摘要
在传统神经网络(BP)模型的基础上,利用广义回归神经网络(GRNN)对处于海洋环境下的金属腐蚀速率进行了预测,将环境温度、含氧量、pH值、盐度及电位作为输入,其腐蚀速率作为输出,并与实际腐蚀速率进行比较。结果表明:采用GRNN预测时,选取默认扩展速度值,其预测平均误差为5.72%,高于采用BP神经网络预测时的6.56%,采用交叉验证方法选取最优扩展速度值,最优扩展速度值下其预测的平均误差为2.38%,说明采用GRNN对海洋环境下的金属材料腐蚀速率进行预测在技术上可行,并具有较高的预测精度,对全面了解海洋金属结构物的运行状态及腐蚀情况有重要意义。
On the basis of the traditional neural network model, the paper predicted the corrosion rate of metal under the marine environment by using the generalized regression neural network （GRNN）. Environmental temperature, oxygen content, pH value, salinity and potential were used as input, and the corrosion rate was the output, which was compared with the actual corrosion rate. Results showed that the default S value was selected when the generalized regression neural network was employed. The average prediction error was 5.72%, higher than that of the baek propagation （BP） neural network. In addition, the optimal S value was selected by cross validation, and the average error of the prediction was 2.38 % under the optimal S value. Therefore, it was technically feasible to predict the corrosion rate of metal materials under the marine environment with the generalized regression neural network, and the system had a high prediction accuracy, which was of great significance for the comprehensive understanding of the operation state and corrosion of the marine metal structures.
作者
马良涛
董海防
朱刚
李进
MA Liang- tao, DONG Hal- fang, ZHU Gang, LI Jin(Wuhan Second Institute of Ship Design and Research, Wuhan 430000, China)
出处
《材料保护》
CAS
CSCD
北大核心
2018年第9期35-39,共5页
Materials Protection
关键词
广义回归神经网络
BP
金属腐蚀速率
最优扩展速度
预测
交叉验证
generalized regression neural network
back propagation
metal corrosion rate
optimal S value
forecast
cross validation
