基于随机森林的粮食单产预测研究

Prediction of Grain Yield Per Unit Area Based on Random Forest

科学预测一定区域的粮食单产水平及其影响因素,对提高粮食产能,保障国家粮食安全,更好地实施耕地保护具有重要理论和实践指导意义。本文以甘肃省及其14个市州为例,从压力-状态-响应-可持续的角度出发,构建粮食单产预测指标体系,采用随机森林(RF)模型对粮食单产进行预测,并对影响粮食单产的因素重要性进行评估。结果表明,(1)RF模型相较于线性回归、神经网络模型预测出的粮食单产更为切合实际,为进行粮食单产的预测提出了新的方法;(2)对粮食单产影响较大的因素重要性排序,前5个因素为:地均化肥投入>灌溉指数>农民人均农业产值>地均农业机械投入水平>单位面积农业产值,为更好地保护耕地、管控粮食生产能力指明了方向。

Scientifically predicting the grain yield level and its influencing factors in a certain area has important theoretical and practical guiding significance for improving grain productivity,ensuring national food security and better implementing cultivated land protection.In this paper,taking Gansu Province and its 14 cities and prefectures as an example,from the perspective of pressure-state-response-sustainability,the index system of grain yield prediction was constructed,and the random forest(RF)model was used to predict grain yield,and the importance of factors affecting grain yield was evaluated.The results show that,(1)compared with linear regression and neural network model,RF model is more practical in predicting grain yield,which provides a new method for predicting grain yield;(2)the first five factors that have great influence on the grain yield are:average fertilizer input>irrigation index>average agricultural output value of farmers>average agricultural machinery input level>agricultural output value per unit area,which points out the direction for better protection of cultivated land and control of grain production capacity.