Research On Pattern Matching Of Financial Time Series Based On RSPP-CNN

In order to reduce the risk of investment and gain excess income,it is very important for investors to master the fluctuation and change law of the financial market.Chart pattern analysis is one of the analysis methods of financial market.Its effectiveness is widely verified.However,it is difficult for investors who are not professional to identify chart pattern successfully.Therefore,how to identify the charts pattern of financial time series automatically and accurately is a topic which is worth studying.The traditional financial time series pattern recognition methods have some shortcomings such as high requirements for users and large impact on the recognition results by data preprocessing.However,convolution neural network(CNN)has been proved to have good ability of image processing,graphics can be self-learning features and identification.In order to solve the problems of traditional recognition methods,this paper introduces a deep residual learning(CNN)and space Pyramid pooling layer(SPP)based on the convolution neural network(CNN),and proposes a RSPP-CNN depth neural network structure,which is more suitable for identifying the chart form of the financial time series.The main work of this article is as follows:1)Sort out the charts of common financial time series,describe their morphological characteristics and causes,and give mathematical definition and representation method;2)Analyze the shortcomings of the traditional CNN network structure to identify the chart form of the financial time series,and design the RSPP-CNN deep neural network structure: using the residual learning module and the spatial Pyramid pool layer(SPP)to improve the traditional CNN network structure to solve the gradient disappearance(explosion)and network degradation and the problem of inability to accept multiscale input to some extent;3)Building a financial time series chart recognition model based on RSPP-CNN.The model is divided into 3 parts: first,the financial time sequence diagram form database is set up,and the data acquisition method is given.Then a modified wavelet threshold denoising method is used to preprocess the financial time sequence diagram form data.Finally,the pre processed data are input into the trained RSPP-CNN network to complete recognition;4)Set up a set of controlled experiments to verify the model,and verify the superiority of the model to the recognition of financial time series charts.The structure of RSPP-CNN deep neural network proposed in this paper solves the problem of gradient disappearance(explosion),network degradation and inability to accept multiscale input in traditional CNN,and the RSPP-CNN based pattern recognition model of time series chart solves the shortcomings of the traditional pattern recognition method,which has high requirements for users and large impact on the recognition results by data preprocessing.The experimental results show that the model has a good recognition effect.In the future,the research can be expanded to design automatic trading system.