摘要
目的探究多变量LSTM模型,旨在精准预测大蒜种植面积,缓解大蒜价格波动、促进大蒜产业可持续发展.方法考虑大蒜价格、产量、自然灾害和出口量等多个影响因素,提出一种基于多变量长短期记忆模型(Long Short-Term Memory,LSTM)的大蒜种植面积预测方法.通过构建多变量LSTM模型,使用2003—2022年的大蒜价格、产量、自然灾害和出口量数据作为输入变量,以大蒜种植面积为输出变量进行训练和预测.结果多变量LSTM模型的平均绝对百分比误差(MA PE)为3.73%,相较于单变量LSTM模型、向量自回归模型(Vector Autoregression Model,VAR)和自回归积分滑动平均模型(Autoregressive Integrated Moving Average Model,ARIMA)具有更低的误差,显示出更高的预测准确性.结论模型能挖掘大蒜种植面积、大蒜价格、产量、受灾面积和出口量等变量间的相关性,能够较为准确地预测大蒜种植面积.
Objective In order to accurately predict the planting area of garlic in order to mitigate garlic price fluctuations and promote the sustainable development of the garlic industry.Methods This article proposes a garlic planting area prediction method based on a multivariate Long Short-Term Memory(LSTM)model,which considers multiple influencing factors such as garlic prices,production,natural disasters,and export volume.By constructing a multivariate LSTM model and utilizing data on garlic prices,production,natural disasters,and export volume from 2003to 2022 as input variables,with garlic planting area as the output variable for training and prediction.Results Experimental results demonstrate that the multivariate LSTM model achieves an average absolute percentage error(MAPE)of 3.73%,showing lower errors compared to single-variable LSTM models,Vector Autoregression models(VAR),and Autoregressive Integrated Moving Average Models(ARIMA),indicating higher prediction accuracy.Conclusion This article model is able to unveil the relationships between variables such as garlic cultivation area,garlic prices,yield,affected area,and export volume,enabling more accurate predictions of garlic cultivation area.
作者
胡彦军
HU Yanjun(School of Applied Engineering,Henan University of Science and Technology,Sanmenxia 472000,China)
出处
《河南科技学院学报(自然科学版)》
2026年第1期75-83,共9页
Journal of Henan Institute of Science and Technology(Natural Science Edition)
基金
教育部科技发展中心中国高校产学研创新基金项目(2021LDA10003)。
