Application of ESN prediction model based on compressed sensing in stock market

Abstract

Echo State Network (ESN) is often used for time series prediction and chaotic series prediction.But in the training process of ESN model, the reserve pool will involve a large number of calculations, so the reserve pool may have node redundancy, resulting in inaccurate training model. It is very important to select the active nodes in the reserve pool because the predicted nodes are very few compared with all the nodes in the reserve pool. Based on the characteristics of the original vector, compressed sensing can reduce the dimension of the vector so as to obtain the activated nodes. In this paper, the ESN model is added with compressed sensing, which simplifies the reserve pool part of the ESN model. The simulation results show that our method can not only reduce the calculation amount of training effectively, but also improve the accuracy.