基于改进ELM神经网络的客户满意度评价模型

摘要

使用一种动态递归网络———ELM神经网络来模拟专家打分进行电力客户满意度测评。仿真结果表明， ELM神经网络具有训练速度快和结构简单的特点，能较准确地反映客户满意度。同时，针对ELM神经网络基于梯度下降算法调整权值和阈值，容易陷入局部最优的缺陷，提出了利用入侵杂草算法（ IWO）优化ELM神经网络的连接权值系数。神经网络权值优化是一个大规模多峰优化问题，已有文献证明IWO算法对于解决高维度、多峰优化问题具有明显优势。新方法有效弥补了单一算法的不足，拥有ELM神经网络动态记忆的能力以及入侵杂草算法全局收敛性强的特点。实例计算证明，改进ELM神经网络可以建立精度更高的电力客户满意度评价模型，保证专家评价系统的一致性和稳定性，是一种行之有效的评价方法。%A dynamic recurrent neural network, namely ELM neural network simulating the assessment of expert scoring has been used to evaluate the electric power customer satisfaction. The calculation of real examples shows that this method is capable to reflect the lev⁃els of customer satisfaction accurately with the advantages of fast training speed and simple structure. At the same a method for optimi⁃zing the connecting weight value coefficient of ELM neural network is presented by using the global searching ability of IWO. The opti⁃mization of neural network parameters is a large scale multimodal optimization problem and the tests show that IWO has obvious advan⁃tages in solving high⁃dimensional multimodal optimization problem particularly. This new approach combines the merits of ELM neural network that has the ability of dynamic memory and the strong global searching capability of IWO which exactly makes up the shortcom⁃ings of single algorithm. The simulations reveal that neural network optimized by IWO is able to build a higher precision modal for the e⁃valuation of electric power customer satisfaction and guarantee the uniformity and stability of expert evaluating system.