Research On Agricultural Product Price Prediction Based On Deep Learning

Agriculture is the foundation industry of China’s national economy,and an important support for ensuring national food security,maintaining social stability,promoting rural revitalization,and realizing common prosperity.Agricultural product prices are important signals that reflect market supply and demand,affect production costs and benefits,guide production and consumption behavior.They are also important factors in safeguarding farmers’ interests,promoting agricultural efficiency and income growth,and achieving high-quality and efficient development.In recent years,due to various factors,China’s agricultural product prices have fluctuated abnormally several times.These fluctuations have seriously affected people’s normal lives and the country’s economic growth.Improving the accuracy and reliability of agricultural product price forecasting can help all parties make more reasonable decisions,thereby mitigating the harm caused by agricultural product price fluctuations.The research content of this article is as follows:(1)The data of agricultural product prices,related economic and agricultural production variables from 2012 to 2021 were collected and preprocessed.A Prophet model was built to predict the prices of agricultural products,and the fluctuation characteristics of agricultural product prices were analyzed by decomposing them into trend,holiday,weekly cycle,and yearly cycle components.After analyzing the fluctuation of agricultural product prices,the conclusion was drawn that in recent years,agricultural product prices have shown an upward trend but with a slow growth rate,and cyclic fluctuations affected by factors such as holidays and growth cycles.(2)The PSO-Prophet model was constructed by optimizing the parameters of the Prophet model using the particle swarm optimization algorithm(PSO).The PSO-Prophet model improved the accuracy of agricultural product price prediction,with the average absolute percentage error(MAPE),root mean square error(RMSE),and average absolute error(MAE)of garlic price prediction reduced by 79.65%,73.92%,and 79.39% compared to the Prophet model,respectively.The predicted results of bean prices were reduced by 34.89%,34.72%,and 32.00%,while the predicted results of carrot prices were reduced by 38.35%,49.67%,and 57.47%.The predicted results of luffa prices were reduced by 61.45%,56.99%,and 61.12%.(3)Granger causality test was used to analyze the relationship between five economic and agricultural production variables and agricultural product prices.The variables most relevant to the prices of four agricultural products were selected to construct a deep learning prediction model that incorporates influencing factors.RNN,LSTM,and GRU neural network models were used to predict agricultural product prices,and the optimal parameters were found using the grid search method.To further improve the prediction performance of the model,the selected economic and agricultural production variable data were used as model inputs.The experimental results show that adding influencing factors can improve the prediction accuracy of the deep learning model,and the LSTM model with influencing factors added has the highest prediction accuracy.To address the issue of poor performance of the PSO-Prophet model in predicting fluctuations but good performance in predicting trends,as well as the problem of deviation in the prediction values of the LSTM model with influencing factors added in predicting fluctuations,a weighted combination method was used to combine the two models and construct the PSO-Prophet-LSTM prediction model.The PSO-Prophet-LSTM model has better prediction performance and good generalization ability.Compared with the LSTM model with influencing factors added,the MAPE,RMSE,and MAE for the four agricultural product prices are reduced by 11.74%,4.08%,9.90%,0.83%,0.79%,3.09%,3.24%,-11.00%,0.69%,2.15%,-8.00%,and 2.43%,respectively.Compared with the PSO-Prophet model,they are reduced by 31.39%,27.68%,31.37%,75.86%,76.63%,76.40%,86.15%,81.77%,83.83%,46.10%,48.66%,and 46.44%,respectively.The combination model can effectively improve the shortcomings of single models in predicting agricultural product prices and accurately predict the trend and fluctuation changes in agricultural product prices.