Application Of Variable Structure Temporal Neural Network Model In Stock Forecasting

In the era of big data,it has promoted the progress and innovation of people's way of thinking and methodology,and helped people understand the world from a comprehensive and diversified perspective.Data analysis technology based on data mining can mine the value behind the data and be applied in many fields of society.The stock market is regarded as the barometer of the national economy,which plays an important role in promoting economic development and consolidating social harmony.The correct prediction of the stock future trend has always been a hot research topic for scholars at home and abroad.The time series of stock price is affected by complex factors,which is non-linear and non-stationary.Therefore,it is difficult to predict the exact change of stock price.Wavelet analysis method is used to process the time series before prediction.The time series is divided into low frequency and high frequency sequences,and the characteristics of time series can be analyzed from the details.Temporal data mining can make full use of the time attributes of stock data,which is conducive to the discovery of internal knowledge between data.In addition,neural network model has strong nonlinear processing ability and learning ability,which is an effective method for stock data prediction.The selection of period is an important basis for stock market prediction.Compared with the original data,the frequency series generated by wavelet transform have some changes.It is no longer applicable to continue to use fixed empirical analysis period and single neural network structure for prediction.In this paper,the temporal data model and neural network model are combined,and the Mallat algorithm based on multi-resolution analysis is used to preprocess the stock data,then the time series data are divided into high frequency and low frequency sequences.According to the data characteristics of different sequences,sequences are transformed into temporal data set.Particle swarm optimization algorithm is used to find the neural network structure in different sequences.Different scorrespond to different neural network structures and then the variable structure temporal neuralnetwork prediction model is established.The experimental results show that the design of variable structure temporal neural network model is feasible,reasonable and effective.Compared with the unimproved neural network and SVM method,the model in this paper has better stability and lower prediction error in stock prediction.