基于广义回归网络-改进差分进化算法的污闪电压预测研究  

Research of polluted flashover voltage forecasting on the basis of general regression neural network-modified differential evolution

作　　者：帅海燕[1,2] 龚庆武[1] 张园园[1] 

机构地区：[1]武汉大学电气工程学院,湖北武汉430072 [2]武汉交通职业学院,湖北武汉430064

出　　处：《电力系统保护与控制》2010年第19期107-113,136,共8页Power System Protection and Control

基　　金：科技部科学技术重点项目(NCSIE-2006-JKZX-174)

摘　　要：将基于实验数据和数学公式计算值用广义回归网络-改进差分进化算法为绝缘子污闪电压建立一种新的预测模型,以绝缘子的盘径、高度,爬电距离、形状因素四个结构参数及等值附盐密度为输入参数来预测污闪电压值。广义回归网络不需设定模型的形式,但其平滑因子参数需优化估值。为了克服传统差分进化算法优化参数时的弱点,改进差分算法引入Powell寻优法以提高算法搜优速度,同时引入混沌优化法以提高种群多样性,降低算法陷入局部最优的概率。仿真结果表明与GRNN-DE及多元线性回归相比,GRNN-MDE具有更为优良的预报能力,稳定性也更好,将它应用于绝缘子污闪电压的预测,效果更好。Based on the data derived from experimental measurements and a mathematical model ,the paper constructs a new critical flashover voltage forecasting model using general regression neural network-modified differential evolution. The model uses the four characteristics of insulator ,n amely ,d iameter ,h eight, c reepage distance and form factor, and equivalent salt deposit density as the inputs to estimate the critical flashover voltage. The general regression neural network does not need the fixed model form, but the smoothing factors should be valued optimally .I n order to overcome the flaws lying in basic differential evolution, modified differential evolution introduces Powell searching operation to expedite the speed of the algorithm and invites chaos optimization to improve the diversity of populations and reduce the algorithm’s probabilities of slumping a local optimal solution .The four results show that compared with GRNN-DE and multivariate linear regression（ MLR）,GRNN-MDE has more excellent forecasting capability and eminent stability , which, once is used to forecast contaminated insulator critical flashover voltage ,w orks better. This work is supported by Ministry of Science and Technology of China （NCSIE-2006-JKZX-174）.

关 键 词：广义回归网络 差分进化算法 Powell寻优法 混沌优化法 绝缘子结构参数 临界闪络电压 

分 类 号：TM852[电气工程—高电压与绝缘技术] TM216.04[一般工业技术—材料科学与工程]