LSTM模型在耕地面积预测领域的构建与应用  

Establishment and application of LSTM model for cultivated land area prediction

作　　者：向雁 侯艳林 姜文来[2] 陈印军[2] 成良强 XIANG Yan;HOU Yanlin;JIANG Wenlai;CHEN Yinjun;CHENG Liangqiang(Tourism Management School,Guizhou University of Commerce,Guiyang 550014,China;Institute of Agricultural Resources and Agricultural Regionalization,Chinese Academy of Agricultural Sciences,Beijing 100081,China;Oil Research Institute,Guizhou Academy of Agricultural Sciences,Guiyang 550009,China)

机构地区：[1]贵州商学院旅游管理学院,贵阳550014 [2]中国农业科学院农业资源与农业区划研究所,北京100081 [3]贵州省农业科学院油料研究所,贵阳550009

出　　处：《科技导报》2021年第9期100-108,共9页Science & Technology Review

基　　金：中国农业科学院科技创新工程协同创新任务项目(CAAS-ZDRW202012)。

摘　　要：长短期记忆(LSTM)模型广泛应用于系统故障、交通流量、股票指数、紧急事件、碳排放、石油产量、农区地下水位等多个领域,均表现了出色的预测性能。为了丰富耕地面积预测方法、提升耕地预测精度,将LSTM模型引入耕地面积预测。选择常用的趋势外推模型、指数平滑模型、灰色模型、移动平均自回归、支持向量机、NAR动态神经网络等6类模型进行对比,并以耕地变化趋势比较复杂的黑龙江省和变化趋势比较单一的辽宁省、吉林省作为案例进行分析,以验证LSTM模型耕地面积预测效果。结果表明,从均方根误差(RMSE)、平均绝对误差(MAPE)这2个指标的综合评价来看,LSTM模型拟合和预测效果均为最优。根据LSTM模型预测,2018—2030年黑龙江、吉林、辽宁3省的耕地面积将呈持续减少的趋势,耕地减少速度均有放缓之势。The long-short term memory model(LSTM)is a special recurrent neural network structure,which is widely used in system failure,traffic flow,stock index,emergency event,carbon emission,water table depth,and other fields,showing excellent prediction performance.This paper introduces the LSTM model into forecasting cultivated land area to enrich predicting methods and improve prediction accuracy.To verify the validity of the LSTM model in cultivated land area prediction,TE,GM,ES,ARIMA,SVM and NARNET models are selected for comparison,in which Heilongjiang,Jilin and Liaoning provinces are taken as case areas for revealing evaluation effects of different time series models.The results indicate that the prediction effect of LSTM is better than other models in terms of the comprehensive evaluation of RMSE and MAPE.Finally,according to LSTM forecast,the cultivated land areas of Heilongjiang,Jilin and Liaoning provinces will continue to decrease from 2018 to 2030 and the decrease rate will slow down.

关 键 词：耕地 预测模型 深度学习 神经网络 LSTM模型 

分 类 号：F323.211[经济管理—产业经济] TP183[自动化与计算机技术—控制理论与控制工程]