Study On Mining The Influencing Factors Of Hog Price And Predicting The Price Of Hog

Pork is an important food related to the national economy and the people’s livelihood,and the price of hogs has naturally also attracted widespread attention from the society,however,in recent years,the price of hogs fluctuated abnormally,often falling into a vicious circle of"expensive hogs hurt the ordinary people,and cheap hogs hurt farmers".The uncertainty of hog price is not only detrimental to the healthy development of the pig industry,but also to the healthy development of the national economy.Therefore,mining these significant factors affecting the price of hogs and making accurate predictions on the market price of hogs can not only reduce the economic losses of producers and operators,but also ensure the healthy development of pig industry and provide favorable protection for the interests of the general public and with realistic meaning.Although up to now,scholars at home and abroad have done more and more research in the field of mining and prediction of influencing factors of hog price,including neural network,time series,traditional machine learning,combinatorial model and other research methods,and have achieved some research results,however,through in-depth investigation of the current situation,it is not difficult to find that there are still some problems in the mining research of influencing factors of pig price,such as single mining method,and the excavated influencing factors do not have relevance,representativeness and independence(referring to the situation that the influencing factors are independent of each other and have no collinearity);in the research of hog price prediction,there are still some problems,such as the prediction accuracy is not high enough and it is easy to fall into local minima.1)This paper aims at the problems of unrepresentativeness,independence and relevance in the mining of influencing factors of hog price,this paper integrates a variety of feature engineering technologies and proposes a multi-model integrated voting mining algorithm.The experimental results show that the multi-model integrated voting algorithm has certain advantages over other mining algorithms,and the average price of slaughtered pigs,the average price of sows and the price of fattening pig mixture ingredients are mined from the weekly data of raw pig and feed prices.These three factors are the significant influencing factors of hog price.2)This paper aims at the problems in the research of hog price prediction,such as the prediction accuracy is not high enough、it is easy to fall into local minimum value、the lack of data on many influencing factors of hog price in real pig production,and even only a single hog price data without other influencing factors,focusing on the direction of time series prediction,combined with deep learning technology,this paper uses deep learning time series prediction methods to predict hog prices,and proposes a bidirectional long short-term memory with self attention mechanism(Bilstm+Self-Attention).The experimental results show that the prediction result of Bilstm+Self-Attention model is the best,with MSE(mean square error)of 0.186232,MAE(mean absolute error)of 0.218997 and R~2(coefficient of determination)of 0.997639;Comparing the prediction results of Bilstm+Self-Attention model with the prediction results of xgboost algorithm optimized by grid search parameters,the three regression prediction and evaluation indicators(MSE,MAE and R~2)have increased by 59.5%,30.6%and 0.35%respectively.This shows that Bilstm+Self-Attention algorithm can fully learn the nonlinear change law between hog prices,and can provide a potential and effective scheme for the construction of hog price monitoring system and early warning model.