电网碳钢材料腐蚀预测模型研究  

Corrosion Prediction Model of Carbon Steel for Power Grid

作　　者：何成 游溢 王宗江 黄路遥 张强 陈云 卢壹梁 杨丙坤 王晓芳 HE Cheng;YOU Yi;WANG Zongjiang;HUANG Luyao;ZHANG Qiang;CHEN Yun;LU Yiliang;YANG Bingkun;WANG Xiaofang(Electric Power Science and Research Institute of State Grid Xinjiang Electric Power Co.,Ltd.,Urumqi 830011,China;China Electric Power Research Institute,Beijing 100192,China;State Grid Smart Grid Research Institute Co.,Ltd.,Beijing 102209,China)

机构地区：[1]国网新疆电力有限公司电力科学研究院,乌鲁木齐830011 [2]中国电力科学研究院有限公司,北京100192 [3]国网智能电网研究有限公司,北京102209

出　　处：《装备环境工程》2025年第2期151-159,共9页Equipment Environmental Engineering

基　　金：国网新疆电力有限公司科技项目(5230DK230016)。

摘　　要：目的研究新疆地区典型环境因素对典型电网用碳钢材料腐蚀的影响。方法根据采集到的新疆地区155组腐蚀及环境数据,通过支持向量回归(SVR)、梯度提升算法(GBoost)、皮尔逊相关系数(PCC)、逐点互信息(PMI)和随机森林(RF)等5种方式对输入变量进行特征选择,分析11种典型环境因素对典型电网用碳钢材料腐蚀速率的影响,选择重要性排序前5输入变量并集,进行Sobol敏感性分析。采用经主成分分析进行降维处理构建的SVR模型和利用梯度提升算法优化损失函数构建的GBoost模型预测腐蚀速率,研究2种模型的性能和预测能力。结果对新疆地区典型电网用碳钢材料碳钢大气腐蚀影响较为重要的前5个特征依次是年降水量、年均湿度、年均温差、PM_(10)和O_(3),GBoost模型相比于SVR模型在预测影响碳钢腐蚀的环境因素方面表现出较高的准确性和可靠性。结论GBoost模型具有更好地预测泛化能力和模型解释力,能够有效捕捉碳钢腐蚀速率与环境因素之间的复杂关系。The work aims to investigate the influence of typical environmental factors on the corrosion rate of carbon steel used in Xinjiang power grid.Based on 155 sets of corrosion and environmental data collected from Xinjiang,feature selection of input variables was accomplished through the application of five distinct methods,namely Support Vector Regression(SVR),Gradient Boosting algorithm(GBoost),Pearson Correlation Coefficient(PCC),Pointwise Mutual Information(PMI),and Ran-dom Forest(RF).The influence of 11 typical environmental factors on the corrosion rate of typical carbon steel used in power grids was analyzed.The top five input variables ranked by their significance,were chosen and amalgamated for the Sobol sensi-tivity examination.Principal component analysis(PCA)was conducted to construct a SVR model and GBoost was used to op-timize loss function to construct a GBoost model to predict the corrosion rate.The performance and prediction ability of the two models were studied.The result showed that the top five environmental factors are,in turn,annual precipitation,annual humidi-ty,annual temperature difference,PM_(10) and O_(3).Compared with the SVR model,the GBoost model shows higher accuracy and reliability in predicting the correction rate of carbon steel in Xinjiang.In conclusion,the GBoost model has better predictive generalization ability and model explanatory ability,and can effectively capture the complex relationship between the corrosion rate of carbon steel and environmental factors.

关 键 词：电网 碳钢 大气腐蚀 机器学习 寿命预测 预测精度 

分 类 号：TG172[金属学及工艺—金属表面处理]