论文部分内容阅读
神经网络的输入变量、隐含层结点以及中心的选择对模型的性能都有重大的影响,以前的研究一般只考虑优化网络的参数或其结点数。为解决这个问题,提出了一种新的全局优化算法来自动选择RBF神经网络的输入变量和结点数目,并同时优化其参数。在提出的算法中,RBF网络的结点数目、输入变量的选择和参数都采用二进制编码,并用遗传算法来优化。为提高算法的性能和收敛速度,在遗传算法优化的同时引入了一种高性能的基于梯度的局部搜索算子(结构化的非线性参数优化方法)来优化RBF网络中的参数。Box-Jenkins煤气炉标准时间序列的预测问题被用来检验算法的性能。实验结果表明,提出的算法可以得到非常“紧凑”的RBF网络,且其性能优于其他一些算法。
The input variables of neural network, the hidden layer nodes and the choice of center all have a great influence on the performance of the model. In the past, only the parameters of optimization network or the number of nodes were considered. In order to solve this problem, a new global optimization algorithm is proposed to automatically select the input variables and the number of nodes in the RBF neural network and to optimize its parameters simultaneously. In the proposed algorithm, the number of RBF network nodes, the choice of input variables and parameters are used binary coding, and genetic algorithms to optimize. In order to improve the performance and convergence speed of the algorithm, a high-performance gradient-based local search operator (structured non-linear parameter optimization method) is introduced to optimize the parameters in the RBF network. The Box-Jenkins gas stove standard time series prediction problem was used to test the performance of the algorithm. The experimental results show that the proposed algorithm can get very “compact” RBF network and its performance is better than some other algorithms.