method is proposed to improve the accuracy,efficiency and robustness of the PNPU.First,a full-flow simulator for the PNPU is developed from the back end simulator to the high-level training framework;Second,the full system architecture and the complete photonic chip design enable the simulator to closely model the real system;Third,the nonidealities of the photonic chip are evaluated for the PNPU design.The average test accuracy exceeds 98%,and the computing power exceeds 100TOPS.展开更多
Background Visual perception of face images or face pareidolia can be evaluated with event-related potentials(ERP)for healthy subjects and patients with neurological conditions.In this study,we aimed to analyse event-...Background Visual perception of face images or face pareidolia can be evaluated with event-related potentials(ERP)for healthy subjects and patients with neurological conditions.In this study,we aimed to analyse event-related potential components such as P100,N100,N170,and vertex-positive potential(VPP)in response to face pareidolia perception in temporal lobe epilepsy(TLE)patients.Methods ERPs were recorded during the pareidolia test.Waveforms were analzyed and current source density(CSD)maps were generated.Results CSD profles were shown to be interpretable when face and face pareidolia conditions.N100,P100,and N170 components showed larger amplitudes and longer latency in epilepsy patients in response to face pareidolia stimuli compared to real face images.However,the N170 component latency did not difer signifcantly between epilepsy patients and healthy participants,while the larger amplitude and longer latency of N100 and P100 responses were evoked in healthy patients.Conclusions Our results indicate a diference in the neural mechanisms of processing real face information and pareidolia face-like information in TLE patients.展开更多
In the real world, the inputs of many complicated systems are time-varying functions or processes. In order to predict the outputs of these systems with high speed and accuracy, this paper proposes a time series predi...In the real world, the inputs of many complicated systems are time-varying functions or processes. In order to predict the outputs of these systems with high speed and accuracy, this paper proposes a time series prediction model based on the wavelet process neural network, and develops the corresponding learning algorithm based on the expansion of the orthogonal basis functions. The effectiveness of the proposed time series prediction model and its learning algorithm is proved by the Macke-Glass time series prediction, and the comparative prediction results indicate that the proposed time series prediction model based on the wavelet process neural network seems to perform well and appears suitable for using as a good tool to predict the highly complex nonlinear time series.展开更多
This paper develops a deep learning tool based on neural processes(NPs)called the Peri-Net-Pro,to predict the crack patterns in a moving disk and classifies them according to the classification modes with quantified u...This paper develops a deep learning tool based on neural processes(NPs)called the Peri-Net-Pro,to predict the crack patterns in a moving disk and classifies them according to the classification modes with quantified uncertainties.In particular,image classification and regression studies are conducted by means of convolutional neural networks(CNNs)and NPs.First,the amount and quality of the data are enhanced by using peridynamics to theoretically compensate for the problems of the finite element method(FEM)in generating crack pattern images.Second,case studies are conducted with the prototype microelastic brittle(PMB),linear peridynamic solid(LPS),and viscoelastic solid(VES)models obtained by using the peridynamic theory.The case studies are performed to classify the images by using CNNs and determine the suitability of the PMB,LBS,and VES models.Finally,a regression analysis is performed on the crack pattern images with NPs to predict the crack patterns.The regression analysis results confirm that the variance decreases when the number of epochs increases by using the NPs.The training results gradually improve,and the variance ranges decrease to less than 0.035.The main finding of this study is that the NPs enable accurate predictions,even with missing or insufficient training data.The results demonstrate that if the context points are set to the 10th,100th,300th,and 784th,the training information is deliberately omitted for the context points of the 10th,100th,and 300th,and the predictions are different when the context points are significantly lower.However,the comparison of the results of the 100th and 784th context points shows that the predicted results are similar because of the Gaussian processes in the NPs.Therefore,if the NPs are employed for training,the missing information of the training data can be supplemented to predict the results.展开更多
In China, there are approximately 20 million people suffering from peripheral nerve injury and this number is increasing at a rate of 2 million per year. These patients cannot live or work independently and are a heav...In China, there are approximately 20 million people suffering from peripheral nerve injury and this number is increasing at a rate of 2 million per year. These patients cannot live or work independently and are a heavy responsibility on both family and society because of extreme disability and dysfunction caused by peripheral nerve injury （PNI）. Thus, repair of PNI has become a major public health issue in China.展开更多
