Dynamics is a key issue about understanding recurrent neural networks(RNNs).Because of the complexity,the problem still remains unanswered in spite of many important progresses.Echo state network(ESN)is a simple appro...Dynamics is a key issue about understanding recurrent neural networks(RNNs).Because of the complexity,the problem still remains unanswered in spite of many important progresses.Echo state network(ESN)is a simple approach to design RNNs.It is possible to investigate ESNs’dynamics deeply.However,most of dynamic studies have mainly concentrated on the shallow ESNs and seldom of them explain the dynamics of the deep ones.Therefore,this paper investigates the dynamics of four typical ESNs under a unified theoretical framework.These ESNs contain both the shallow versions and the deep ones.This investigation is helpful to clarify the dynamics of ESNs in a general sense.Also,the short-term memory(STM)of different ESNs is analyzed,which is closely related to the dynamics.This analysis is helpful to determine the hyper-parameters of ESNs for given problems.In addition,the problem-solving abilities of ESNs are investigated through modeling two time series tasks.It further explains the influence of the dynamics on ESN’s performance.展开更多
An accurate landslide displacement prediction is an important part of landslide warning system. Aiming at the dynamic characteristics of landslide evolution and the shortcomings of traditional static prediction models...An accurate landslide displacement prediction is an important part of landslide warning system. Aiming at the dynamic characteristics of landslide evolution and the shortcomings of traditional static prediction models, this paper proposes a dynamic prediction model of landslide displacement based on singular spectrum analysis(SSA) and stack long short-term memory(SLSTM) network. The SSA is used to decompose the landslide accumulated displacement time series data into trend term and periodic term displacement subsequences. A cubic polynomial function is used to predict the trend term displacement subsequence, and the SLSTM neural network is used to predict the periodic term displacement subsequence. At the same time, the Bayesian optimization algorithm is used to determine that the SLSTM network input sequence length is 12 and the number of hidden layer nodes is 18. The SLSTM network is updated by adding predicted values to the training set to achieve dynamic displacement prediction. Finally, the accumulated landslide displacement is obtained by superimposing the predicted value of each displacement subsequence. The proposed model was verified on the Xintan landslide in Hubei Province, China. The results show that when predicting the displacement of the periodic term, the SLSTM network has higher prediction accuracy than the support vector machine(SVM) and auto regressive integrated moving average(ARIMA). The mean relative error(MRE) is reduced by 4.099% and 3.548% respectively, while the root mean square error(RMSE) is reduced by 5.830 mm and 3.854 mm respectively. It is concluded that the SLSTM network model can better simulate the dynamic characteristics of landslides.展开更多
As a high efficiency hydrogen-to-power device,proton exchange membrane fuel cell(PEMFC)attracts much attention,especially for the automotive applications.Real-time prediction of output voltage and area specific resist...As a high efficiency hydrogen-to-power device,proton exchange membrane fuel cell(PEMFC)attracts much attention,especially for the automotive applications.Real-time prediction of output voltage and area specific resistance(ASR)via the on-board model is critical to monitor the health state of the automotive PEMFC stack.In this study,we use a transient PEMFC system model for dynamic process simulation of PEMFC to generate the dataset,and a long short-term memory(LSTM)deep learning model is developed to predict the dynamic per-formance of PEMFC.The results show that the developed LSTM deep learning model has much better perfor-mance than other models.A sensitivity analysis on the input features is performed,and three insensitive features are removed,that could slightly improve the prediction accuracy and significantly reduce the data volume.The neural structure,sequence duration,and sampling frequency are optimized.We find that the optimal sequence data duration for predicting ASR is 5 s or 20 s,and that for predicting output voltage is 40 s.The sampling frequency can be reduced from 10 Hz to 0.5 Hz and 0.25 Hz,which slightly affects the prediction accuracy,but obviously reduces the data volume and computation amount.展开更多
基金Sponsored by the Shandong Provincial Natural Science Foundation(Grant No.ZR2021MF105).
