分类梯度提升算法(CatBoost)与蝙蝠算法(Bat)耦合建模预测中国西北部地区水面蒸发量  

Coupling CatBoost Model with Bat Algorithm to Simulate the Pan Evaporation in Northwest China

作　　者：董力铭 曾文治[1] 雷国庆 DONG Li-ming;ZENG Wen-zhi;LEI Guo-qing(State Key Laboratory of Water Resources and Hydropower Engineering Science,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学水资源与水电工程科学国家重点实验室,武汉430072

出　　处：《节水灌溉》2021年第2期63-69,共7页Water Saving Irrigation

基　　金：国家自然科学基金项目(51879196)。

摘　　要：为预测干旱、半干旱地区的水面蒸发量,以实测水面蒸发量为基准,利用我国西北部地区的45个气象站实测逐日最低气温、最高气温、相对湿度、风速及太阳辐射,建立基于蝙蝠算法的分类梯度提升算法模型(Bat-CB)并与原分类梯度提升算法模型(CatBoost)及随机森林模型(RF)进行对比。结果表明Bat-CB模型的预测能力与稳定性(RMSE为0.859~2.227 mm/d;MAE为0.540~1.328 mm/d;NSE为0.625~0.894;MAPE为0.162~0.328)最优,而CatBoost模型(RMSE为0.897~2.754 mm/d;MAE为0.531~1.77 mm/d;NSE为0.147~0.869;MAPE为0.161~0.421)对水面蒸发的预测能力略优于随机森林模型(RMSE为1.005~3.604 mm/d;MAE为0.644~2.479 mm/d;NSE为-1.242~0.894;MAPE为0.176~0.686)。此外,Bat-CB模型对于水面蒸发的季节及空间变异性具有较好的适应能力,而随机森林模型在夏季水面蒸发预测中表现最差。研究成果可用于干旱、半干旱地区水面蒸发量的预测。In order to predict the water surface evaporation in arid and semi-arid regions,the measured pan evaporation of 45 weather stations in Northwest China were used as the benchmark,the daily minimum temperature,maximum temperature,relative humidity,wind speed and solar radiation were applied to simulate the pan evaporation using the original classification gradient boosting algorithm model(Catboost),the random forest model(RF)and the Catboost model coupled with bat algorithm(Bat-CB)model respectively.The results indicated that the prediction ability and stability of the Bat-CB model(RMSE:0.859~2.227 mm/d;MAE:0.540~1.328 mm/d;NSE:0.625~0.894;MAPE:0.162~0.328)were the best.The CatBoost model(RMSE:0.897~2.754 mm/d;MAE:0.531~1.77 mm/d;NSE:0.147~0.869;MAPE:0.161~0.421)was slightly better than the RF model in predicting water surface evaporation(RMSE:1.005~3.604 mm/d;MAE:0.644~2.479 mm/d;NSE:~1.242~0.894;MAPE:0.176~0.686).In addition,the Bat-CB model had good adaptability to the seasonal and spatial variability of water surface evaporation,while the random forest model performed the worst in summer water surface evaporation prediction.

关 键 词：水面蒸发 机器学习 蝙蝠算法 随机森林模型 分类梯度提升算法 

分 类 号：S27[农业科学—农业水土工程]