Abstract: An optimization method of RBF neural network based on simulated annealing and hybrid hierarchy genetic algorithm is put forward. In this method, the network topology, centers and radius of RBF neural network are optimized by hybrid hierarchy genetic algorithm, the probabilities of cross and mutation in genetic algorithm are controlled by simulated annealing algorithm, and the output weights of network are calculated by least square method. To validate the feasibility and effectiveness, this method and other four methods are implemented in typical case, the result shows that the accuracy of the proposed method is obviously higher than other methods. The feasibility and superiority of the method are validated.

Key words: Simulated Annealing（SA）, Hybrid Hierarchy Genetic Algorithm（HHGA）, Radial Basic Function Neural Network（RBFNN）, fault diagnosis

摘要： 提出一种利用模拟退火和混合递阶遗传算法优化RBF神经网络的方法。通过利用混合递阶遗传算法对RBF神经网络的拓扑结构、径向基中心和半径进行参数寻优，引入模拟退火算法对交叉和变异概率进行控制，采用最小二乘法确定网络的输出权值。将此方法应用于典型实例，并与其他四种方法进行对比，通过试验结果证明了该方法的准确率明显优于其他四种方法，方法的可行性和优越性得到验证。

关键词: 模拟退火, 混合递阶遗传算法, 径向基神经网络, 故障诊断