This paper presents a novel method for inferring the odor based on neural activities observed from rats' main olfactory bulbs.Multi-channel extra-cellular single unit recordings are done by micro-wire electrodes(T...This paper presents a novel method for inferring the odor based on neural activities observed from rats' main olfactory bulbs.Multi-channel extra-cellular single unit recordings are done by micro-wire electrodes(Tungsten,50 μm,32 channels)implanted in the mitral/tufted cell layers of the main olfactory bulb of the anesthetized rats to obtain neural responses to various odors.Neural responses as a key feature are measured by subtraction firing rates before stimulus from after.For odor inference,a decoding method is developed based on the ML estimation.The results show that the average decoding accuracy is about 100.0%,96.0%,and 80.0% with three rats,respectively.This work has profound implications for a novel brain-machine interface system for odor inference.展开更多
Predicting the time-varying auto-spectral density of a spacecraft in high-altitude orbits requires an accurate model for the non-stationary random vibration signals with densely spaced modal frequency. The traditional...Predicting the time-varying auto-spectral density of a spacecraft in high-altitude orbits requires an accurate model for the non-stationary random vibration signals with densely spaced modal frequency. The traditional time-varying algorithm limits prediction accuracy, thus affecting a number of operational decisions. To solve this problem, a time-varying auto regressive （TVAR） model based on the process neural network （PNN） and the empirical mode decomposition （EMD） is proposed. The time-varying system is tracked on-line by establishing a time-varying parameter model, and then the relevant parameter spectrum is obtained. Firstly, the EMD method is utilized to decompose the signal into several intrinsic mode functions （IMFs）. Then for each IMF, the PNN is established and the time-varying auto-spectral density is obtained. Finally, the time-frequency distribution of the signals can be reconstructed by linear superposition. The simulation and the analytical results from an example demonstrate that this approach possesses simplicity, effectiveness, and feasibility, as well as higher frequency resolution.展开更多
Because there is neither waste rock nor mill tailings in the gypsum mine, and the buildings on the goaf of gypsum mine are needed to be protected, the research proposed the scheme of the clay filling technology. Gypsu...Because there is neither waste rock nor mill tailings in the gypsum mine, and the buildings on the goaf of gypsum mine are needed to be protected, the research proposed the scheme of the clay filling technology. Gypsum, cement, lime and water glass were used as adhesive, and the strength of different material ratios were investigated in this study. The influence factors of clay strength were obtained in the order of cement, gypsum, water glass and lime. The results show that the cement content is the determinant influence factor, and gypsum has positive effects, while the water glass can enhance both clay strength and the fluidity of the filing slurry. Furthermore, combining chaotic optimization method with neural network, the optimal ratio of composite cementing agent was obtained. The results show that the optimal ratio of water glass, cement, lime and clay (in quality) is 1.17:6.74:4.17:87.92 in the process of bottom self-flow filling, while the optimal ratio is 1.78:9.58:4.71:83.93 for roof-contacted filling. A novel filling process to fill in gypsum mine goaf with clay is established. The engineering practice shows that the filling cost is low, thus, notable economic benefit is achieved.展开更多
The effluent total phosphorus(ETP) is an important parameter to evaluate the performance of wastewater treatment process(WWTP). In this study, a novel method, using a data-derived soft-sensor method, is proposed to ob...The effluent total phosphorus(ETP) is an important parameter to evaluate the performance of wastewater treatment process(WWTP). In this study, a novel method, using a data-derived soft-sensor method, is proposed to obtain the reliable values of ETP online. First, a partial least square(PLS) method is introduced to select the related secondary variables of ETP based on the experimental data. Second, a radial basis function neural network(RBFNN) is developed to identify the relationship between the related secondary variables and ETP. This RBFNN easily optimizes the model parameters to improve the generalization ability of the soft-sensor. Finally, a monitoring system, based on the above PLS and RBFNN, named PLS-RBFNN-based soft-sensor system, is developed and tested in a real WWTP. Experimental results show that the proposed monitoring system can obtain the values of ETP online and own better predicting performance than some existing methods.展开更多
Three recent breakthroughs due to AI in arts and science serve as motivation:An award winning digital image,protein folding,fast matrix multiplication.Many recent developments in artificial neural networks,particularl...Three recent breakthroughs due to AI in arts and science serve as motivation:An award winning digital image,protein folding,fast matrix multiplication.Many recent developments in artificial neural networks,particularly deep learning(DL),applied and relevant to computational mechanics(solid,fluids,finite-element technology)are reviewed in detail.Both hybrid and pure machine learning(ML)methods are discussed.Hybrid methods combine traditional PDE discretizations with ML methods either(1)to help model complex nonlinear constitutive relations,(2)to nonlinearly reduce the model order for efficient simulation(turbulence),or(3)to accelerate the simulation by predicting certain components in the traditional integration methods.Here,methods(1)and(2)relied on Long-Short-Term Memory(LSTM)architecture,with