风力发电机温度时序预测方法优化  

Optimization of Wind Turbine Generator Temperature Time Series Prediction Method

作　　者：王言国[1] 秦冠军[1] 兰金江 WANG Yan-guo;QIN Guan-jun;LAN Jin-jiang(NR Electric Co.,Ltd.,Nanjing 211101,China;China Three Gorges Renewables(Group)Co.,Ltd.,Beijing 100032,China)

机构地区：[1]南京南瑞继保电气有限公司,江苏南京211101 [2]中国三峡新能源(集团)股份有限公司,北京100032

出　　处：《计算机技术与发展》2023年第9期215-220,共6页Computer Technology and Development

基　　金：中国长江三峡集团公司“三峡新能源集中监控系统”项目(JD-SX-01001)。

摘　　要：发电机作为风机核心部件,其温度过热往往是发电机故障的综合表现,业界和学术界对风机发电机温度预测做了很多的研究。该文在前人研究的基础上,对发电机温度时序预测模型构建中的多个环节进行了深入优化:通过数据集转换将时间序列数据集转换为有监督学习数据集,进而采用有监督学习算法;风机的实时测量值有数百个之多,通过量化自变量和因变量之间的非线性关系,进行更合理的特征筛选,剔除与预测目标弱相关或不相关的特征;基于无监督学习算法,实现了训练集中异常数据的自动过滤,并通过对数十个风场数据质量的统计分析,剔除影响建模精度约5%的脏数据;最后构建多个算法模型,并通过在相同计算环境上的横向对比实验,以均方误差、解释方差、R2 score作为衡量模型精度的指标,从各种建模算法中选择精度较高的算法。提出的一系列改进措施为提高风力发电机温度时序预测模型精度提供了参考。As the core component of the wind turbine generator,the overheating of the generator is often the comprehensive performance of the generator fault.The industry and academia have done a lot of research on the temperature prediction of the wind turbine generator.Based on previous research,we have deeply optimized several links in the construction of the generator temperature time series prediction model:the time series dataset is converted into a supervised learning dataset through dataset conversion,and the supervised learning algorithm is used.There are hundreds of real-time measured values of generator.Through quantifying the nonlinear relationship between independent variables and dependent variables,more reasonable feature screening is carried out to eliminate the features that are weakly related or unrelated to the prediction target.Based on the unsupervised learning algorithm,the automatic filtering of abnormal data in the training set is realized,and through the statistical analysis of the data quality of tens of wind farms,the dirty data that affects the modeling accuracy by about 5%is eliminated.Finally,several algorithm models are constructed.Through horizontal comparison experiments in the same computing environment,mean square error,explanatory variance and R2 score are used as indicators to measure the accuracy of the model,and algorithms with higher accuracy are selected from various modeling algorithms.A series of improvement measures proposed provide a reference for improving the accuracy of wind turbine generator temperature time series prediction model.

关 键 词：风力发电机 时序预测 数据变换 特征优选 孤立森林 XGBoost 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TK83[自动化与计算机技术—控制科学与工程]