Research On Prediction Of Economic Based On Time Series And LSTM Model

The development of economic globalization has brought us new opportunities and challenges.While bringing our national enterprises to the world,it has also involved our economy in the tide of the world economy.The impact of economic globalization is becoming more and more obvious.Therefore,it is more and more important to forecast economic development trend,formulate appropriate economic policies and avoid risks in advance.On the issue of economic forecasting,this paper analyses the characteristics,classification and main forecasting methods of economic forecasting,including time series forecasting method,econometrics method,multivariate statistical analysis method,input-output analysis method,trend curve analysis method,neural network,etc.These methods mostly depend on mathematical models or the summary of practical experience of researchers.These models are simple and the accuracy is limited.With the development of artificial intelligence and big data technology,researchers pay attention to BPNN.They use BPNN to do economic forecasting,and has achieved good results.However,BPNN has a shortcoming that it does not consider the time-series characteristics.In conclusion,this paper chooses LSTM to do experiments.GDP and CPI are two kinds of important macro-economy indicators,and they are used as predictors in this paper.Firstly,it evaluates the importance of the influencing factors,select the indicators with high scores and establishes the evaluation index system of economic prediction.Secondly,different activation functions and different optimizers were used in LSTM model for comparison.Meanwhile,dropout technology is used to avoid over-fitting.Through these discussions,the LSTM model which is most suitable for economic forecasting is constructed.After that,this paper use improved bat algorithm to optimize LSTM model.And considering the timeliness of influencing factors,this paper establishes a model make up of GM(1,1)and improved LSTM model.GM(1,1)model is suitable for small sample prediction,so we use GM(1,1)model to get the forecast value of single feature,and then use improved LSTM model to predict GDP and CPI.Because GM(1,1)is not good at the long-term prediction,and it is easy to be influenced by random factors.To solve this problem,this paper do weighted averaging of prediction results of the unbiased grey model and the improved grey model based on sliding average method and traditional grey model.Then we pass weighted mean value of combination grey model to LSTM model.Finally,the paper designs and implements the display system of economic forecasting results.In this paper,prediction results of the several common models are compared.It is found that LSTM model has higher prediction accuracy and it is suitable for solving regional economic prediction problems.Three ways of improvement have effectively improved the effect of the model.Combining grey model with LSTM can better predict the future direction of economic development,so as to formulate appropriate economic policies and help the economy develop better and faster.