Prediction Of Chaotic Time Series Based On Machine Learning With Composite Kernel

With the chaos theory and the application of chaos technology increasing, the chaotic time series analysis and prediction has become a hotspot in the research field of chaos signal. The application of chaos technology development involves many fields, including:acoustic, optical, chemical, hydrology, chaotic characteristics of earthquake, the" Butterfly Effect" of weather forecast, and the stock market chaos characteristic.At present, the prediction of the chaotic time series methods include:global prediction method, local prediction method, based on the maximum Lyapunov exponent prediction method, a nonlinear adaptive filter prediction method and based on kernel machine learning prediction method. In the kernel machine learning algorithm, The support vector machine (Support Vector Machine, SVM) and the relevance vector machine (Relevance Vector Machine, RVM) have paid close attention to by domestic and foreign scholars,because of the good robustness, strong generalization ability, fast convergence speed, high accuracy.In SVM and RVM, the single kernel function because of its some properties, resulting in they may have good generalization ability and learning ability,affecting the forecasting accuracy. This paper uses global and local kernels linear combination to construct combined kernel function, which combines the advantages of the two, and has good generalization ability and learning ability. The combined kernel function is used in SVM and RVM on the typical chaotic time series prediction algorithm. The simulation results show that the predicted ability of combined kernel function is better than the single kernel function.The combined kernel is applied to the prediction of actual chaotic sequences. Achieving the effective prediction of the sunspot series and power load. The predicted results show: The combination of kernel functions can accurately predict the chaotic sequence in the actual project, to better serve the engineering practice.