method(3)relying on convolutional neural networks.Pure ML methods to solve(nonlinear)PDEs are represented by Physics-Informed Neural network(PINN)methods,which could be combined with attention mechanism to address discontinuous solutions.Both LSTM and attention architectures,together with modern and generalized classic optimizers to include stochasticity for DL networks,are extensively reviewed.Kernel machines,including Gaussian processes,are provided to sufficient depth for more advanced works such as shallow networks with infinite width.Not only addressing experts,readers are assumed familiar with computational mechanics,but not with DL,whose concepts and applications are built up from the basics,aiming at bringing first-time learners quickly to the forefront of research.History and limitations of AI are recounted and discussed,with particular attention at pointing out misstatements or misconceptions of the classics,even in well-known references.Positioning and pointing control of a large-deformable beam is given as an example.展开更多
Flexible sensing array integrated with multiple sensors is an attractive approach for flight parameter detection.However,the poor resolution of flexible sensors and time-consuming neural network processes mitigate the...Flexible sensing array integrated with multiple sensors is an attractive approach for flight parameter detection.However,the poor resolution of flexible sensors and time-consuming neural network processes mitigate their accuracy and adaptability in predicting flight parameters.Here we present an ultra-thin flexible sensing patch with a new configuration,comprising a differential pressure sensor array and a vector flow velocity sensor.The capacitive differential pressure sensor array is fabricated by a multilayer polyimide bonding technique,reaching a resolution of 0.14 Pa.To solve flight parameters with the flexible sensing patch,we develop an analytical pressure-velocity fusion algorithm,enabling fast response and high accuracy in flight parameter detection.The average errors in calculating the angle of attack,angle of sideslip,and airspeed are 0.22°,0.35°,and 0.73 m s^(-1),respectively.The high-resolution flexible sensors and novel analytical pressure-velocity fusion algorithm pave the way for flexible sensing patch-based air data sensing techniques.展开更多
The data-driven Intelligent Transportation System(ITS)provides great support to travel decisions and system management but inevitably encounters the issue of data missing in monitoring systems.Hence,network-wide traff...The data-driven Intelligent Transportation System(ITS)provides great support to travel decisions and system management but inevitably encounters the issue of data missing in monitoring systems.Hence,network-wide traffic state prediction and imputation is critical to recognizing the system level state of a transportation network.Abundant research works have adopted various approaches for traffic prediction and imputation.However,previous methods ignore the reliability analysis of the predicted/imputed traffic information.Thus,this study originally proposes an attentive graph neural process(AGNP)method for network-level short-term traffic speed prediction and imputation,simultaneously considering reliability.Firstly,the Gaussian process(GP)is used to model the observed traffic speed state.Such a stochastic process is further learned by the proposed AGNP method,which is utilized for inferring the congestion state on the remaining unobserved road segments.Data from a transportation network in Anhui Province,China,is used to conduct three experiments with increasing missing data ratio for model testing.Based on comparisons against other machine learning models,the results show that the proposed AGNP model can impute traffic networks and predict traffic speed with high-level performance.With the probabilistic confidence provided by the AGNP,reliability analysis is conducted both numerically and visually to show that the predicted distributions are beneficial to guide traffic control strategies and travel plans.展开更多
Natural perceptual systems,shaped by millions of years of evolution,exhibit unparalleled environmental adaptability and energy-efficient information processing that surpass existing engineering technologies[[1],[2],[3...Natural perceptual systems,shaped by millions of years of evolution,exhibit unparalleled environmental adaptability and energy-efficient information processing that surpass existing engineering technologies[[1],[2],[3]].Particularly,the avian visual system of raptors-especially eagles has emerged as a paradigm for bioinspired optoelectronics,owing to its integrated multispectral sensing,dynamic range modulation,and hierarchical neural processing[4].展开更多
基金supported by the National Natural Science Foundation of China(Grant No.61827817)。
文摘The photonic neural processing unit(PNPU)demonstrates ultrahigh inference speed with low energy consumption,and it has become a promising hardware artificial intelligence(AI)accelerator.However,the nonidealities of the photonic device and the peripheral circuit make the practical application much more complex.Rather than optimizing the photonic device,the architecture,and the algorithm individually,a joint device-architecture-algorithm codesign method is proposed to improve the accuracy,efficiency and robustness of the PNPU.First,a full-flow simulator for the PNPU is developed from the back end simulator to the high-level training framework;Second,the full system architecture and the complete photonic chip design enable the simulator to closely model the real system;Third,the nonidealities of the photonic chip are evaluated for the PNPU design.The average test accuracy exceeds 98%,and the computing power exceeds 100TOPS.
