基于组合算法的RBF神经网络列车轮对缺陷识别

Identification of Train Wheel Set Defects Based on Composite Algorithm RBF Neural Network

提出了一种基于动态聚类和遗传算法相结合的组合RBF网络训练方法;采用动态聚类法对样本数据进行聚类,使RBF神经网络的隐含层节点中心数在训练过程中自动确定,使用经验公式作为标准,选取最优聚类数,采用遗传算法对隐层中心和宽度以及隐层到输出层的权值进行优化,在全局范围内寻找网络的最优模型;最后对轮对缺陷进行纹理特征提取,并组成训练样本和测试样本,输入到网络进行训练与测试;实验结果表明,与传统方法比较,该组合方法具有较高的识别率。

A new training method based on dynamic clustering algorithm and Genetic algorithm is presented for RBF. The sample data is classified by dynamic clustering algorithm and neural network hidden layer nodes are identified automatically in the training course. An experience formula is used to optimize them as the standard, then genetic algorithm is used to optimize the centers and widths of RBF and the weights between hidden layer and output layer and an optimization model is achieved. Finally the features of wheel set defects by texture feature extraction are used as the train and test sample and used in the training and testing of networks. Simulation results show that the composite algorithm can reach a high recognition rate than RBF neural network.