Prediction And Reconstruction Of Chaotic Time Series Based On Hybrid Artificial Neural Network

Chaotic time series data exist in many fields,such as meteorology,medicine and ecology.Extracting the characteristics of chaotic time series to predict and reconstruct it can discover the essence of system evolution and the laws behind it,which is of great scientific and technical significance to society.Due to the great nonlinear characteristics of chaotic time series,it is difficult to predict and reconstruct chaotic time series.Due to their strong nonlinear learning ability,recurrent neural network and convolutional neural network have gradually become the focus of chaotic time series forecasting research.However,due to the inherent structural characteristics of the network,the RNN model faces problems such as gradient disappearance and unfavorable parallel computing,which affects the training speed and accuracy.The traditional convolutional neural network model is mainly designed for image data processing,so the prediction of chaotic time series using the parallel capability of convolutional neural network still needs further research.In addition to predicting chaotic time series,it can also be reconstructed.The reconstruction of the chaos generated by the stable chaos generator can be applied in the fields of chaos security communication and so on.Considering that silicon-based optical chaos has the advantages of compatibility with complementary metal-oxide-semiconductor integration processes,ultra-small size,and high bandwidth.Therefore,using a silicon-based optical chaos generator to generate stable chaotic time series data is a good chaos generation scheme.In addition,using neural networks to reconstruct chaos can reduce the cost and complexity of chaos source synchronization.However,to map the neural network to computing devices,it is necessary to consider the limited computing and storage resources of embedded and mobile devices,so the neural network model used for chaos reconstruction needs to have a simple structure.In order to improve the accuracy and speed of chaotic time series prediction,this thesis proposes a TCN-CBAM neural network model based on Temporal Convolutional Network(TCN)and Convolutional Block Attention Module(CBAM)for Chaotic time series forecasting.The interior of TCN is composed of a large number of special convolution calculations,using techniques including causal convolution,dilated convolution and residual connection.CBAM combines the spatial attention mechanism and the channel attention mechanism,which is beneficial to improve the prediction accuracy.In this thesis,the proposed TCN-CBAM model is used to conduct various comparative experiments in multiple classical chaotic systems(Chen chaotic system,Lorenz chaotic system and sunspots),and it is better than the comparison model in terms of prediction results.In addition,the TCN-CBAM model also has the shortest training time among the comparison models.In addition,this thesis proposes a stacked hybrid neural network for high-precision reconstruction of silicon-based optical chaos.The network combines the advantages of convolutional neural network and long short-term memory network,and uses different types of network models to extract the features of chaotic time series,and then integrates these two different features through full connection.In this thesis,a theoretical model of silicon-based optical chaos is used to generate chaotic time series data.And the proposed stacked hybrid neural network model is used for the chaotic reconstruction task.In the experiments in this thesis,the one-dimensional,two-dimensional and three-dimensional chaos are reconstructed respectively.Experimental results show that the proposed model outperforms other comparative models in multiple indicators.