Research On Combined Forecasting Model For Time Series Data

With the development of technology,time series forecasting approaches are constantly being improved.At present,the commonly used prediction approaches include traditional time series methods and machine learning methods.These methods are easy to operate and have good prediction performance,so they are widely used in some industries.However,there are differences in the forecasting performance of the same model in different data series,which cause the unstable prediction characteristics for these approaches.The combined forecasting method can make full use of the advantages of each model in an ensemble manner to improve the forecasting accuracy and versatility of the model.The Echo State Network(ESN)has good application performance in time series prediction.Therefore,this paper builds an ESN-based combination model.The proposed ensemble method is optimized from two aspects: Firstly,for the current parameter optimization algorithms of neural networks are mostly heuristics,this paper introduces bayesian inference to the regularization loss function of the echo state network ensemble model to optimize model parameters,which aims to improve the prediction accuracy of the model.Secondly,in view of the problem that the combination method of fixed weight cannot dynamically update the weight coefficients which causes low combination efficiency,this paper uses the generalized mixture function to dynamically update the weight of the combined model on the basis of the improved ESN combination model,so as to enhance the capabilities of the proposed model to dynamically capture data information.Compared with other variable weight calculation methods,the construction of generalized mixture functions is simpler and computationally cheaper.Finally,the validity of the combined model is verified through experimental simulation data and actual data.The results show that the proposed combined forecasting model BESN-GM has higher prediction accuracy and better robustness than other prediction models.