Application Of Sparse Modeling Methods In Time Series Prediction

In recent years,the application of machine learning algorithms in time ser ies prediction has been widely concerned by scholars both at home and abroad.Sparse Representation(SR)is a typical sparse modeling machine learning method.As a memory-based modeling method which is different from the previous method s,SR has become a hot research topic with important theory and application value.Especially with the increasing maturity of wind power generation technology,the scale of wind farm is increasing,the influence of wind power generation on power grid is more and more signific ant,and the prediction of wind power time series is of great significance to the development of power system.The current modeling methods such as neural networks and support vector machines have are difficult to be defined and their generalization capacity is limited.Therefore,it is important to explore the application of sparse modeling in wind power time series prediction.The dissertation mainly investigates the application of sparse modeling methods in time series prediction to meet the requirements of practical application to predict the accuracy,and provides a new idea for the application of memory-based machine learning method in time series prediction.The main contents of the dissertation include the following aspects:(1)The basic theory of SR method is analyzed,involving two kinds of sparse vector solving theories based on greedy algorithm and relaxation algorithm respectively,in the meanwhile the theory about construction and learning algorithm of dictionary is analyzed.(2)Several kinds of sparse representation methods are introduced into chaotic time series prediction model,which are applied to the decomposition and reconstruction of the input data of the time series respectively in the product form of the super-complete dictionary and the sparse vector to extract the implicit information.The sparse vector and output data obtained from the solution are substituted into SVM prediction model.The SR-SVM combined forecasting model is established and compared with the single SVM method in t he reference chaotic time series to verify the feasibility of the method.(3)A kind of sparse coding prediction model based on adaptive data dictionary is proposed.The input and output data of historical time series are constructed to the input and output dictionaries respectively,which are composed of dictionary pairs.Then,for the delay input data vector,the dissertation uses the sparse coding methods to establish a prediction model with a dictionary pair,to predict the value by the inner product of the output dictionary and the sparse vector.At the same time,the dictionary of adaptive update strategy is considered to achieve on-line prediction with higher accuracy.The method is applied to chaotic time series prediction as well as the direct and indirect prediction of short-term wind power in different regions,meanwhile the validity of the method is verified by comparison with the existing methods under the same conditions.