Application Of Hybrid Model Based On ARIMA And LSTM In Vegetable Price Prediction

Agriculture is the base of the national economy,and the agricultural product market is vital to Chinese socialist market economy.With the rapid development of the market economy,accurate control of agricultural products'price can guarantee the economic benefits of the majority of laborers and promote the development of the industry.From the national perspective,vegetables are important part of agriculture and an important economic crop.From the perspective of citizens,vegetables are rigid necessity of life,and frequent price fluctuations will impose great effects on farmers and consumers.The vegetable price time series is affected by many complex factors,and the influencing factors are coupled with each other,making the price data with non-linear characteristics such as time-varying and seasonal.Among the existing agricultural product price prediction methods,the prediction accuracy of the Autoregressive Integrated Moving Average Model(ARIMA)is not good,and this paper consider using the deep learning method to mine the nonlinear relationship between variables.Based on the idea of error compensation,the Long-Short Term Memory(LSTM)method is used to correct the prediction residual error,thereby establishing a hybrid model with linear and nonlinear methods.The predicted vegetable prices are finally obtained by combining the prediction of two methods.The operation steps of the hybrid model are as follows:This paper selects the time series data of seven vegetable prices in Dalian from 2013 to 2018.Firstly,the linear component is modeled and the error term is calculated using the ARIMA model;In the second step,the LSTM algorithm is used to carry out nonlinear error correction.For different vegetable varieties,this paper choose the hyper-parameters combination with the smallest error;finally,the linear prediction value calculated by ARIMA and the nonlinear error correction value obtained by the LSTM algorithm are summed to obtain the vegetable price prediction result of Dalian.The conclusion shows that the LSTM deep learning method can better reflect the nonlinear law in the vegetable price time series,and effectively correct the error of the linear prediction method.Compared with single ARIMA model,hybrid model improves the prediction accuracy and reflects superiority.The innovation of this paper mainly lies in the following three aspects:First,the application innovation of hybrid model.The combination method of ARIMA model and LSTM algorithm is applied to the prediction of vegetable prices,which not only improves the accuracy of prediction,but also satisfies the model's explanatory;the second is to reconstruct phase space during the training process of deep learning,selecting different numbers of input variables for training,and the time span is respectively week,month,quarter,half year,and one year;the third is that most scholars predict the price of a single agricultural product,while this paper selected seven kinds of vegetables.Different vegetable varieties also have different parameter configurations,which are more applicable.