基于机器学习的继电保护误动作识别方法研究

打开文本图片集

RelayProtectionMisoperationIdentificationBasedon Machine Learning

HUANG Kui （State GridFujian Electric PowerCo.，Ltd.，UltraHigh Voltage Branch，Fuzhou35OoO1，China）

Abstract： [Purposes] To address the diffculties and low accuracy in identifying relay protection misoperations in power systems，a machine learning-based method for misoperation identification is proposed. [Methods] First， the sampled data were normalized and outliers were removed. Then，feature parameters such as the rate of change of voltage amplitude and current mutation were extracted using waveletFourier transform. Subsequently，three base classifiers—support vector machine （SVM），random forest， and deep neural network （DNN）—were constructed.Finally，a multi-model ensemble was implemented using a weighted voting fusion strategy with weights optimized bya genetic algorithm.[Findings] Experimental results show that the proposed method achieves an identification accuracy of 96.8% under load mutation conditions，which is 15.3% higher than that of traditional methods.[Conclusions] The method significantly improves the accuracy of misoperation identification and has important engineering application value.

Keywords： relay protection; misoperation identification; machine learning; feature extraction; model fusion

0 引言

随着电网规模的不断扩大以及新能源并网数量的持续增多，继电保护面临着更为复杂的运行环境，误动作发生的概率明显提升。（剩余4800字）