摘要
针对带钢弯曲质量预测时因样本的误判率不平衡而导致质量预测误差较大、异常检出率较低的问题,引入了决策函数中的少数类样本支持向量所占的权重系数μ,建立了基于改进权重系数的加权支持向量机(μ+SVM)的热轧带钢弯曲质量预测分析模型。结果表明:通过比较其与随机采样+支持向量机方法、人工合成数据+支持向量机方法的预测结果,验证了μ+SVM支持向量机方法的优越性。采用支持向量机方法得到的最小误判率为0.32,标准差为0.38,更加符合实际检测需求。
Aiming at the problems of big quality prediction error and low abnormal detection rate due to the imbalance of sample misjudgment rate in the prediction of bending quality of steel strip, the weight coefficient μ of minority sample support vector in decision function was introduced, and the prediction and analysis model of bending quality of hot rolled steel strip based on weighted support vector machine(μ+SVM) with improved weight coefficient was established. The results show that the superiority of μ+SVM support vector machine method is verified by comparing the prediction results with that of random sampling+support vector machine method and artificial synthetic data +support vector machine method. The minimum error rate of 0.32 and standard deviation of 0.38 are obtained by using support vector machine, which is more in line with actual testing requirements.
作者
闵建
MIN Jian(Department of Mechanical Engineering,Taizhou Technician College,Taizhou 225300,China)
出处
《热加工工艺》
CSCD
北大核心
2018年第23期165-167,共3页
Hot Working Technology
关键词
质量预测
弯曲性能
支持向量机
权重系数
quality prediction
bending performance
support vector machine (SVM)
weight coefficient