Optimization of echo state network behavior space based on microbial genetic algorithm

Aiming at the difficulty of selecting the parameters of the echo state network reservoir, this paper proposes a method for optimizing ESN parameters based on behavior space. Its essence is to build an ESN behavior space through generalization rank, kernel rank, and memory capacity, and use microbial genetic algorithms to find rules for reservoir parameter selection and the minimum behavior configuration required for learning tasks. This method overcomes the shortcomings of traditional ESN such as multiple parameters and long optimization time, and improves optimization efficiency and network learning performance. The experimental results show that the MGA-ESN method proposed in this paper is basically close to the optimal network structure, the learning performance is better than the GESN network, and the reasons that affect the performance of the ESN network can be explained through the behavior space.