基于XGBoost和网格搜索的变压器油中溶解气体含量预测

Prediction on Dissolved Gas Content in Transformer Oil Based on XGBoost and GridSearchCV

对变压器的故障检测通常使用油中溶解气分析的方法,目前对变压器油中溶解气体的含量预测逐渐成为了研究热点.将XGBoost算法应用于油中溶解气体含量的预测,并且与支持向量机等5种算法进行比较,验证了XGBoost算法的可行性和精确性;并构建了XGBoost和网格搜索相结合的模型,该模型利用网格搜索法对XGBoost算法进行超参数筛选和优化,显著提升了XGBoost算法的性能.最后,利用内蒙古根河市110kV变压器油中溶解气体数据,经过网格搜索法筛选出5个超参数,并对其优化.油中溶解气体含量预测仿真结果表明,超参数优化后的XGBoost算法的均方根误差和平均绝对百分比误差都有所下降,预测精度明显优于未进行超参数优化的模型.

As one of the key equipment in the operation of power grid,the working state of transformer is closely related to the safe operation of power system.At present,dissolved gas analysis method is usually used in transformer fault detection.The prediction of dissolved gas content in transformer oil has gradually become a hot topic.In this paper,XGBoost algorithm is applied to the prediction of dissolved gas content in oil,and compared with five algorithms such as support vector machine.The feasibility and accuracy of XGBoost algorithm are verified.Then,a model combining XGBoost and GridSearchCV is proposed.The model uses GridSearchCV method to screen and optimize the super parameters of XGBoost algorithm,which significantly improves the performance of XGBoost algorithm.Finally,using the dissolved gas data in 110 kV transformer oil in Genhe city,Inner Mongolia,the feasibility and accuracy of XGBoost applied to the prediction of dissolved gas content in oil are verified through simulation experiments.Further,five super parameters are selected and optimized by GridSearchCV method.The simulation results show that,the root mean square error and average absolute percentage error of XGBoost algorithm after superparametric optimization are reduced,and the prediction accuracy is obviously better than that of the model without superparametric optimization.