Forecast Of Agricultural Production Value In Zhejiang Province Based On Time Series

Agricultural output value is an important indicator of Zhejiang’s economic source and an important source for measuring farmers’ income in Zhejiang Province.Therefore,it is effective to predict the agricultural production value of Zhejiang Province,study the change law of agricultural production value and accurately optimize its growth trend.The forecast is of great significance for ensuring the stability of Zhejiang’s economic development and assisting the Zhejiang Provincial Department of Agriculture in making scientific and effective decisions.This paper uses data from Zhejiang Province from 1978 to 2016 as a data set,The data from 1978 to 2013 is the training set,and the 2014 to 2016 is the verification set.Firstly,using the ARIMAX model based on statistical methods,the data is tested for stationarity and processing.The model is fixed by autocorrelation and partial correlation function graphs.By comparing the AIC and BIC information criteria of each model,the optimal one is selected.The relevant data models were compared and analyzed to find the minimum AIC and BIC values,and the agricultural production value prediction was carried out on the numerical values to obtain the first set of model data.The model has an RMSE value of 228.7452 and a MAE value of 224.0467.Then use the LSTM model to predict the data.Firstly,the main component is extracted from the data,the appropriate principal component is selected,the prediction of the long-term memory cycle neural network is performed,and the neural network model is parameter-tuned.The obtained model has an RMSE value of 0.0062 and a MAE value of 0.0049,which is more accurate than the ARIMAX model.Finally,the two models are integrated,and the ARIMA model is used to extract the characteristics of the data after principal component analysis,so that the circulating neural network can predict the total agricultural production value of Zhejiang Province 2017 to 2019.The obtained ARIMA-LSTM model has an RMSE of 0.0071 and a MAE value of 0.0056 compared with the ARIMAX model.Although the accuracy of the training model is slightly lower than that of the training model,the ARIMA is used for feature extraction,which makes the model more scalable.