文摘Dynamics is a key issue about understanding recurrent neural networks(RNNs).Because of the complexity,the problem still remains unanswered in spite of many important progresses.Echo state network(ESN)is a simple approach to design RNNs.It is possible to investigate ESNs’dynamics deeply.However,most of dynamic studies have mainly concentrated on the shallow ESNs and seldom of them explain the dynamics of the deep ones.Therefore,this paper investigates the dynamics of four typical ESNs under a unified theoretical framework.These ESNs contain both the shallow versions and the deep ones.This investigation is helpful to clarify the dynamics of ESNs in a general sense.Also,the short-term memory(STM)of different ESNs is analyzed,which is closely related to the dynamics.This analysis is helpful to determine the hyper-parameters of ESNs for given problems.In addition,the problem-solving abilities of ESNs are investigated through modeling two time series tasks.It further explains the influence of the dynamics on ESN’s performance.
基金supported by the Natural Science Foundation of Shaanxi Province under Grant 2019JQ206in part by the Science and Technology Department of Shaanxi Province under Grant 2020CGXNG-009in part by the Education Department of Shaanxi Province under Grant 17JK0346。
文摘An accurate landslide displacement prediction is an important part of landslide warning system. Aiming at the dynamic characteristics of landslide evolution and the shortcomings of traditional static prediction models, this paper proposes a dynamic prediction model of landslide displacement based on singular spectrum analysis(SSA) and stack long short-term memory(SLSTM) network. The SSA is used to decompose the landslide accumulated displacement time series data into trend term and periodic term displacement subsequences. A cubic polynomial function is used to predict the trend term displacement subsequence, and the SLSTM neural network is used to predict the periodic term displacement subsequence. At the same time, the Bayesian optimization algorithm is used to determine that the SLSTM network input sequence length is 12 and the number of hidden layer nodes is 18. The SLSTM network is updated by adding predicted values to the training set to achieve dynamic displacement prediction. Finally, the accumulated landslide displacement is obtained by superimposing the predicted value of each displacement subsequence. The proposed model was verified on the Xintan landslide in Hubei Province, China. The results show that when predicting the displacement of the periodic term, the SLSTM network has higher prediction accuracy than the support vector machine(SVM) and auto regressive integrated moving average(ARIMA). The mean relative error(MRE) is reduced by 4.099% and 3.548% respectively, while the root mean square error(RMSE) is reduced by 5.830 mm and 3.854 mm respectively. It is concluded that the SLSTM network model can better simulate the dynamic characteristics of landslides.
基金This research is supported by the National Natural Science Founda-tion of China(No.52176196)the National Key Research and Devel-opment Program of China(No.2022YFE0103100)+1 种基金the China Postdoctoral Science Foundation(No.2021TQ0235)the Hong Kong Scholars Program(No.XJ2021033).
文摘As a high efficiency hydrogen-to-power device,proton exchange membrane fuel cell(PEMFC)attracts much attention,especially for the automotive applications.Real-time prediction of output voltage and area specific resistance(ASR)via the on-board model is critical to monitor the health state of the automotive PEMFC stack.In this study,we use a transient PEMFC system model for dynamic process simulation of PEMFC to generate the dataset,and a long short-term memory(LSTM)deep learning model is developed to predict the dynamic per-formance of PEMFC.The results show that the developed LSTM deep learning model has much better perfor-mance than other models.A sensitivity analysis on the input features is performed,and three insensitive features are removed,that could slightly improve the prediction accuracy and significantly reduce the data volume.The neural structure,sequence duration,and sampling frequency are optimized.We find that the optimal sequence data duration for predicting ASR is 5 s or 20 s,and that for predicting output voltage is 40 s.The sampling frequency can be reduced from 10 Hz to 0.5 Hz and 0.25 Hz,which slightly affects the prediction accuracy,but obviously reduces the data volume and computation amount.
