Jiangsu Province GDP Forecast Analysis Based On ARIMA Model And BP Neural Network Model

GDP refers to the total value of labor services and final products produced by all resident units in a country or region within a period of time.It represents the economic situation of a country or region.It is closely related to employment,economic growth rate,inflation,etc.GDP is narrow economic growth.Therefore,it is of great importance for us to grasp the characteristics of the short-term change of GDP and make short-term forecast of GDP.Jiangsu's GDP has been slightly lower than Guangdong's in recent years.For Jiangsu province,which is a big GDP province,it is very important to predict the GDP value.This paper mainly selects the GDP data of Jiangsu province from 1970 to 2018,in which the training model USES the data from 1970 to 2016,while the test model USES the data of the remaining years.The GDP data of Jiangsu province belong to the non-stationary time series and have the characteristics of linearity and nonlinearity.Based on these data,four models were established in this paper.The software R and SAS were mainly used to forecast Jiangsu's GDP.Finally,the prediction accuracy of our model is compared and analyzed to identify the optimal model.Firstly,ARIMA model was established to determine the relative optimal model ARIMA(0,2,2)according to the minimum information criterion and model parameter conditions.This model is a linear model of GDP data of Jiangsu province,and the GDP value of Jiangsu province was predicted.Secondly,the BP neural network of nonlinear model was established,and the appropriate activation function was selected to train the neural network structure,so as to dig out the nonlinear characteristics of GDP data.Then,the GDP data has the characteristics of linearity and nonlinearity,and the single model has some defects.Therefore,it is decided to build a combination model,which USES ARIMA model to predict the linear part of GDP,and BP neural network to predict the nonlinear residual of the model,and then adds the predicted values of the two models to get the final predicted values of the model.The empirical analysis shows that the predictive ability of the combined model is better than that of BP neural network.Finally,although the combined model is better than BP neural network in GDP prediction,it still has some defects,so it needs to be improved to some extent.Can be seen from the sequence diagram GDP data of Jiangsu province has the certain trend of exponential growth,so this article realization to data smoothly through the logarithmic difference and second order difference of two methods,then the ARIMA model is set up,and then get the weighted average of the two models of predicted values,thus improved the combination model of linear part of the predicted value,the residual part of the two models,weighted average,using BP neural network to predict nonlinear residual,add those two parts of the predicted values,the combination of the improved model predictive value.To sum up,in this paper,in order to predict GDP value in jiangsu province,chose the following four models:ARIMA model,BP neural network model,combination model and the combination of the improved model,and then compared the four kinds of prediction relative error of the model,it is concluded that the combination of the improved model of short-term prediction effect is better than the other three models,ARIMA model in a relatively long time prediction effect is better,higher prediction precision.Therefore,this paper models all the GDP data,USES the improved combined model to predict jiangsu's GDP in 2019,and USES the ARIMA model to predict jiangsu's GDP in 2020 and 2021.