Echo state network optimization method and device based on adaptive backtracking search

Abstract

The embodiment of the invention provides an echo state network optimization method and echo state network optimization equipment based on adaptive backtracking search. The method comprises the following steps: initializing parameters of an adaptive backtracking search algorithm, acquiring a plurality of populations, forming a connection weight matrix of an echo state network by using a certain individual in the plurality of populations, and acquiring an output value and an adaptive value of the echo state network according to the connection weight matrix and the state matrix; if the current population algebra is smaller than the maximum population algebra, a self-adaptive backtracking search algorithm is adopted to obtain an experimental population and a certain individual in the experimental population, another connection weight matrix is formed, and the output value and the adaptive value of the echo state network are obtained according to the connection weight matrix and the state matrix; and if the adaptive value of the output value is smaller than the adaptive value of the previous output value, replacing the previous individual with the next individual to optimize the echo state network. According to the embodiment of the invention, the system identification and time sequence prediction capability of the echo state network can be improved.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of neural networks, in particular to a method and device for optimizing an echo state network based on adaptive backtracking search. Background technique [0002] Artificial Neural Network (ANN) is a data-driven, adaptive artificial intelligence model. ANN is very suitable for modeling time series problems, mainly because ANN has the following characteristics: (1) Different from traditional mathematical modeling methods, ANN is an adaptive modeling method driven by data; (2) It is different from traditional mathematical modeling methods. Compared with regression-based methods, ANN has better model generalization ability; (3) ANN can theoretically approach any dynamic system infinitely with arbitrary precision, and has good nonlinear mapping ability. The powerful approximation ability makes ANN widely used in time series forecasting problems. However, in the face of specific time serie...

AI Technical Summary

Problems solved by technology

Although the linear regression method can save the cost of network training, it will lead to the problem of model overfitting during the training process of the network, thus affecting the prediction accuracy

Images