组合人工神经网络模型预测海水腐蚀速度的研究

PREDICTION OF MARINE CORROSION USING A COMBINED ARTIFICIAL NEURAL NETWORK MODEL

针对误差反传(BP)人工神经网络在海水腐蚀预测建模中的样本数量有限和隐含层单元数难以确定的问题,提出了结合使用自组织特征映射(SOM)网络和径向基函数(RBF)人工神经网络方法预测海水腐蚀速度.首先采用SOM对样本进行分类,再使用RBF进行样本训练和建模,并对A3钢和16Mn钢的海水腐蚀速度进行了预测.结果表明,该方法能够增强了网络局部泛化能力,提高了预测精度和计算速度.

Considering the limitation of the number of samples and the difficulty of the selection of the suitable number in hidden layer cells during constructing a BP prediction model of marine corrosion system, a combined ANN method with SOM （self-organizing feature map） and RBF （radial base function） was proposed to predict marine corrosion rate. By this method, the training and predicting were divided into two steps： clustering the samples at first, and then training and predicting them at the clustered areas, respectively, by SOM and RBF neural network. The prediction results of the marine corrosion of A3 and 16Mn steel showed that this method could improve the ability of local generalization of the network ,which made the prediction much better than the normal BP network in terms of precision and speed.