基于STL-Informer-ARIMA组合模型的猪肉价格预测方法研究
Pork Price Prediction Method Based on STL-Informer-ARIMA Combined Model
王杰 1董国奥 1李俊清2
作者信息
- 1. 山东农业大学信息科学与工程学院,山东 泰安 271018
- 2. 山东农业大学信息科学与工程学院,山东 泰安 271018;山东农业大学农业大数据研究中心,山东 泰安 271018
- 折叠
摘要
合理预测猪肉价格对稳定生猪市场价格波动及促进猪产业的健康持续发展具有重要意义.本文深入研究了猪肉价格的影响因素,整合了29种相关价格数据.通过分析数据特征,针对Informer模型在猪肉价格数据提取方面的局限性,对Informer模型进行改进,将自注意力机制ProbAttention更换为Synthesizer模型,引入了价格波动模块.在此基础上,本文提出了一种新的价格预测组合模型STL-Informer-ARIMA,模型结合了随机森林(Random Forest)和递归特征消除(Recursive Feature Elimination)进行特征选择,利用季节性和趋势分解法(Seasonal and Trend Decomposition Using Loess)对猪肉(白条猪)价格进行分解,采用ARIMA模型对季节项进行预测,同时针对趋势项和残差项采用改进的Informer模型进行预测.实验表明,STL-Informer-ARIMA组合模型的MSE为0.532,MAE为0.446,RMSE为0.729,MAPE为0.030,R2为0.958,相较于LSTM、SVR和GRU等常用价格预测模型,本文的组合模型有效提升了猪肉价格预测的准确性和可靠性.
Abstract
Reasonable pork prices forecasting is significant for stabilizing the fluctuation of pork market prices and promoting the healthy and sustainable development of the pig industry.We conducted an in-depth study of the influencing factors of pork prices and integrated 29 related price data.Based on the analysis of the data characteristics and the limitation of the Informer model in pork price data extraction,it is improved by replacing the self-attention mechanism ProbAttention with the Synthesizer model's mechanism,and the price fluctuation module is introduced.On this basis,we propose a new combined price forecasting model STL-Informer-ARIMA,which combines Random Forest and Recursive Feature Elimination for feature selection.The model utilizes the Seasonal and Trend Decomposition Using Loess method to decompose the pork(lean hogs)prices,employs the ARIMA model to predict the seasonal component,and adopts the improved Informer model to predict the trend and residual components.Experimental results show that the MSE of the STL-Informer-ARIMA combined model is 0.532,MAE is 0.446,RMSE is 0.729,MAPE is 0.030,and R² is 0.958.Compared with commonly used price forecasting models such as LSTM,SVR,and GRU,the combination model has improved the accuracy and reliability of pork price prediction.
关键词
猪肉/价格/特征选择/改进的Informer/组合模型Key words
Pork/prices/feature selection/improved Informer/combinatorial modeling引用本文复制引用
基金项目
国家重点研发计划政府间/港澳台重点专项(2019YFE010380)
山东省科技型中小企业创新能力提升工程项目()
出版年
2024
国家科技期刊平台
NSTL
万方数据