基于深度学习与内核岭回归的电力系统鲁棒状态估计

Robust State Estimation of Power System Based on Deep Learning and Kernel Ridge Regression

针对电力系统状态估计中量测数据存在非高斯噪声、估计结果精度有限和时效性不高的问题,该文提出了一种结合深度学习框架与内核岭回归的电力系统预测辅助鲁棒状态估计方法。首先,根据电力系统的历史运行数据,将基于注意力机制的卷积神经网络与长短期记忆神经网络相结合构建预测模型。其次,通过支持向量机来检测异常与缺失数据,实现数据分类,并基于改进的预测模型,在量测异常情况下输出状态量;再次,使用内核岭回归模型建立量测量与状态量的非线性映射函数,对含有非高斯噪声的量测量进行滤波;最后,在IEEE 118节点和IEEE 300节点测试系统上进行数值仿真,结果表明本文所提方法具有较高的精确性和鲁棒性。

To solve the problems of non-Gaussian noise in measurement data,limited accuracy of estimation results,and low efficiency in power system state estimation,a robust forecasting-aided state estimation method(FASE)for power system based on deep learning framework and kernel ridge regression(KRRSE)is proposed.Firstly,according to the historical operation data of the power system,the convolutional neural network based on attention mechanism combined with the neural network of long-short term memory(ATT-CNN-LSTM)is used to formulate the prediction model.Secondly,the support vector machine(SVM)is used to detect abnormal and missing data and realize data classification,and output state estimate result based on the ATT-CNN-LSTM model under abnormal condition.Thirdly,the nonlinear mapping function between the quantity measurement and the state quantity is established by using the kernel ridge regression model,which can filter out the non-Gaussian noise in measurement data.Finally,case studies on the IEEE 118-bus system and 300-bus system verify the high accuracy and robustness of the proposed KRRSE method.