基于CatBoost的温室日参考作物蒸发蒸腾量估算模型研究  

A CatBoost Model for Simulating the Daily Reference Evapotranspiration in Greenhouse

作　　者：牛曼丽 李红岺 李新旭[1] NIU Man-li;LI Hong-ling;LI Xin-xu(Beijing Agricultural Technology Extension Station,Beijing 100029,China)

机构地区：[1]北京市农业技术推广站,北京100029

出　　处：《节水灌溉》2022年第1期14-19,共6页Water Saving Irrigation

基　　金：北京市科技计划“工厂化番茄质量控制技术研究与应用”(Z201100008020013)。

摘　　要：参考作物蒸发蒸腾量(Reference Evapotranspiration,ET_(0))是估算作物需水量、制定灌溉制度、提高用水效率,实现农业节水的重要参数。针对传统Penman-Monteith(P-M)公式计算作物蒸发蒸腾量需要参数多,计算复杂等问题,提出了一种基于支持分类特征的梯度提升决策树(CatBoost)算法估算温室日参考作物蒸发蒸腾量。以温室修正型Penman-Monteith公式计算的ET_(0)作为标准值,通过Pearson’s方法对输入参数与ET_(0)之间的相关性进行分析,组合不同输入特征向量。当输入参数组合为3参数,即平均室内温度、平均相对湿度、累积太阳辐射时,CatBoost性能最优,测试集估算精度MAE为0.220 mm/d,RMSE为0.310 mm/d。进一步对比了6种其他机器学习模型(XGBoost、AdaBoost、随机森林、决策树、KNN、SVM)的估算精度,结果表明CatBoost模型具有最佳的估算精度和稳定性,能够较好地模拟参考作物蒸发蒸腾量。构建的日参考作物蒸发蒸腾量估算模型为水肥精准化管理、灌溉控制系统研发提供了一种新的思路。Reference Evapotranspiration(ET_(0))is an important parameter for estimating crop water requirement,formulating irrigation system and improving water efficiency to achieve the goal of saving agricultural water saving.In order to solve the problem that the traditional Penman-Monteith(P-M)formula requires too many parameters and complex calculation,a gradient boosting decision trees(CatBoost algorithm based on supported categorical features was proposed to estimate the daily reference evapotranspiration of greenhouse crops.The ET_(0)calculated by greenhouse modified Penman-Monteith formula was taken as the standard value,and the correlation between input parameters and ET_(0)was analyzed by Pearson′s method,and different input feature vectors were combined as input.When the input parameter combination includes 3 parameters,namely average indoor temperature,average relative humidity and cumulative solar radiation,the performance of CatBoost model was the best with MAE=0.220 mm/d and RMSE=0.310 mm/d respectively.Furthermore,the estimation accuracy between six other machine learning models(XGBoost,AdaBoost,Random forest,decision tree,KNN,SVM)was presented.The results showed that CatBoost model had the estimation accuracy and stability,and could simulate the evapotranspiration of reference crops better.The evapotranspiration estimation model constructed in this paper provides a new approach to realize accurate management of water and fertilizer,development of irrigation control system.

关 键 词：CatBoost 参考作物蒸发蒸腾量 估算模型 

分 类 号：S274[农业科学—农业水土工程] TP181[农业科学—农业工程]