Application Of ARIMA-BiLSTM Model In Vegetable Price Prediction Under Different Combination Modes

Agriculture is an important pillar of the national economy.The stability of the agricultural product market is of great significance to the stable development of the national economy.As a major category of agricultural products,the market price prediction of vegetable agricultural products is of great significance to vegetable growers,consumers and ensuring the balance of regional vegetable supply and demand.However,vegetables are fresh agricultural products,and their prices are vulnerable to external shocks from both supply and demand.Vegetable prices fluctuate frequently and have a great impact on Residents’ lives.Therefore,a method that can further improve the accurate prediction of vegetable prices is needed.Agriculture is an important pillar of the national economy,and the stability of the agricultural product market is a strong support for the stable and healthy operation of the national economy.As a major category of agricultural products,the market price prediction of vegetable agricultural products is of great significance for vegetable growers,consumers and ensuring the balance of regional vegetable supply and demand.In the actual production and circulation process,due to the seasonal changes of vegetables and the characteristics of fresh agricultural products,their prices are vulnerable to the external impact of both supply and demand,and the prices will change greatly in the same year.At the same time,due to the cobweb effect,the prices will also fluctuate significantly in different years.The operation of the vegetable market is uncertain.Vegetable prices fluctuate frequently,which makes it more difficult to predict vegetable prices.Sorting out the current research on vegetable price prediction,it is found that in the past,it is mostly single model prediction,and the combined model method is a research direction of sequence prediction in recent years,but the research on vegetable price prediction is insufficient.Combined with the seasonal and cobweb effect characteristics of the vegetable market,this paper further analyzes the mechanism behind ARIMA model and BiLSTM neural network,and constructs the vegetable price prediction method of ARIMA-BiLSTM combined model.Different from previous studies,the analysis of model fitting effect is explained from the perspective of historical price simulation to future price prediction.In addition,the future price prediction is divided into ultra short-term,short-term and long-term to demonstrate the application scenarios of the model,the following conclusions are drawn:(1)The two combined forecasting models can improve the deficiency of single model in vegetable price forecasting.The ARIMA-BiLSTM combined model with error correction is constructed.The predicted residuals of ARIMA model are fitted by BiLSTM,and the ARIMA prediction results are corrected.The fitting accuracy of historical price data of potatoes,cucumbers,celery,rape and cabbage reaches 92.05%,93.06%,86.38%,93.72%and 88.99%respectively.Compared with a single prediction model,the prediction accuracy of error correction combined model is improved,the highest percentage point was 11.03 percentage points.The ARIMA-BiLSTM combined model of component decomposition is constructed.Combined with the obvious seasonal characteristics of vegetable price fluctuations,the X-11 decomposition technology is used to decompose the original price into three components:trend,seasonal and irregular changes.The trend components are predicted by ARIMA model,and the seasonal and irregular changes are predicted by BiLSTM.Finally,the results are combined.The accuracy of the combined model in predicting the historical prices of potatoes,cucumbers,celery,rape and cabbage reached 93.27%,93.21%,88.80%,89.60%and 89.87%respectively,which was also significantly improved compared with ARIMA model and BiLSTM single prediction model.(2)In the future price prediction stage,the application scenarios of each model are viewed from the ultra short-term,short-term and long-term stages in the future.The conclusions are as follows:a.the longer the time span,the worse the prediction effect of the model dominated by the traditional time series model,and the neural network model will be better;b.In the short-term and long-term price prediction stage of vegetables,the ARIMA-BiLSTM combined model of component decomposition has significant advantages,and the prediction mean square error is only 0.031285 and 0.075309.(3)From the perspective of the essential mechanism and theoretical significance of the model,it is concluded that the performance of the ARIMA-BiLSTM combined model of component decomposition is better than the ARIMA-BiLSTM combined model with error correction,and it is more stable.At the same time,it has more practical significance in the comparison and analysis of the prediction results with a single prediction model.The above conclusions based on five vegetable objects well prove the universality and effectiveness of the ARIMA-BiLSTM combined model of component decomposition in the field of vegetable price prediction.At the end of the paper,some suggestions are put forward for the combined model of component decomposition to guide vegetable production and circulation,and some directions are pointed out for the later improvement of the model.