基于CatBoost的恶劣天气下电网设备故障预测模型  

CatBoost-based fault prediction model for grid equipment under severe weather

作　　者：江上航 王国庆 朱建明 黄钧 JIANG Shanghang;WANG Guoqing;ZHU Jianming;HUANG Jun(School of Engineering Science,University of Chinese Academy of Sciences,Beijing 100190,China;School of Emergency Management,University of Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]中国科学院大学工程科学学院,北京100049 [2]中国科学院大学应急管理科学与工程学院,北京100049

出　　处：《安全与环境工程》2025年第2期133-142,共10页Safety and Environmental Engineering

基　　金：国家自然科学基金项目(72074202)。

摘　　要：恶劣天气对电网设备故障的发生有着显著影响,进而威胁电网的安全运行。综合气象特征与地理信息特征,提出一种恶劣天气下基于CatBoost(分类数据提升树)算法的电网设备故障预测模型。首先收集故障信息、实时气象和地理信息等原始数据,并使用MICE Forest(多重插补随机森林)算法填补缺失值;然后根据数据特征属性进行特征衍生得到时间、气象等方面的8个新特征,并利用递归特征消除方法结合交叉验证方法,评估特征的重要度,选取重要度最高的特征作为输入来构建模型;接着以最高准确率为优化目标,使用贝叶斯算法得到模型最优参数;最后在测试集上对模型准确率、精确率和调和均值F1进行验证。结果表明:提取得到的新特征有助于提高模型的预测精度,并且,与其他模型相比具有更高的预测精度,其准确率为0.831,精确率为0.858,调和均值为0.756。Severe weather has a significant impact on the failure of power grid equipment,affecting the safe operation of the power grid.A power grid equipment fault prediction model based on CatBoost algorithm is proposed by integrating meteorological and geographic information characteristics under adverse weather conditions.Firstly,raw data such as fault information,real-time weather and geographic information were collected,and the MICE Forest(multiple imputation by chained equations forest)algorithm was used to fill in missing values.Then,based on the feature attributes,eight new features in terms of time,weather,etc.were derived.The recursive feature elimination cross validation method was used to evaluate the importance of the features,and the most important feature was obtained as input to build the model.The highest accuracy was used as the optimization objective,and the Bayesian algorithm was used to obtain the optimal parameters of the model.Finally,the accuracy,precision,and harmonic mean F1 of the model was validated on the test set.The results show that the extracted new features help improve the prediction accuracy of the model,and have higher prediction accuracy compared to other models,with an accuracy of 0.831,an accuracy of 0.858,and a harmonic mean of 0.756.

关 键 词：恶劣天气 CatBoost 特征工程 电网设备故障 预测模型 

分 类 号：X934[环境科学与工程—安全科学]