Computational Research and Implementation of Prediction of Pork Price Based on Deeplearning

Abstract

Abstract Social network public opinion has a wide range of social influence, which is one of the main external factors causing the fluctuation of pork market. However, previous studies have considered less factors outside the market, so the prediction accuracy is not high. A model based on LSTM about public opinion is proposed in this paper. Through multi-modal information fusion, the model makes use of the complementary advantages of multi-source information to enhance the estimated performance of the model. In addition, considering the influence of different fusion methods on the estimated performance of the model, six fusion methods are tried: Add, Sub, Max, Min, Con and AVG. For the sake of verify the effect of the model, the multimodal combination estimated model and the traditional regression model are used for comparative prediction. The results show that: in the task of pork price forecasting, the indicators of Multi_BERT_LSTM (CON) combined with Con mode are better than the contrast model, and the average absolute percentage error MAPE is increased by 17% compared with the optimal traditional forecasting model, and better prediction effect is achieved.