摘要
针对大气环境下电网设备中金属材料的腐蚀速率预测问题,提出一种基于遗传算法(genetic algorithm,GA)优化Stacking集成学习算法模型,挖掘大气环境因子与镀锌钢腐蚀速率的关系。该模型为双层结构,融合了多个预测模型的优点。通过GA算法优化第一层各个初级学习器的待调参数,将初级学习器学习到的数据交给第二层次级学习器做进一步拟合。同时,结合K折交叉验证的方式有效降低过拟合现象。结合Spearman相关系数和随机森林特征重要性评估方法,筛选出与镀锌钢腐蚀速率相关性最高的5个环境因子作为输入,由此展开镀锌钢腐蚀速率预测研究。试验结果表明,相较于单一的机器学习模型,该模型能有效提高预测镀锌钢材料腐蚀速率的拟合度,降低预测误差。
In order to investigate the relationship between atmospheric environment factors and the corrosion rate of galvanized steel,a genetic algorithm(GA)based optimized Stacking integrated learning algorithm model is proposed for predicting the corrosion rate of metallic materials in grid equipment.The model,which has a two-layer structure,combines the benefits of many prediction methods.The GA method is used to tailor each primary learner's parameters in the first layer,and the secondary learner in the second layer receives the primary learner's data for further fitting.At the same time,K-fold cross-validation is combined with other techniques to successfully mitigate the overfitting phenomena.By combining Spearman's correlation coefficient and the random forest feature importance assessment method,the five environmental factors with the strongest correlation to the corrosion rate of galvanized steel were chosen as inputs,and the study of predicting the corrosion rate of galvanized steel was subsequently conducted.In comparison to a single machine learning model,the experimental findings demonstrate that the model can effectively enhance the fit and decrease the error in forecasting the corrosion rate of galvanized steel materials.
作者
田辉
樊志彬
王倩
米春旭
TIAN Hui;FAN Zhibin;WANG Qian;MI Chunxu(State Grid Shandong Electric Power Research Institute,Jinan 250003,China)
出处
《山东电力技术》
2023年第10期43-49,共7页
Shandong Electric Power
基金
国家电网有限公司科技项目“基于电网大气腐蚀图的数据挖掘及电网设备服役寿命评价技术研究”(5200-202016471A-0-0-00)。
关键词
大气腐蚀
环境因子
电网设备
机器学习
腐蚀速率
atmospheric corrosion
environmental factors
power grid equipment
machine learning
corrosion rate
