Research On Time Series Data Prediction Algorithm Based On Deep Learning

Time series data contains valuable information,which can reveal the development rules of different phenomena in many fields such as network,meteorology,electricity,transportation,etc.Based on LSTM neural network,this paper studies the problem of feature redundancy and combination model in time series prediction,solves the problem of limited prediction accuracy caused by poor input characteristics of the model,and realizes the optimization of prediction accuracy using combination model.First of all,this paper introduces the advantages and disadvantages of existing time series prediction technologies.Through the study of feature selection and combination models,the problem of feature redundancy and the problem of limited prediction accuracy of a single model in time series prediction are introduced.Then,the influence of different input parameters in LSTM neural network on the prediction performance of the model is studied to select the appropriate input parameters.On this basis,aiming at the problem of feature redundancy in the LSTM model,the advantages and disadvantages of different feature selection algorithms are compared,and the appropriate feature selection algorithm is selected and optimized to obtain the optimized feature selection algorithm.The algorithm is applied to LSTM model to verify the effectiveness of feature selection algorithm applied to LSTM model.Finally,aiming at the problem of limited prediction accuracy of a single model,this paper analyzes the relevant theories of combined model technology in time series prediction in detail,and proposes a combined model optimization algorithm,which combines the AR model with the optimized LSTM model to obtain a combined prediction model.The combination prediction model is applied to the actual prediction problem,which effectively improves the target prediction accuracy.Theoretical research and experimental results show that the time series prediction algorithm used in this paper,including the optimized feature selection algorithm and the combination model optimization algorithm,can be well combined with LSTM neural network,which not only overcomes the problems of poor input characteristics and limited prediction accuracy of the model,but also improves the stability and reliability of the prediction model,which is of great significance to the research of time series prediction.