基于门控循环单元网络的农产品价格预测模型构建  

Construction of agricultural product price forecasting model based on gated recurrent unit

作　　者：王桂红[1] 潘栋 刘向锋 WANG Guihong;PAN Dong;LIU Xiangfeng(College of Information and Electrical Engineering,Shenyang Agricultural University,Shenyang 110866,China)

机构地区：[1]沈阳农业大学信息与电气工程学院,沈阳110866

出　　处：《沈阳师范大学学报（自然科学版）》2022年第5期451-456,共6页Journal of Shenyang Normal University:Natural Science Edition

基　　金：辽宁省社会科学规划基金资助项目(L12DJY055)。

摘　　要：农产品的价格预测对指导农业生产、调节市场消费供给有重要意义。首先提出规范产品名称、简化类别与统一单位等数据规范化原则,再采用滚动交叉验证的方法将规范化处理后的数据划分为训练集与验证集。针对基于回归分析的传统预测模型存在准确度和效率均偏低的问题,研究使用泛化能力较强的门控循环单元(gated recurrent unit,GRU)神经网络来构建价格预测模型,根据对比与消融实验调整模型的超参数与优化器,通过Dropout方法对模型进行优化。实验结果表明,最佳模型误差度为0.0435,耗时为10.8 min,在准确度和效率方面达到既定研究目标。该模型普适性较强,对于具有时序性的其他农产品数据也具有良好预测效果,可将模型增加查询接口并集成到相关WEB系统中。研究的实验方法与模型参数也可作为其他神经网络在价格预测方面研究的参考。The price forecast of agricultural products is of great significance in guiding agricultural production and regulating market consumption supply.In this study,the data normalization principles such as standardizing product names,simplifying categories and unifying units are first proposed,and then the normalized data is divided into training sets and test sets by rolling cross validation.In view of the low accuracy and efficiency of the traditional prediction model based on regression analysis,the GRU neural network with strong generalization ability was used to build the price prediction model.The model′s super parameters and optimizer were adjusted according to the comparison and ablation experiments,and the model was optimized through the Dropout method.The experimental results show that the error of the optimal model is 0.0435,and the time consumption is 10.8 minutes.The accuracy and efficiency of the model meet the established research goals.The model has strong universality and good prediction effect for other agricultural product data with time series.The model can be added with query interface and integrated into relevant WEB systems.The experimental methods and model parameters can be used as a reference for other neural networks in price forecasting.

关 键 词：数据规范化 门控网络 价格预测 模型优化 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]