文献类型：期刊文献

中文题名：基于聚类算法与树模型的弹簧钢脱碳质量分析预测方法

英文题名：Analysis and Prediction Method for Decarburization Quality of the Spring Steels Based on Clustering Algorithm and Tree Model

作者：唐思瑜[1];吕立华[2];孙丽华[1];姜庆超[1]

机构：[1]华东理工大学能源化工过程智能制造教育部重点实验室;[2]宝山钢铁股份有限公司中央研究院智能所

年份：2025

卷号：52

期号：3

起止页码：392

中文期刊名：化工自动化及仪表

外文期刊名：Control and Instruments in Chemical Industry

基金：国家自然科学基金优秀青年基金(批准号:62322309)资助的课题。

语种：中文

中文关键词：K近邻;孤立森林;随机森林;弹簧钢脱碳;互信息

外文关键词：K-nearest neighbors;isolation forest;random forest;spring steel decarburization;mutual information

摘要：弹簧钢的脱碳质量对成品钢的性能至关重要,因此提出了一种基于聚类算法与树模型的弹簧钢脱碳质量分析预测方法。首先,采用K近邻(KNN)算法解决数据稀疏性问题;其次,采用孤立森林(IF)算法应对数据离散度高的问题;接着,采用互信息与随机森林结合的混合式特征选择算法对经过上述两种方法处理后的数据进行特征优选;最后,基于随机森林建立弹簧钢脱碳质量预测模型。通过某弹簧钢脱碳数据对所提方法进行验证,通过KNN与IF算法提升数据质量,基于混合式特征选择结果构建随机森林脱碳质量预测模型,平均准确率为93.02%,平均F1分数为87.87%,验证了所提方法的有效性与应用潜力。
The spring steels’decarburization quality influences the performance of finished steels much.In this paper,the method to analyze and predict decarburization quality of spring steels based on both clustering algorithm and tree model was proposed.In which,having the K-nearest neighbor algorithm used to solve data sparsity;and then,having the isolation forest algorithm adopted to solve high data dispersion,including having the hybrid feature selection algorithm which combining mutual information and random forest employed to perform optimal feature selection on the data processed by above-mentioned two methods;and finally,having the random forest based to establish the spring steel’s decarburization quality prediction model.Through making use of the spring steel’s decarburization data,the proposed method was verified,including having the KNN and IF algorithms adopted to improve data quality.Having the hybrid feature selection results based to establish the random forest decarburization quality prediction model can bring about an average accuracy of 93.02%and the average F1 score of 87.87%,which verifies the effectiveness and application potential of the proposed method.