Research And Application Of Deep Neural Network Prediction Fused With Variational Mode Decomposition

Variational mode decomposition is a completely non-recursive and adaptive signal processing method.It has the advantage in predetermining the number of mode decompositions.Based on the solid mathematical theory,it has more flexible application characteristics.For stock price data with nonlinear chaotic phenomena,its effective data processing methods and prediction models can provide researchers with new directions.In this paper,the classical chaotic Lorenz system is simulated by the deep neural network method fused with variational mode decomposition,and the daily closing price of the CSI300 index is used as the empirical analysis object.The specific research work is as follows:The first chapter is the introduction,which mainly introduces the research status,background and significance of stock data at home and abroad,as well as the main structure of the paper.The second chapter is the basic theory,which introduces the neural network algorithm steps,the process of obtaining the optimal solution of variational mode decomposition,chaotic theory,phase space reconstruction and other theories.The third chapter is the research on the regression model of support vector machine fused with varitional mode decomposition.After the initial values of x,y and z in the classical chaotic Lorenz equation are determined in the simulation research part,the Lorenz-y simulation time series is obtained by integrating the fourth-order Runge-Kutta algorithm.After decomposing the sequence into two modes using the VMD method,the modes are predicted by SVR model,which verifies that the VMD-SVR model can effectively predict the chaotic time series.In the empirical analysis,after the sequence is decomposed,the SVR model is used to predict the decomposed modes respectively,and the predictions of each mode are added to obtain the predictions of the fused model.The trend of the actual value and the predicted value is almost the same.The fourth chapter is the research of deep neural network fused with variational mode decomposition.In the simulation research part,the Lorenz-x simulation time series is selected.After the sequence is decomposed by the VMD method,the modes are predicted by DNN,and adds the predictions of each mode to obtain the predictions of the fused model.Both the test set and training set have set fit well,which verifies the effectiveness of the fused method for predicting chaotic time series.In the empirical analysis,the delay time for calculating the stock index sequence is 11,the embedding dimension is 4,and the space vector is obtained by reconstructing the phase space according to these two parameters.The Lyapunov index of the sequence is 0.0203,which shows that the sequence has chaotic characteristics.By training deep neural network with different structures,when the number of hidden layers is 2 and each layer has 9 nodes,the average absolute error value of the deep neural network is relatively the smallest.Finally the variational mode decomposition method is used to decompose the stock index sequence,and 5 modes with chaotic characteristics are obtained.The above deep neural network is used to predict each mode separately.Sum the obtained predictions to obtain the prediction of the deep neural network hybrid model fused with variational mode decomposition.The average absolute error of the hybrid model is 33.4333,which improves the prediction accuracy of single neural network and traditional linear model.This hybrid prediction model based on signal decomposition and phase space reconstruction technology has better effects and more ideal prediction in practice.