Research And Implementation Of Time Series Prediction System Based On Echo State Neural Network

Forecasting time series under different backgrounds can predict the development trend of the industry and better guide people's social life.Time series prediction methods based on Echo State Neural Network(ESN)have gradually become one of the hotspots of research.When traditional ESN forecasts financial time series or power system time series,it is difficult to achieve the best forecasting effect,and for time series with high complexity,ESN cannot well explore its laws.This paper proposes a time series prediction method based on Cauchy mutation and leadership mechanism of particle swarm optimization(ACSIL-PSO)algorithm to optimize ESN.This method is difficult to ensure the adaptability and prediction accuracy of ESN when the wiener-hopf method is used to determine the output connection weight in the training process of ESN.In the ESN training process,the ACSIL-PSO algorithm is used to optimize the output connection rights of the ESN to ensure the adaptability of the ESN.And the ACSIL-PSO algorithm introduces the ideas of leaders and competitors,uses the leader particle to better lead other particles to search,and updates the global optimal position through the IGCM mutation strategy to reduce the possibility of particles falling into the local optimal.In the process of particle learning and communication,the method of sorting and communication is used to strengthen the learning between particles and improve the speed of particle convergence and prediction accuracy.Further analysis of the forecast data,it is found that the time series with high complexity,using traditional forecasting methods,often can not get better forecast results.Therefore,a complete integrated empirical mode decomposition(CEEMDAN)with adaptive noise is proposed.The particle swarm optimization algorithm based on Cauchy mutation(DWPSO)optimizes the combined forecasting method of ESN.First,use CEEMDAN to decompose the time series into several sub-sequences,use PE to detect the complexity of the sub-sequences,merge the sub-sequences close to PE,and then use the DWPSO algorithm to optimize the prediction method of ESN,predict the merged sequence,and finally divide the sequences The predicted value of is superimposed to get the actual predicted value.This method reduces the complexity of the time series,reduces the number of predictions and the amount of calculation,thereby ensuring the adaptability of the ESN.The DWPSO algorithm uses adaptive Cauchy mutation for the optimal position of the group to increase the possibility of the group escaping from the local optimum.At the same time,it uses an adaptive inertial weight strategy during the particle search process to accelerate the convergence of the particle swarm.The two algorithms are compared with other algorithms through experiments,and the prediction accuracy has been improved to a certain extent.Finally,a time series forecasting system is designed and implemented,which realizes the functions of data processing,extraction and forecasting,and at the same time the forecasting results are visualized.