Fuzzy neural network optimization and network traffic forecasting based on improved differential evolution
Fuzzy neural network optimization and network traffic forecasting based on improved differential evolution
复制标题
DOI：
10.1016/j.future.2017.08.041
复制
发表时间：
2018-04-01
影响因子：
7.5
通讯作者：
Lu, Huaiwei
中科院分区：
文献类型：
--
作者：
Hou, Yue;Zhao, Long;Lu, Huaiwei
传统的模糊神经网络在进行参数辨识时，往往采用BP算法进行参数优化。然而，BP算法容易陷入局部极值。针对该方法的不足，本文将差分进化算法与BP算法相结合，提出了一种改进的差分进化BP算法来优化模糊神经网络预测网络流量。针对差分进化算法存在的收敛速度慢、易早熟等问题，提出了一种基于自适应变异算子和高斯扰动交叉算子的改进差分进化算法，旨在对标准差分进化算法的变异和交叉算子的设计进行改进。为了验证该算法的有效性，将优化后的模糊神经网络预测算法应用于四个标准测试函数和实际网络流量。仿真结果表明，该算法的收敛速度和预测精度均优于传统的模糊神经网络算法。它不仅提高了模糊神经网络的泛化能力，而且提高了网络流量的预测精度。(C)2017爱思唯尔B.V.保留所有权利。
The traditional fuzzy neural network often uses BP algorithm to optimize parameters when conducting parameter identification. However, BP algorithm tends to be trapped in local extremum. In view of the shortcomings of this method, this paper combines the differential evolution algorithm with the BP algorithm, and proposes an improved differential evolution BP algorithm to optimize the fuzzy neural network forecasting network traffic. In order to solve problems such as slow convergence speed and tendency of premature convergence existing in differential evolution algorithm, an improved differential evolution algorithm using the adaptive mutation operator and Gaussian disturbance crossover operator aims to improve the mutation of standard differential evolution algorithm and the design of crossover operators. To validate the effectiveness of it, this optimized fuzzy neural network forecasting algorithm is applied to four standard test functions and the actual network traffic. Simulation results show that the convergence speed and forecasting accuracy of the proposed algorithm are better than those of the traditional fuzzy neural network algorithm. It improves not only the generalization ability of the fuzzy neural network but also the forecasting accuracy of the network traffic. (C) 2017 Elsevier B.V. All rights reserved.