摘要
现有蔬菜价格预测模型多针对单一品种且稳定性与适用性不足,鉴于此提出一种基于HP滤波法(Hodrick-Prescott filter)与经验模态分解法(Empirical mode decomposition,EMD)分解数据,并耦合卷积神经网络(Convolutional neural network,CNN)与长短期记忆模型(Long short-term memory,LSTM)的蔬菜价格预测模型。HP-EMD方法将价格序列分解为意义明确的分量以分析价格的波动规律,CNN-LSTM方法提取分量特征以提高模型的稳定性。以云南省2019—2021年西红柿、芹菜、菠菜、大白菜和大蒜的价格数据进行模型验证。结果表明:该模型预测的西红柿价格平均相对误差为5.03%、决定系数为0.85、均方根误差为0.30元(人民币,下同)∕kg,DM检验(Diebold mariano test)表明该模型显著优于其他模型。其他蔬菜预测结果的决定系数也均在0.8以上,表明该模型具有良好的适用性。
The existing vegetable price prediction models are mostly for single species and are not stable and applicable enough.We propose a method based on HP filtration(Hodric-Prescott filtration)and empirical mode decomposition(EMD)to decompose the data,and coupled with convolutional neural network(CNN)and long short-term memory.HP-EMD method decomposes the price series into meaningful components to analyze the price fluctuation pattern,and the CNN-LSTM method extracts the component features to improve the stability of the model.The model was validated with the price data of tomato,celery,spinach,cabbage and garlic in Yunnan Province from 2019—2021.The results showed that the model predicted tomato prices with an average relative error of 5.03%,coefficient of determination of 0.85,and root mean square error of 0.30 yuan∕kg,and the DM test(Diebold mariano test)indicated that the model significantly outperformed the other models.The coefficients of determination of the other vegetable forecasts were also above 0.8,indicating that the model had good applicability.
作者
何志亚
刘闯
武官府
刘云贵
马建强
HE Zhiya;LIU Chuang;WU Guanfu;LIU Yungui;MA Jianqiang(Honghe Hani and Yi Autonomous Prefecture Water Conservancy and Hydropower Survey and Design Institute,Mengzi 661199,China;College of Agricultural Science and Engineering,Hohai University,Nanjing 211100,China;Research Institute of Geological Survey,Consultation and Planning for Water Resources and Hydropower Projects of Honghe Hani and Yi Autonomous Prefecture,Mengzi 661199,China)
出处
《上海农业学报》
2024年第2期109-117,共9页
Acta Agriculturae Shanghai
基金
国家自然科学基金项目(51609082)。
关键词
蔬菜价格
CNN
LSTM
经验模态分解
HP滤波
Vegetable prices
CNN
LSTM
Empirical modal decomposition
Hodric-Prescott filtration
