摘要
针对短时间序列农产品价格预测中存在的非平稳性、噪声干扰及多因素耦合问题,本研究提出了一种基于多特征融合的组合预测模型。本研究以重庆市大白菜价格为研究对象,整合Pearson相关性分析、经验模态分解(EMD)和Savitzky-Golay(SG)滤波器,构建了“Pearson-EMD-RF-SG”预测框架。首先,通过Pearson分析筛选出老南瓜、韭菜等10种竞争性农产品的关键影响因子;其次,利用EMD分解原始价格序列,提取高频本征模态函数(IMF)及残差趋势分量;最后,结合随机森林(RF)模型进行多尺度特征训练,并通过SG滤波器优化预测结果的平滑性。实验采用2022—2024年130期大白菜价格数据,对比支持向量回归(SVR)、极致梯度提升(XGBoost)和多层感知机(MLP)等模型,结果表明:完整模型的RMSE(0.126)、MAE(0.082)和R2(0.901)显著优于单一模型,消融实验进一步验证了各模块的贡献度,移除SG或EMD后预测误差分别增加21.5%和34.69%。研究表明,多特征融合策略能有效捕捉价格波动的非线性动态,为短时序农产品市场预测提供了高精度解决方案,对智慧农业的供应链优化与政策调控具有实践价值。
To address the challenges of non-stationarity,noise interference,and multi-factor coupling in short-term agricultural price prediction,this study proposed a hybrid forecasting model based on multi-feature fusion.Focusing on Chinese cabbage prices in Chongqing,the"Pearson-EMD-RF-SG"framework integrates Pearson correlation analysis,Empirical Mode Decomposition(EMD),and Savitzky-Golay(SG)filter.Key steps include:(1)screening influential factors from 10 competitive crops(e.g.,pumpkin,leek)via Pearson analysis;(2)decomposing price series into intrinsic mode functions(IMFs)and residual trends using EMD;(3)training multi-scale features with Random Forest(RF)and refining predictions via SG smoothing.Experimental results from 130-period data(2022-2024)demonstrate the model's superiority,with RMSE(0.126),MAE(0.082),and R2(0.901)outperforming benchmarks(SVR,XGBoost,MLP).Ablation studies reveal that removing SG or EMD modules increases prediction errors by 21.5%and 34.69%,respectively.The study proved that multi-feature fusion effectively captures nonlinear dynamics in price fluctuations,offering a high-precision solution for short-time series forecasting and supporting decision-making in smart agriculture supply chains.
作者
廖宇诚
王元玲
赵天凡
Liao Yucheng;Wang Yuanling;Zhao Tianfan(Chongqing College of Mobile Communication,Chongqing 401420;Chongqing Key Laboratory of Public Big Data Security Technology,Chongqing 401420)
出处
《农业展望》
2025年第7期3-10,共8页
Agricultural Outlook
基金
重庆移通学院校级应用研究项目(KY2024013)。
关键词
农产品价格预测
消融实验
多特征融合
随机森林
agricultural products price prediction
ablation experiments
multi-feature fusion
Random Forest
