基于极限学习机的电气设备运行数据异常检测研究

Study on anomaly detection of operation data of electrical equipmentbased on extreme learning machine

由于现有的数据异常检测方法检测结果与实际负荷变化情况之间存在显著差异,为此,文章研究基于极限学习机的电气设备运行数据异常检测。文章利用经验模态分解(Empirical Mode Decomposition, EMD)对电气设备状态噪声信号进行处理,将重构信号与核函数的乘积进行双重求和,通过信号在模糊空间内的重新映射,完成特征提取。运用极限学习机优化神经网络训练过程,以输出与目标向量维度相匹配的预测结果。结合Parzen窗概率密度估计方法对设备运行状态展开判断。通过实时采集设备数据,输入已训练的概率神经网络(Probabilistic Neural Networks, PNN)模型中,输出设备当前运行状态的概率分布,根据分布判断设备是否处于异常状态。实验结果表明,实验组在400 min时准确地捕捉到电气设备负荷的异常变化,与实际情况高度吻合,有效验证所提出的异常检测方法的准确性。

Due to the significant difference between the detection results of existing data anomaly detection methods and the actual load changes,this article studies the detection of electrical equipment operation data anomalies based on extreme learning machines.The article uses empirical mode decomposition(EMD)to process electrical equipment state noise signals,and double sums the product of the reconstructed signal and the kernel function.By remapping the signal in fuzzy space,feature extraction is completed.Optimize the neural network training process using extreme learning machines to output prediction results that match the target vector dimension.Use the Parzen window probability density estimation method to assess the operational status of devices.By collecting real-time device data and inputting it into a trained probabilistic neural network(PNN)model,the probability distribution of the current operating state of the device is output to determine whether the device is in an abnormal state based on the distribution.The experimental results showed that the experimental group accurately captured the abnormal changes in electrical equipment load at 400 minutes,which was highly consistent with the actual situation and effectively verified the accuracy of the proposed anomaly detection method.