混合优化RBF-BP网络的板形缺陷识别

针对传统板形模式识别方法存在精度低、鲁棒性弱的问题,提出了一种混合优化RBF-BP组合神经网络板形模式识别方法。首先利用自组织映射网络(SOM)对样本聚类,利用聚类后的网络拓扑结构确定RBF的中心,并计算RBF的宽度,克服了传统聚类算法随机选取中心导致聚类结果不稳定的问题。然后利用遗传算法(GA)良好的全局搜索能力优化整个网络的权值。RBF-BP组合神经网络是由一个RBF子网和一BP子网串联构成的,该网络同时具备BP神经网络能较好地预测未知样本的能力以及RBF神经网络的逼近速度快的优点。并以某900HC可逆冷轧机板形识别为应用背景,在MATLAB2010a环境下进行仿真实验,结果表明混合优化RBF-BP组合神经网络的板形模式识别方法能够识别出常见的板形缺陷,提高了板形缺陷识别精度并具有较好的鲁棒性,可以满足板带轧机高精度的板形控制要求。

Flatness Defect Recognition Based on RBF-BP Network by Hybrid Optimization

The problem of the traditional flatness pattern recognition method is low accuracy and weak robustness, flatness pattern recognition method based on RBF-BP network by hybrid optimization is proposed. Firstly, self-organizing map network(SOM) is used to cluster samples, and the network topology structure after clustering is used to determine the center of RBF, and the width of RBF is calculated, which overcomes the problem of unstable clustering results caused by random selection of centers in traditional clustering algorithm. Then the weight of the whole network is optimized by using the fine global search ability of genetic algorithm(GA). RBF-BP combined neural network is composed of a RBF sub-network and a BP sub-network in series. It has the ability of BP neural network to predict unknown samples and the advantage of fast approximation speed of RBF neural network. Taking the shape recognition of a 900 HC reversible cold rolling mill as the application background, the simulation experiment was carried out under the environment of MATLAB 2010 a. The results show that the method of shape pattern recognition based on hybrid optimization RBF-BP neural network can recognize common shape defects, improve the accuracy of flatness defect recognition and possesses good robustness, which can meet the requirements of high precision flatness control of strip mill.