Chaotic Time Series Prediction Based On Deep Learning

Chaotic time series has the external characteristics similar to random motion,and shows a disordered random state.However,its inherent nonlinear dynamic system makes it possible to predict chaotic time series within a certain period.At present,the chaotic time series prediction is mainly divided into single-step prediction and multi-step prediction.Establishing chaotic time series prediction methods with high precision can promote the development of chaotic time series prediction.Therefore,this paper focuses on the in-depth study of chaotic time series prediction.On the basis of studying the status for chaotic time series prediction at home and abroad and the existing prediction methods for chaotic time series,this paper proposed a single-step prediction method for chaotic time series with high precision,in order to solve the problem of large prediction error caused by considering prediction as a simple regression problem and using the single model in the existing single-step prediction methods for chaotic time series.This method takes the directional trend into account the single-step prediction for chaotic time series,improves the training objective function,uses two kinds of prediction models,starts from multi-angle prediction,and combines the prediction results through the optimization algorithm,so as to improve the precision of chaotic time series prediction and achieve the single-step prediction for chaotic time series with high precision.At the same time,considering the problem of data leakage in most multi-step prediction methods,a multi-step prediction method for chaotic time series with high precision was proposed.This method preprocesses chaotic time series by overlapping slices for several times,improves the noise reduction threshold function,and then reorganizes the processed sequence data,so as to reduce the noise interference to prediction,concurrently prevent the problem of data leakage,and achieve the multi-step prediction for chaotic time series with high precision.Finally,the comparative analyses were performed on the proposed methods with the widely used chaotic time series prediction methods.From the angles of validity and error,it is verified that the proposed single-step and multi-step prediction methods for chaotic time series with high precision meet the high-precision prediction requirements of chaotic time series and have certain validity and superiority.