改进极限学习机应用于电网故障诊断

Improved Extreme Learning Machine Applied in Power System Fault Diagnosis

电网故障诊断中交叉数据模式识别问题占据重要位置,传统的人工智能方法处理效果不甚理想。提出运用改进极限学习机进行故障诊断的算法,随机选取输入权值向量和隐含层的偏差,并且利用最小二乘法分析计算输出权值,以达到提高故障诊断容错性的目的。仿真结果表明:在保护动作信息不完备的情况下,该算法的故障判断准确性明显优于BP神经网络,该算法对存在一定错误数据的故障信息也具有良好的识别能力。

The cross data pattern recognition plays a very important role in the fault diagnosis in the power grid,but the effect of the traditional artificial intelligence method is limited.This paper proposes that the improved extreme learning ma-chine be used for fault diagnosis,the input weight vector and the deviation of the hidden layer be randomly selected,and the output value be calculated and analyzed by using least square method in order to improve the fault toleration. The simulation results show that, in the condition of incomplete protection action information,the accuracy of the fault judgment algorithm of is better than BP neural network. In addition,the algorithm has better recognition ability for the fault information containing certain error data.