Research On Parameter Optimization Method Of Reservoir Computing Model

Echo state network adopts large-scale sparsely connected reservoir as the hidden layer and only adjusts the output connection weight,which makes the whole training stage simple and efficient and becomes an important processing method for nonlinear time series prediction task.However,when the traditional reservoir parameter is initialized randomly according to the experience,it is easy to fall into the local optimal value,which degrades the network prediction performance.Taking it as the research object,the parameter optimization method of this model is studied from the perspective of swarm intelligence optimization.The main research contents are as follows:1)To solve the problem of poor convergence and slow convergence of particle swarm optimization.The echo state network based on ant lion optimizer is proposed to select the optimal input,reservoir,feedback weight parameter,through the process of ant lion catching ants,to improve the prediction performance of echo state network.In chaotic time series prediction task,the simulation results showed that the model had obvious advantages in nonlinear approximation performance and convergence speed compared with particle swarm optimization.2)To solve the problem that it is difficult to get the optimal value for the ant lion optimizer with too many parameters,the teaching-learning-based optimization algorithm is applied to optimize the input,reservoir,feedback weight parameter,and the weight parameters are updated through the two stages of mutual learning between teachers and students,so as to improve the prediction performance and retain part of the randomness of the parameters.In chaotic time series prediction task,the simulation results showed that the model had obvious advantages in nonlinear approximation performance and convergence speed compared with particle swarm optimization and ant lion optimizer.3)Aiming at the poor adaptive and slow convergence of the teaching-learning-based optimization,the improved method of the teaching and learning algorithm is studied,and the elitist,adaptive teaching factor,linear decreasing learning weight and self-driven strategy are combined with reservoir to further improve the prediction performance of the model.To verify the validity of this model and solve the problem of air quality prediction,the improved model is tested in chaotic time series and used to solve the practical problem of air quality prediction.Emulation results show that the improved methods have more significant advantages in prediction performance compared with traditional teaching-learning-based optimization,particle swarm optimization and ant lion optimizer methods.Further in the air quality prediction task,its effectiveness is verified.Figure 23;Table 19;Reference 45...