基于RF特征优选的AEA⁃ResNet柱上断路器运行状态诊断

Operation status diagnosis of pole-mounted breakers based on RF feature optimization and AEA-ResNet

为实现柱上断路器运行状态的智能高效诊断,提出一种基于随机森林(random forest,RF)的特征优选算法,并利用遗传模拟退火算法(annealing evolution algorithm,AEA)优化残差神经网络(residual neural network,ResNet),实现设备状态的智能预测。首先构建包含22维特征的断路器运行状态数据库,通过RF算法计算各特征的重要度指标,并通过序列反向搜索的方式保留11维特征作为后续模型的输入。然后,利用AEA算法对ResNet的网络结构进行迭代优化,识别最优参数用于模型预测。最后,仿真结果表明,RF算法可有效避免特征冗余,提高模型的预测性能。与传统预测模型相比,AEA-ResNet模型可以显著提升预测准确率,尤其在少数类样本的召回率和精度方面优势明显。

To evaluate the operation status of pole-mounted breakers in an intelligent and efficient manner,the random forest(RF)is em-ployed for feature optimization,and the annealing evolution algorithm(AEA)is applied to optimize the parameters of the re-sidual neural network(ResNet).Firstly,a database is constructed,encompassing 22-dimensional operational features of pole-mounted breakers,with their importance indices calculated using the RF.Through the reverse sequence search method,11 features are determined as inputs.Subsequently,the AEA is employed to optimize the network parameters of ResNet.Sim-ulation results indicate that the RF effectively eliminates feature redundancy and improves the prediction performance of the model.In comparison to traditional prediction models,the proposed AEA-ResNet method significantly improves the accuracy,especially in the recall and precision of minority samples.