基于GA-RF模型土壤坚实度对黑土区大豆产量的影响  

Effects of soil penetration resistance on soybean yield in black soil region based on GA-RF model

作　　者：周修理[1] 张萍萍 秦娜 霍东旭 乔金友[2,3] ZHOU Xiuli;ZHANG Pingping;QIN Na;HUO Dongxu;QIAO Jinyou(School of Electrical and Information,Northeast Agricultural University,Harbin 150030,China;School of Engineering,Northeast Agricultural University,Harbin 150030,China;Heilongjiang Province Major Crops Production Mechanization and Materialization of Technology Innovation Center,Harbin 150030,China)

机构地区：[1]东北农业大学电气与信息学院,哈尔滨150030 [2]东北农业大学工程学院,哈尔滨150030 [3]黑龙江省主要农作物生产机械化材料化技术创新中心,哈尔滨150030

出　　处：《东北农业大学学报》2022年第10期67-75,共9页Journal of Northeast Agricultural University

基　　金：国家重点研发计划项目(2021YFD2000405-2);大豆产业技术体系岗位科学家任务(CARS-04-PS24)。

摘　　要：为探究机械压实土壤坚实度与大豆产量之间关系,以黑土区机械压实试验为基础,基于遗传优化算法(Genetic algorithm,GA)原理对随机森林(Random forest,RF)进行改进,克服传统RF模型参数选择主观性、泛化性差问题,建立基于改进随机森林(GA-RF)土壤坚实度对大豆产量影响预测模型,实现机械压实风险有效评估。结果表明,各深度土壤坚实度与大豆产量间均为负相关关系,但不同深度土壤坚实度对大豆产量影响程度不同,表层(0~30 cm)土壤坚实度对大豆产量影响最大;采用GA-RF模型预测土壤坚实度对大豆产量影响准确率达95.12%,较传统RF模型提高7.31%,与其他常用机器学习模型相比,GA-RF模型预测准确率及宏平均后查准率、召回率、F1值更优。GA-RF对丰富和完善RF方法具有一定理论意义,预测结果可为保护黑土资源、促进农业机械化高质量转型提供理论和技术支撑。In order to explore the relationship between soil penetration resistance and soybean yield after mechanical compaction,this research used the data from mechanical compaction test in black soil area,improved random forest(RF)based on the principle of genetic algorithm(GA),and thus established a prediction model of the impact of soil penetration resistance on soybean yield based on improved Random forest(GA-RF),which overcame the subjectivity of parameter selection of traditional RF model.The research realized effective assessment for the risk of mechanical compaction.The results showed that there was a negative correlation between soil penetration resistance at all the depths and soybean yield,but the influence of soil penetration resistance at different depths on soybean yield was different,the surface soil firmness(0-30 cm)had the largest influence on soybean yield.The accuracy of GA-RF model in predicting the impact of soil penetration resistance on soybean yield was 95.12%,which was 7.31%higher than that of RF.Compared with common machine learning models,GA-RF model had better accuracy and macro average precision,recall and F1 value.The improved GA-RF model enriched and improved the RF method,which had certain theoretical significance.The prediction could provide theoretical and technical support for protecting black soil resources and promote the high-quality transformation of agricultural mechanization.

关 键 词：机械压实 土壤坚实度 大豆产量 随机森林 遗传算法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] S565.1[自动化与计算机技术—控制科学与工程] S233.1[农业科学—作物学]