基于气象特征挖掘与AdaBoost-MEA-ELM模型的绝缘子盐密预测  

Prediction of Insulator ESDD Based on Meteorological Feature Mining and AdaBoost-MEA-ELM Model

作　　者：王尧平 李特 姜凯华 李文辉 吴强[1] 王羽[1] WANG Yaoping;LI Te;JIANG Kaihua;LI Wenhui;WU Qiang;WANG Yu(School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China;State Grid Zhejiang Electric Power Research Institute,Hangzhou 310014,China;State Grid Taizhou Power Supply Company,Taizhou 318000,China)

机构地区：[1]武汉大学电气与自动化学院,湖北武汉430072 [2]国网浙江省电力有限公司电力科学研究院,浙江杭州310014 [3]国网浙江省电力有限公司台州供电公司,浙江台州318000

出　　处：《中国电力》2023年第9期157-167,共11页Electric Power

基　　金：国网浙江省电力有限公司科技项目(B311DS221005)。

摘　　要：为及时掌握输电线路绝缘子污秽情况,提出了一种基于气象的绝缘子盐密预测方法。挖掘了与积污相关性更强的气象特征,通过随机森林评估了气象特征的重要程度,结合序列前向搜索确立了最佳气象特征子集。基于台州市自然积污测试数据,使用极限学习机(extreme learning machine,ELM)建立盐密预测基础模型,并使用思维进化算法(mind evolution algorithm,MEA)对其初始权值与阈值进行优化,通过自适应提升(adaptive boosting,AdaBoost)算法集成进一步提高模型精度。结果表明:AdaBoost-MEA-ELM模型盐密预测平均绝对误差为0.0032mg/cm2,相比原始ELM模型误差降低58.97%,优化效果显著;与其他模型对比验证了AdaBoost-MEA-ELM模型的性能以及3种算法结合的合理性;通过k折交叉验证获得了训练数据改变时模型误差的变化情况,进一步验证了模型的泛化性与稳定性。In order to obtain the pollution condition of transmission line insulators in time,a method of insulator equivalent salt deposit density(ESDD)prediction based on meteorological data is proposed in this paper.The meteorological features that are more closely related to insulator pollution degree are mined,and the importance of each meteorological feature is evaluated by the random forest algorithm.Combined with the sequential forward search method,the optimal subset of meteorological features for ESDD prediction model could be determined.Based on the natural pollution test data of Taizhou City,the basic ESDD prediction model was established by using extreme learning machine(ELM),and its initial weights and thresholds were optimized by the mind evolution algorithm(MEA).Then the AdaBoost algorithm was applied to further improve the accuracy of the model.The results show that the average absolute error of ESDD prediction of AdaBoost-MEA-ELM model is 0.0032 mg/cm2,which is 58.97%lower than that of the original ELM model.Compared with other common models,the performance of the proposed model and the rationality of the combination of these three algorithms are verified.The variation of prediction error when training data changed was obtained by k-fold verification method,which further prove the generalization performance and stability of the model.

关 键 词：等值盐密预测 气象特征 随机森林 极限学习机 思维进化 ADABOOST算法 

分 类 号：TM216[一般工业技术—材料科学与工程]