基于改进LSTM模型的农产品短期价格预测方法  

A Short-term Price Forecast Method for Agricultural Products Based on Improved LSTM Model

作　　者：张保国 任万明 吴兵 ZHANG Baoguo;REN Wanming;WU Bing(Shandong MGdaas Systems Co.,Ltd.,Jinan,Shandong 250100,China;Shandong Modern Agricultural Rural Development Research Center,Jinan,Shandong 250100,China;Jinan Agricultural and Rural Information Center,Jinan,Shandong 250100,China)

机构地区：[1]山东麦港数据系统有限公司,山东济南250100 [2]山东省现代农业农村发展研究中心,山东济南250100 [3]济南市农业农村信息中心,山东济南250100

出　　处：《热带农业科学》2021年第4期131-136,共6页Chinese Journal of Tropical Agriculture

基　　金：国家重点研发计划项目(No.SQ2020YFF0401705);济南市农业科技创新计划项目(No.201906)。

摘　　要：农产品价格与人民的生活息息相关,既关乎消费者的切身利益,也是政府农产品生产调控决策的重要依据。由于自然灾害、重大疫情等各种因素影响,农产品短期价格具有波动大、非线性的特点。而剧烈起伏的波动往往对预测结果产生不好的影响,已有的方法在具有变化大且非线性特点的农产品短期价格预测上表现不理想。本文提出了一种改进的长短期记忆网络(LSTM)模型,该模型能够多维度分析农产品历史价格变化情况,获取价格周期性变化规律;并且在LSTM的基础上增加了前置门,将历史价格信息与价格波动信息相结合,减少数据异常值的波动对预测结果影响,有效地提高了农产品短期价格预测的准确性。实验数据表明,本文所提模型在预测结果的准确性等方面明显优于现有的其他对比模型。The price of agricultural products is closely related to people's lives,concerns the vital interests of consumers,and is also an important basis for government decision-making.Due to natural disasters and major epidemics,the short-term price of agricultural products is characteristics of great fluctuation and nonlinearity.However,dramatic fluctuations in price often have a negative impact on the forecast results,and the existing methods do not perform well in short-term price prediction with considerable changes and nonlinear characteristics.An improved long short-term memory(LSTM)model is introduced to analyze the changes in historical price to acquire the periodic changes in price.In this model a front gate was set on the basis of LSTM to include the price historical data and price fluctuation data to reduce the influence of fluctuations of abnormal price data on the predictions.The results showed that the improved model is obviously better than the other state-of-the-art models in prediction accuracy.

关 键 词：农产品 短期价格预测 前置门 长短期记忆网络 

分 类 号：F323.7[经济管理—产业经济]