Research And Application Of Hog Price Forecasting Method Based On Time Series Data

The pig industry plays an important role in China’s agricultural economy.However,the pig market price has been shrouded in the shadow of frequent fluctuations,bringing negative impacts on residents’consumption and industrial development.Reasonable price prediction can not only help farmers grasp the market situation and reduce potential loss risks,but also help government departments formulate regulatory measures,which is of great significance to ensure the stability of production capacity and prices,and promote the healthy development of the industry.However,the complex sequence characteristics of pig prices pose great challenges to forecasting research.Considering the accuracy of pig price prediction and the information needs of various decision-making scenarios in the process of breeding investment,the transmission characteristics of price fluctuations along the vertical production chain and the horizontal circulation chain are taken as the entry point and research are conducted on short-term and long-term forecasting methods for pig prices based on time series data.The main work of the paper is as follows:(1)In terms of short-term prediction,a correlation STGCN model based on graph neural networks is proposed in view of exploratory analysis of the spatial transmission effect of hog prices.The model consists of a data-driven graph generation module,a spatiotemporal convolutional network,and a linear autoregressive module.The graph generation module measures the correlation of price fluctuation characteristics and constructs a graph adjacency matrix to describe inter-provincial price dependence.The spatio-temporal convolutional network uses spectral convolution of redefined Laplacian matrices to fit a directed band weight graph,and combines gated convolutional networks to extract spatio-temporal features;Linear modules capture trend changes to alleviate the scale insensitivity of nonlinear models.The experimental results show that the proposed method has achieved good results on existing data.(2)In terms of long-term prediction,an HP-Light GBM-segment attention hybrid model based on a decomposition-ensemble framework is proposed.Box-Cox transform are used to adjust the distribution of price data in different periods and unilateral HP filtering are chosen for sequence decomposition to avoid information leakage.A dynamic regression model based on LightGBM and a encoder-decoder model with segment-wise attention are respectively designed for the decomposed trend series and pseudo periodic fluctuation series.The experiment is conducted on the national weekly data set of average prices of live pigs and pork,and the results show that the proposed method is superior to other benchmark models.(3)Finally,this paper designs and implements a data visualization system.The system integrates the research results on the long-term and short-term prediction issues mentioned above,achieving a visual display of pig industry chain data and prediction results,and providing services to farmers through WeChat applets.