甘蔗糖业2025，Vol.54Issue(4)：11-18,8.DOI:10.3969/j.issn.1005-9695.2025.04.002
基于随机森林的多蔗区甘蔗产量预测
Prediction of Sugarcane Yield in Multi Sugarcane-Growing Area based on Random Forest
摘要
Abstract
In order to improve the prediction accuracy of sugarcane yield at the county scale in the core sugarcane-producing region of Guangxi,the dynamic factor analysis and machine learning modeling were carried out by using the daily meteorological data and sugarcane yield per unit area data of 30 counties in Guangxi from 1993 to 2022.Based on the previous research and literature review,key meteorological factors(such as precipitation,average temperature,minimum temperature)were precisely identified and captured for segmented sugarcane growth phases(seedling,elongation and maturity stage),and a county-scale random forest algorithm was used to build the prediction model of sugarcane yield per unit in multi sugarcane-producing area.Model evaluation results demonstrated a mean absolute error(MAE)of 2.95 t/hm2 and a mean absolute percentage error(MAPE)of 3.83％on the test set.The model exhibited exceptional accuracy,particularly in the medium yield range of 45～75 t/hm2,where the scatter plot showed strong agreement between predicted and observed values,with a MAPE of 1.92％.These results indicate the model can effectively capture the sugarcane yield variation trends,and highlight the advantage of random forest in dealing with nonlinear relationships and complex interactions.关键词
甘蔗/随机森林回归/产量预测/气象因子Key words
Sugarcane/Random forest regression/Production forecast/Meteorological factors分类
农业科技引用本文复制引用
成振华,匡昭敏,陶伟,欧钊荣,马瑞升..基于随机森林的多蔗区甘蔗产量预测[J].甘蔗糖业,2025,54(4):11-18,8.基金项目
中国气象局科技项目"揭榜挂帅""基于天空地一体化融合数据的精细化甘蔗农情灾情评估预警和产量预报技术与数算一体化平台研发"(CMAJBGS202322) （CMAJBGS202322）
广西科技重大专项"高效低损甘蔗机械化收获关键技术集成应用与示范"(桂科AA22117007) （桂科AA22117007）
