基于集成学习的甘肃省粮食产量预测  

Grain Yield Prediction in Gansu Province Based on Integrated Learning

作　　者：涂丽珍 郭小燕[1] 冯浩 赵志刚 张中铭 TU Lizhen;GUO Xiaoyan;FENG Hao;et al

机构地区：[1]甘肃农业大学信息科学技术学院,兰州730000

出　　处：《智慧农业导刊》2025年第5期9-13,共5页JOURNAL OF SMART AGRICULTURE

基　　金：甘肃农业大学大学生创新创业训练项目(202416003,202416004)。

摘　　要：针对单个预测模型的侧重点不同导致粮食产量预测不够精准的问题,提出的方法是设计一个组合预测模型,选取粮食种植面积、农用机械总动力、农用化肥施用量、成灾面积、灌溉面积、最高温、最低温和日照时数这8个影响因素;选择并训练随机森林RF、梯度提升树GBDT和XGBoost 3种模型作为基模型,采用线性回归作为第二层模型集成输出最终的粮食产量预测的结果。该堆叠模型的决定系数为0.98,大于单个基模型的决定系数,同时均方根误差、平均绝对误差、均方误差也降低到最小,分别为6.32、4.32和40.00。结果表明,与单个模型相比,堆叠模型对于粮食产量预测具有更高的准确性和更强的鲁棒性。Aiming at the problem that the different focus points of individual prediction models lead to inaccurate grain yield prediction,the proposed method is to design a combined prediction model and select eight influencing factors:grain planting area,total power of agricultural machinery,agricultural chemical fertilizer application,disaster-stricken area,irrigated area,maximum temperature,minimum temperature and sunshine hours;select and train three models:RF(random forest),gradient lifting tree GBDT and XGBoost as the base models,and use linear regression as the second level model to integrate output the final grain yield prediction results.The coefficient of determination of this stacked model is 0.98,which is greater than the coefficient of determination of a single base model.At the same time,the root-mean-square error,the mean absolute error,and the mean square error are also reduced to a minimum of 6.32,4.32 and 40.00 respectively.The results show that compared with individual models,the stacked model has higher accuracy and stronger robustness for grain yield prediction.

关 键 词：甘肃省 区域粮食安全 粮食产量预测 集成学习 影响因素 

分 类 号：F323[经济管理—产业经济]