Pork Price Forecast Based on the Comparison of KPCA-ARIMA-LSTM and DBN Multi-Model

The pork price is affected by feed prices, national policies and per capita GDP. The price fluctuates greatly, and there is no obvious regularity. Accurately predicting the price of pork is of great significance to stabilizing the agricultural product market. Most scholars take a single model to predict the price, but the accuracy is not high. This paper proposes a combination model of KPCA-ARIMA-LSTM to predict pork prices. Since many factors affect the price of pork, to simplify the calculation and improve the calculation efficiency, this paper firstly uses KPCA to reduce the dimensions of the influencing factors. Due to the high volatility of pork prices, this paper divides the historical pork prices into linear and non-linear parts, uses ARIMA to predict the linear part, and uses LSTM for non-linear features. The neural network makes predictions and combines the results of the two models. In addition, the DBN model is also constructed, and the ARIMA model, LSTM model, ARIMA-LSTM model, and DBN model prediction are compared comprehensively. The experimental results show that the KPCA-ARIMA-LSTM combination model has high prediction accuracy.