基于改进粗糙集约简算法的配电系统在线故障诊断

On-line Fault Diagnose of Distribution System Based on Modified Rough Sets Reduction Algorithm

针对在实际的配电系统故障诊断中,往往要面对从海量数据中找到真正对于诊断结果有帮助的关键数据问题,提出了一类基于遗传算法的粗糙集约简算法,可以在较少信息的情况下获得正确的诊断结果。在分析了遗传算法中各个主要参数对算法结果的影响基础上,重点讨论了适应值函数中目标函数、惩罚函数以及惩罚因子的构造方法,并对于其他关键参数及算法进化过程针对粗糙集约简特点进行了修正。与传统的粗糙集约简算法相比,文中的方法能够有效找到最优的属性约简结果,同时大大提高了算法的效率和实用性。将算法应用于美国PG&E的69节点配电系统进行仿真,对于202个属性,319条记录的复杂数据进行了有效的约简,结果表明算法对于实际的复杂配电系统能够进行故障诊断。

A rough sets reduction algorithm based on genetic algorithm is presented to the puzzle problem of key data acquirement in the real complex distribution system fault diagnosis with thousands of data. By this approach, we can get right diagnosis conclusion with less information. The genetic algorithm effect of genetic parameters to the evolutionary process is analyzed. Furthermore the fitness function, punishing function and punishing factor are emphasized to study. The reduction method with utilization of the capability of searching for global optimum of genetic algorithm achieves better reduction result compared with classical rough sets reduction algorithm. Example of America PG＆E distribution power system with 69 nodes shows that the attribute reduction by genetic algorithm accelerates the evolutionary process and avoids premature convergence effectively for the system with 202 attributes and 319 records. According to the example, it shows that this method makes the feasibility of fault diagnosis in complex distribution system with thousands of data.