文摘Background Visual perception of face images or face pareidolia can be evaluated with event-related potentials(ERP)for healthy subjects and patients with neurological conditions.In this study,we aimed to analyse event-related potential components such as P100,N100,N170,and vertex-positive potential(VPP)in response to face pareidolia perception in temporal lobe epilepsy(TLE)patients.Methods ERPs were recorded during the pareidolia test.Waveforms were analzyed and current source density(CSD)maps were generated.Results CSD profles were shown to be interpretable when face and face pareidolia conditions.N100,P100,and N170 components showed larger amplitudes and longer latency in epilepsy patients in response to face pareidolia stimuli compared to real face images.However,the N170 component latency did not difer signifcantly between epilepsy patients and healthy participants,while the larger amplitude and longer latency of N100 and P100 responses were evoked in healthy patients.Conclusions Our results indicate a diference in the neural mechanisms of processing real face information and pareidolia face-like information in TLE patients.
基金Project supported by the National Natural Science Foundation of China (Grant No 60572174)the Doctoral Fund of Ministry of Education of China (Grant No 20070213072)+2 种基金the 111 Project (Grant No B07018)the China Postdoctoral Science Foundation (Grant No 20070410264)the Development Program for Outstanding Young Teachers in Harbin Institute of Technology (Grant No HITQNJS.2007.010)
文摘In the real world, the inputs of many complicated systems are time-varying functions or processes. In order to predict the outputs of these systems with high speed and accuracy, this paper proposes a time series prediction model based on the wavelet process neural network, and develops the corresponding learning algorithm based on the expansion of the orthogonal basis functions. The effectiveness of the proposed time series prediction model and its learning algorithm is proved by the Macke-Glass time series prediction, and the comparative prediction results indicate that the proposed time series prediction model based on the wavelet process neural network seems to perform well and appears suitable for using as a good tool to predict the highly complex nonlinear time series.
基金Project supported by the National Science Foundation of U.S.A.(Nos.DMS-1555072,DMS-2053746DMS-2134209)+1 种基金the Brookhaven National Laboratory of U.S.A.(No.382247)U.S.Department of Energy(DOE)Office of Science Advanced Scientific Computing Research Program(Nos.DESC0021142 and DE-SC0023161)。
文摘This paper develops a deep learning tool based on neural processes(NPs)called the Peri-Net-Pro,to predict the crack patterns in a moving disk and classifies them according to the classification modes with quantified uncertainties.In particular,image classification and regression studies are conducted by means of convolutional neural networks(CNNs)and NPs.First,the amount and quality of the data are enhanced by using peridynamics to theoretically compensate for the problems of the finite element method(FEM)in generating crack pattern images.Second,case studies are conducted with the prototype microelastic brittle(PMB),linear peridynamic solid(LPS),and viscoelastic solid(VES)models obtained by using the peridynamic theory.The case studies are performed to classify the images by using CNNs and determine the suitability of the PMB,LBS,and VES models.Finally,a regression analysis is performed on the crack pattern images with NPs to predict the crack patterns.The regression analysis results confirm that the variance decreases when the number of epochs increases by using the NPs.The training results gradually improve,and the variance ranges decrease to less than 0.035.The main finding of this study is that the NPs enable accurate predictions,even with missing or insufficient training data.The results demonstrate that if the context points are set to the 10th,100th,300th,and 784th,the training information is deliberately omitted for the context points of the 10th,100th,and 300th,and the predictions are different when the context points are significantly lower.However,the comparison of the results of the 100th and 784th context points shows that the predicted results are similar because of the Gaussian processes in the NPs.Therefore,if the NPs are employed for training,the missing information of the training data can be supplemented to predict the results.
