Research On Time Series Prediction Based On Differential Evolution Algorithm And Echo State Network

Time series widely exist in scientific research and social life,so it is of great research significance to effectively and accurately predict them.In recent years,the echo state network has been favored by many scholars because of its simple training algorithm and strong reserve pool dynamics,and it has gradually become the first choice for time series prediction modeling.Aiming at the problem of adaptability of the echo state network reserve pool,and the reserve pool contains many nodes and the parameters are difficult to adjust,this paper uses a differential evolution algorithm to select the suitable parameter settings for the current time series for the reserve pool.In order to further improve the performance of the algorithm,the differential evolution algorithm is improved in two aspects.On the one hand,a mutation strategy pool is constructed to select a suitable mutation strategy for each individual in the population with a probability,increasing the probability of new individuals entering the offspring.On the other hand,the two control parameters of the differential evolution algorithm,scaling factor and crossover probability,are adaptively changed according to the fitness of the individuals,so that the population iteration is developing in a better direction.In order to verify the effectiveness of the proposed model,different time series are used for simulation experiments.The experimental results show that the proposed model has higher prediction accuracy,better adaptability and convergence.Since the single-target differential evolution algorithm cannot guarantee the stability of prediction,this paper further adopts the multi-target differential evolution algorithm to optimize the reserve pool parameters to solve the network adaptability problem.Choose two objective functions to ensure the readiness and stability of the prediction.In the multi-objective differential evolution algorithm,in order to balance the convergence and diversity of solutions,a good pareto front approximation is obtained,and a global optimal solution is obtained.In this paper,?-supported populations are used for decomposition,and the division of two subsets in each subpopulation represents the convergence and diversity of the solution,ensuring that the population iteration process is constantly approaching the real frontier.At the same time,the scaling factor and crossover probability are adaptively changed according to the sum of individuals,and a suitable scaling factor and crossover probability are found for each individual to generate a new individual.It can be seen from the simulation experiment results that the proposed model has higher prediction accuracy and better stability.