Research On Pork Price Prediction Model Based On Deep Learning

According to the latest statistical data,over 70% of the total meat consumption of residents is pork consumption.In recent years,due to the introduction of policies and the outbreak of epidemics,pork prices have frequently fluctuated greatly,affecting the planning of pig farmers’ industry and residents’ meat consumption.Therefore,establishing a scientific and effective prediction model to provide a digital basis for the government and other relevant departments to make political decisions can avoid deviation of pig farmers’ breeding plans from market demand and promote the stable development of the pork industry.In recent years,with the popularization and widespread application of deep learning,people’s profits in production have greatly improved,and life has become more convenient.The long-term and short-term memory network model is the most typical and popular model in deep learning.The improved two-way long-term and short-term memory network model based on the long-term and short-term memory network model can learn the temporal correlation between data in a two-way way,and strengthen the extraction of time series features.However,there are still three shortcomings:first,it is difficult to determine the number of neurons in the hidden layer,which affects the fitting effect of the model;Second,it is difficult to determine the learning rate of the model,which affects the training effect of the model;Third,the batch size is difficult to determine,which affects the training performance of the model.The pork price is a time series with nonlinear characteristics,and the forecasting model is difficult to capture its dynamic change law.In this paper,considering the two factors that affect the prediction results,the parameters of the bidirectional long-term and shortterm memory network model and the nonlinear characteristics of the time series,a pork price prediction method based on the bidirectional long-term and short-term memory network model based on variational mode decomposition and Bayesian optimization is proposed.The main research contents of this topic are as follows:(1)Aiming at the problem that manually selecting parameters of a bidirectional long-term and short-term memory network model affects the prediction effect of the model,a Bayesian optimization method for bidirectional long-term and long-term memory network models is proposed.Bayesian optimization algorithms can obtain ideal solutions after only a few objective function evaluations and can continuously update hyperparameters.They are suitable for solving problems with high computational costs,unknown derivatives,and the need to solve global minimum values.The Bayesian optimization algorithm is introduced to optimize the number of hidden layers,the number of hidden layer neurons,the learning rate,and the batch size of the bidirectional long-term and short-term memory network model,and a prediction model is established based on the optimization results.The experimental results show that the optimization model is effective,and its MAE,RMSE,and MAPE values have decreased.(2)Due to the fluctuation and nonlinearity of pork price time series,a chimpanzee optimized variational modal decomposition algorithm was selected to decompose the data.The chimpanzee optimized variational modal decomposition algorithm can decompose the data into subcomponents with relatively simple fluctuations,which can fully extract the detailed features of the pork price time series.Subsequently,a bidirectional long-term and short-term memory network model was used to predict pork prices.The results show that the proposed scheme improves the accuracy of the prediction model and has good stability.(3)Based on the above two improved methods,this paper proposes a CVMD-BO-BiLSTM combined model prediction method.In order to verify the prediction performance of the combined model proposed in this article,a comparative experiment was conducted using a commonly used single prediction model and a combined model.The experimental results show that the CVMD-BOBiLSTM method has higher accuracy and applicability compared to other models,and is suitable for predicting pork prices.