摘要
电网故障的准确判断与精准定位是电网安全运行的重要保障措施,传统电网保护方法是基于专家经验和离线仿真与整定的方法。随着电网不确定性因素与复杂性的大幅提高,这些方法因精确度与准确率不足会引起保护误动作与拒动,给电力系统的安全稳定运行带来巨大风险。为有效提高电网线路故障诊断的速度和准确性,提出一种利用遗传算法(Genetic Algorithm, GA)优化后的粗糙集(Rough Set, RS)-BP(back propagation)神经网络进行电网线路故障定位的方法。首先采用GA结合粗糙集RS对数据进行预处理,得到最小约简属性集,进而简化BP神经网络结构;其次根据遗传算法全局寻优的特点,采用GA优化BP神经网络的初始权值和阈值;然后使用BP算法细化局部搜索,避免陷入局部极小值,最终形成GA优化RS-BP神经网络的故障定位模型。仿真结果对比发现,该模型能够准确有效地进行故障定位,提高故障诊断速度和准确率,具有可行性和有效性。
Accurate judgment and precise positioning of power grid faults are important safeguards for safe operation of power grids.Traditional grid protection methods are based on expert experience and off-line simulation and tuning.With the significant improvement of grid uncertainties and complexity,these methods are accurate.Insufficient accuracy will cause protection failures and refusal,posing a huge risk to the safe and stable operation of the power system.In order to effectively improve the speed and accuracy of power line fault diagnosis,this paper proposes a genetic algorithm(GA)optimized Rough Set(RS)-BP(back propagation)neural network for fault location of power grid lines methods.Firstly,the data is preprocessed by GA combined with rough set RS to obtain the minimum reduction attribute set,which simplifies the BP neural network structure.Secondly,according to the characteristics of global optimization of genetic algorithm,the initial weight and threshold of BP neural network are optimized by GA.Then BP algorithm is used to refine the local search to avoid falling into local minimum values,and finally form the fault location model of GA-optimized RS-BP neural network.The simulation results show that the model can accurately and effectively locate faults and improve the speed and accuracy of fault diagnosis,which is feasible and effective.
作者
柴尔烜
曾平良
马士聪
邢浩
赵兵
CHAI Erxuan;ZENG Pingliang;MA Shicong;XING Hao;ZHAO Bin(School of Automation,Hangzhou Dianzi University,Hangzhou 310018,China;China Electric Power Research Institute,Beijing 100192,China)
出处
《电力科学与工程》
2019年第9期22-28,共7页
Electric Power Science and Engineering
基金
国家电网公司科技项目(大型交直流混联电网故障特征深度学习及智能识别和控制应用研究)
关键词
电网
故障诊断
遗传算法
粗糙集
神经网络
power grid
fault diagnosis
genetic algorithm
rough set
neural network
