Dynamic security region algorithm considering sample imbalance and misclassification cost

Due to the high dimension and redundant features of power system operating data, the disadvantages of manual analysis of power network security region gradually appear. SVM algorithm provides a new solution for mining the boundary of power grid security region. However, the data set of power system safety analysis has the problems of unbalanced samples and unequal misclassification costs. As a result, the traditional SVM algorithm, which takes classification accuracy as the optimization goal, cannot meet the requirements of power grid security region mining. Therefore, this paper solves the above problems by increasing the penalty for misclassification of unstable samples at the algorithmic level. In order to improve the classification accuracy and reduce the missed alarm rate, a dynamic security region algorithm is proposed, which takes into account the imbalance of samples and the difference of misclassification costs. Firstly, by introducing penalty parameters of sample misclassification cost, the influence of penalty parameters on classification accuracy and over-fitting index is analyzed. In order to reduce the rate of missing alarm, the SVM algorithm is modified by increasing the penalty for misclassification of unstable samples. In order to improve the classification accuracy of the algorithm, the concept of gray scale interval is introduced. By introducing grayscale interval into the output category probability, the classification results of samples outside the interval are almost reliable. Samples located within the interval can be further determined by time domain simulation. The balance between classification accuracy and missing alarm rate is realized. Finally, a CEPRI36 nodes system example is used to verify the effectiveness of the model and algorithm.