基金supported by grants from the National Program on Key Basic Research Project of China(973 Program),No.2014CB542200Program for Innovative Research Team in University of Ministry of Education of China,No.IRT1201+1 种基金the National Natural Science Foundation of China,No.31271284,31171150,81171146,30971526,31100860,31040043Program for New Century Excellent Talents in University of Ministry of Education of China,No.BMU20110270
文摘In China, there are approximately 20 million people suffering from peripheral nerve injury and this number is increasing at a rate of 2 million per year. These patients cannot live or work independently and are a heavy responsibility on both family and society because of extreme disability and dysfunction caused by peripheral nerve injury （PNI）. Thus, repair of PNI has become a major public health issue in China.
基金supported by the MKE(The Ministry of Knowledge Economy,Korea)theITRC(Information Technology Research Center)support program(NIPA-2010-C1090-1021-0010)
文摘This paper presents a novel method for inferring the odor based on neural activities observed from rats' main olfactory bulbs.Multi-channel extra-cellular single unit recordings are done by micro-wire electrodes(Tungsten,50 μm,32 channels)implanted in the mitral/tufted cell layers of the main olfactory bulb of the anesthetized rats to obtain neural responses to various odors.Neural responses as a key feature are measured by subtraction firing rates before stimulus from after.For odor inference,a decoding method is developed based on the ML estimation.The results show that the average decoding accuracy is about 100.0%,96.0%,and 80.0% with three rats,respectively.This work has profound implications for a novel brain-machine interface system for odor inference.
基金Aeronautical Science Foundation of China (20071551016)
文摘Predicting the time-varying auto-spectral density of a spacecraft in high-altitude orbits requires an accurate model for the non-stationary random vibration signals with densely spaced modal frequency. The traditional time-varying algorithm limits prediction accuracy, thus affecting a number of operational decisions. To solve this problem, a time-varying auto regressive （TVAR） model based on the process neural network （PNN） and the empirical mode decomposition （EMD） is proposed. The time-varying system is tracked on-line by establishing a time-varying parameter model, and then the relevant parameter spectrum is obtained. Firstly, the EMD method is utilized to decompose the signal into several intrinsic mode functions （IMFs）. Then for each IMF, the PNN is established and the time-varying auto-spectral density is obtained. Finally, the time-frequency distribution of the signals can be reconstructed by linear superposition. The simulation and the analytical results from an example demonstrate that this approach possesses simplicity, effectiveness, and feasibility, as well as higher frequency resolution.
基金supported by the National Basic Research and Development Program of China (No. 2010CB732004)the joint funding of the National Natural Science Foundation and Shanghai Baosteel Group Corporation of China (No. 51074177)
文摘Because there is neither waste rock nor mill tailings in the gypsum mine, and the buildings on the goaf of gypsum mine are needed to be protected, the research proposed the scheme of the clay filling technology. Gypsum, cement, lime and water glass were used as adhesive, and the strength of different material ratios were investigated in this study. The influence factors of clay strength were obtained in the order of cement, gypsum, water glass and lime. The results show that the cement content is the determinant influence factor, and gypsum has positive effects, while the water glass can enhance both clay strength and the fluidity of the filing slurry. Furthermore, combining chaotic optimization method with neural network, the optimal ratio of composite cementing agent was obtained. The results show that the optimal ratio of water glass, cement, lime and clay (in quality) is 1.17:6.74:4.17:87.92 in the process of bottom self-flow filling, while the optimal ratio is 1.78:9.58:4.71:83.93 for roof-contacted filling. A novel filling process to fill in gypsum mine goaf with clay is established. The engineering practice shows that the filling cost is low, thus, notable economic benefit is achieved.
基金Supported by the National Science Foundation of China(61622301,61533002)Beijing Natural Science Foundation(4172005)Major National Science and Technology Project(2017ZX07104)
文摘The effluent total phosphorus(ETP) is an important parameter to evaluate the performance of wastewater treatment process(WWTP). In this study, a novel method, using a data-derived soft-sensor method, is proposed to obtain the reliable values of ETP online. First, a partial least square(PLS) method is introduced to select the related secondary variables of ETP based on the experimental data. Second, a radial basis function neural network(RBFNN) is developed to identify the relationship between the related secondary variables and ETP. This RBFNN easily optimizes the model parameters to improve the generalization ability of the soft-sensor. Finally, a monitoring system, based on the above PLS and RBFNN, named PLS-RBFNN-based soft-sensor system, is developed and tested in a real WWTP. Experimental results show that the proposed monitoring system can obtain the values of ETP online and own better predicting performance than some existing methods.
