Research On Time Series Prediction And Its Application Based On Deep Belief Network

With the rapid development of Internet of Things and Big data processing technology,the acquisition of time series data becomes more conven ient and faster,and the time series data are gradually showing the characteristics of nonlinearity,large capacity,and high complexity at the same time.Hence,the traditional methods of time series prediction have been gradually unable to satisfy the requirements of the current era.In order to capture the representative information from a large amount of time series data,it is becoming increasingly important to research new prediction techniques for complex time series.Deep Belief Network(DBN),as one of the main algorithms of Deep Learning,with its powerful feature extraction and function representation capability and the advantage in dealing with complex nonlinear data and so on,has been widely utilized in image classification,speech recognition and fault diagnosis.But in the field of time series,especially the prediction research and application are far from enough.This paper aims to solve the existing problems in the field of time series prediction.For example,it is difficult to establish accurate physical model for complex series and difficult to characterize the relationship in time series.Therefore,the data-driven prediction technology based on DBN is studied in depth.Firstly,based on the analysis of DBN,a time series prediction mode l is proposed,and the framework and process of this model are described in detail.At the same time,in order to explore the impact of DBN network parameters for prediction results,the three aspects of input node number,network layer number and hidden layer node number are studied respectively.And the performance of DBN is compared with traditional time prediction algorithms using same standard data sets.Secondly,in order to further analyze the prediction performance of the basic DBN algorithm in practical applications,the Prognostics and Health Management(PHM)2012 challenge(Bearing health prediction)is taken as the background,and further research is conducted from three aspects: one-step prediction,multi-step prediction and remaining life prediction.Finally,duo to basic DBN algorithm has poor accuracy in the long-term prediction and lacks uncertainty expression of predicted results,a fusion prediction method based on DBN and Relevance Vector Machine(RVM)is proposed,in which the RVM is as t he prediction layer of basic DBN algorithm.And the fusion algorithm is applied to the remaining life prediction of the lithium-ion battery in University of Maryland,and the performance of the fusion algorithm is also verified and analyzed.The research results show that: compared with traditional time series prediction methods,the basic DBN is more suitable for forecasting high dimensional and high complexity data,and it has excellent short-term prediction performance.However,the long-term prediction ability of the basic DBN prediction algorithm is still lacking,and the uncertainty of the prediction results also could not be given at the same time.The fusion algorithm proposed in this paper based on DBN and RVM,which can not only improve the long-term prediction accuracy,but also solve the problem that the basic DBN algorithm does not have the expression of uncertainty.Because of these advantages,the fusion algorithm becomes more scientific in practical applications.