电力系统动态安全域的LS-SVM在线拟合法

On-line Imitating Method Based on LS-SVM of Dynamic Security Regions in Power System

提出了一种基于支持向量机的电力系统动态安全域在线拟合方法。支持向量机在解决非线性有限样本和高维识别方面有明显优势,但标准支持向量机在学习时需要求解复杂的二次规划问题,耗时较多。为此采用最小二乘支持向量机的二值分类算法,构造了二类和三类分类器对运行点的稳定状态进行判断,以最小二乘线性系统代替二次规划方法的标准支持向量机进行模式识别和函数估计,解决了大样本数据建模和运算速度慢的问题。同时采用回归算法构造稳定裕度拟合器,对系统既定故障下运行点的临界切除时间进行在线拟合并计算出稳定裕度。最后以EPRI-36节点模型为算例进行仿真计算,仿真结果表明该方法避免“维灾难”问题的同时,能更好地拟合动态安全域的边界,且进一步证明了该方法的有效性和准确性。

This paper presents an on-line imitating method of dynamic security regions in power system, based on Support Vector Machine （SVM）. SVM has great ability to solve the problem of nonlinear limited sample and high dimension recognition. But it costs much time for classical SVM to solve the quadratic programming problem, This paper uses the two-value classification arithmetic of Least Squares Support Vector Machine （LS-SVM） to build binary classifier, 3-classification classifiers for judging the state of running points, Instead of quadratic programming method of classical SVM, LS-SVM uses least squares linear system for pattern recognition and function estimation, which solves the problem of great sample for building model, and speeds up operation rate, At the same time, the stability margin imitation machine is built by using the regression arithmetic for imitating the critical cutting-time on-line and computing its transient stability margin, to same fixed fault. It is tested on the EPRI-36 bus model of PSASP, and the result indicates the method can avoid the ＂dimensions misfortune＂ problem with imitating the dynamic security regions＇ boundary better, and further proves the validity and accuracy of the theory.