文摘Three recent breakthroughs due to AI in arts and science serve as motivation:An award winning digital image,protein folding,fast matrix multiplication.Many recent developments in artificial neural networks,particularly deep learning(DL),applied and relevant to computational mechanics(solid,fluids,finite-element technology)are reviewed in detail.Both hybrid and pure machine learning(ML)methods are discussed.Hybrid methods combine traditional PDE discretizations with ML methods either(1)to help model complex nonlinear constitutive relations,(2)to nonlinearly reduce the model order for efficient simulation(turbulence),or(3)to accelerate the simulation by predicting certain components in the traditional integration methods.Here,methods(1)and(2)relied on Long-Short-Term Memory(LSTM)architecture,with method(3)relying on convolutional neural networks.Pure ML methods to solve(nonlinear)PDEs are represented by Physics-Informed Neural network(PINN)methods,which could be combined with attention mechanism to address discontinuous solutions.Both LSTM and attention architectures,together with modern and generalized classic optimizers to include stochasticity for DL networks,are extensively reviewed.Kernel machines,including Gaussian processes,are provided to sufficient depth for more advanced works such as shallow networks with infinite width.Not only addressing experts,readers are assumed familiar with computational mechanics,but not with DL,whose concepts and applications are built up from the basics,aiming at bringing first-time learners quickly to the forefront of research.History and limitations of AI are recounted and discussed,with particular attention at pointing out misstatements or misconceptions of the classics,even in well-known references.Positioning and pointing control of a large-deformable beam is given as an example.
基金supported financially by the National Natural Science Foundation of China(T2121003 received by X.D.,and U23A20638 received by Y.J.)the National Key Research and Development Program of China(2023YFB3208000 and 2023YFB3208001 received by Y.J.).
文摘Flexible sensing array integrated with multiple sensors is an attractive approach for flight parameter detection.However,the poor resolution of flexible sensors and time-consuming neural network processes mitigate their accuracy and adaptability in predicting flight parameters.Here we present an ultra-thin flexible sensing patch with a new configuration,comprising a differential pressure sensor array and a vector flow velocity sensor.The capacitive differential pressure sensor array is fabricated by a multilayer polyimide bonding technique,reaching a resolution of 0.14 Pa.To solve flight parameters with the flexible sensing patch,we develop an analytical pressure-velocity fusion algorithm,enabling fast response and high accuracy in flight parameter detection.The average errors in calculating the angle of attack,angle of sideslip,and airspeed are 0.22°,0.35°,and 0.73 m s^(-1),respectively.The high-resolution flexible sensors and novel analytical pressure-velocity fusion algorithm pave the way for flexible sensing patch-based air data sensing techniques.
基金supported by“Pioneer”and“Leading Goose”R&D Program of Zhejiang(No.2023C03155)Hong Kong Research Grants Council(Nos.HKUST16208920 and T41-603/20R)+1 种基金National Natural Science Foundation of China(Nos.71922019 and 72171210)the Smart Urban Future(SURF)Laboratory,Zhejiang Province.
文摘The data-driven Intelligent Transportation System(ITS)provides great support to travel decisions and system management but inevitably encounters the issue of data missing in monitoring systems.Hence,network-wide traffic state prediction and imputation is critical to recognizing the system level state of a transportation network.Abundant research works have adopted various approaches for traffic prediction and imputation.However,previous methods ignore the reliability analysis of the predicted/imputed traffic information.Thus,this study originally proposes an attentive graph neural process(AGNP)method for network-level short-term traffic speed prediction and imputation,simultaneously considering reliability.Firstly,the Gaussian process(GP)is used to model the observed traffic speed state.Such a stochastic process is further learned by the proposed AGNP method,which is utilized for inferring the congestion state on the remaining unobserved road segments.Data from a transportation network in Anhui Province,China,is used to conduct three experiments with increasing missing data ratio for model testing.Based on comparisons against other machine learning models,the results show that the proposed AGNP model can impute traffic networks and predict traffic speed with high-level performance.With the probabilistic confidence provided by the AGNP,reliability analysis is conducted both numerically and visually to show that the predicted distributions are beneficial to guide traffic control strategies and travel plans.
基金supported by the National Key Research and Development Program of China(2021YFA0717900)the National Natural Science Foundation of China(52273190 and 52121002).
文摘Natural perceptual systems,shaped by millions of years of evolution,exhibit unparalleled environmental adaptability and energy-efficient information processing that surpass existing engineering technologies[[1],[2],[3]].Particularly,the avian visual system of raptors-especially eagles has emerged as a paradigm for bioinspired optoelectronics,owing to its integrated multispectral sensing,dynamic range modulation,and hierarchical neural processing[4].
