基于混合重抽样和LightGBM算法的配变低压跳闸预测

针对配变台区在夏季用电高峰期易频繁跳闸的问题,提出一种基于混合重抽样和LightGBM算法的配变低压跳闸预测模型.为了解决数据分布的边缘化问题,首先采用隔离森林剔除样本中的离群值.其次采用NCL欠抽样与SMOTE过抽样相结合的混合重抽样方法处理训练样本的数据不平衡问题.然后采用混合重采样算法产生的新样本对LightGB...

Low-voltage tripping prediction of a distribution transformer based on hybridresampling and a LightGBM algorithm

There are frequent tripping faults in the distribution transformation area during the summer peak period. A low-voltage trip prediction model based on a hybrid resampling method and the LightGBM algorithm is proposed. First, an isolation forest is used to eliminate outliers in the samples to solve the problem of data distribution marginalization. Secondly, a mixed resampling method combining NCL under-sampling and SMOTE over-sampling is used to handle the data imbalance of training samples. Thirdly, the LightGBM classifier is trained by the new samples generated by the hybrid resampling algorithm. Finally, the probability of low-voltage tripping faults in the target station area is predicted by the well-trained classifier. The experimental results show that the proposed iF-SMOTE-NCL-LightGBM model achieves the highest performance evaluation indicators, among other prediction models, in low-voltage trip prediction, and can effectively predict low-voltage tripping events.This work is supported by the Science and Technology Project of China Southern Power Grid Co., Ltd. (No. GZJKJXM20170049).