Improvement And Application Of Echo State Network Based On Fuzzy Hyperbolic Model

The echo state network as an improved recurrent neural network is proposed. The basic idea is to use a reservoir to replace the hidden layer, which can express the input by the high-dimensional and nonlinear way. The generation process of reservoir is independent of the training process, thus only output weights need to be trained. In order to improve the performance of echo state network, this paper mainly studies the relationship between the parameters, and to carry out the research on the improvement of echo state network.First, the leaky integrator echo state network(Leaky-ESN), as an improved echo state network is proposed. The Leaky-ESN exists a complex nonlinear relationship between parameters, and then setting parameter values is very difficult because of lacking effective guiding principles. Therefore, in order to further enhance modeling ability, we study the relationship between the size, sparsity of the reservoir and the modeling accuracy. This paper not only gives the qualitative description of relationship, but also gives the quantitative description by the least squares fitting method. Second, we proposed an improved echo state network based on fuzzy hyperbolic model(FH-ESN). This model combines the advantages of two kinds of networks. Because the state equation of the two models is similar. We use fuzzy hyperbolic model to explain the useful information of the echo state network. This can make the echo state network have interpretability. And this method improves the accuracy of modeling. The FH-ESN fully reflects the influence of the change of each signal to the current state. Finally, we use the improved echo state network to predict the photovoltaic power generation. Compared with the standard echo state network, the proposed method has higher prediction accuracy.