用PSO算法训练神经网络抑制发电机局放随机脉冲干扰  

Application of PSO Based Neural Network in Suppression of Stochastic Pulse Interference for Partial Discharge Monitoring in Large Generators

作　　者：邵震宇[1] 黄成军[1] 肖燕[1] 赵亚奎[1] 江秀臣[1] 

机构地区：[1]上海交通大学电气工程系,上海市200030

出　　处：《电力系统自动化》2005年第11期49-52,共4页Automation of Electric Power Systems

基　　金：中华电力教育基金会基金资助项目

摘　　要：随机脉冲干扰在局部放电在线监测的各类干扰中是最难抑制的,为此提出了一种基于粒子群优化(PSO)算法训练神经网络的随机脉冲干扰抑制算法。PSO算法的优势在于它能通过粒子间的相互作用而发现复杂搜索空间的最优区域。与传统反向传播(BP)算法相比,采用PSO算法来训练神经网络,可以有效地克服传统算法收敛速度慢、易陷于局部极小值等缺点,并且训练出的神经网络在泛化能力上也有很大的提高。大量实际数据的训练和分析结果表明,该算法在抑制局放随机脉冲干扰上是比较有效的。The stochastic pulse interference is the most difficult to be suppressed among all the interferences in the partial discharge (PD) on-line monitoring. This paper proposes a new pattern recognition method based on neural network to suppress the stochastic pulse interference. The method utilizes the PSO algorithm to speed up the learning of the neural network, and doesn't have the disadvantages of the BP algorithm such as slow convergence and local minimum. The simulation results have validated that the proposed method is correct and effective in suppression of the stochastic pulse interference.

关 键 词：局部放电 随机脉冲干扰 粒子群优化 神经网络 

分 类 号：TM31[电气工程—电机]