基于支持向量机的P91钢蠕变-疲劳寿命预测  

Life Prediction of Creep-fatigue Interaction of P91 Steel Based on SVM

作　　者：纪冬梅[1,2] 轩福贞[1] 涂善东[1] 姚秀平[2] 

机构地区：[1]华东理工大学机械与动力工程学院,上海200237 [2]上海电力学院能源与环境工程学院,上海200237

出　　处：《压力容器》2011年第10期15-21,8,共8页Pressure Vessel Technology

摘　　要：以相关文献中关于P91钢母材和焊材蠕变-疲劳试验数据为样本数据,利用支持向量机方法,以保载时间为模型的输入特征参数,蠕变寿命或疲劳寿命为模型的输出特征参数,通过训练试验数据建立P91钢寿命预测模型,并用部分数据验证模型的预测能力。同时研究了不敏感系数对于模型推广能力的影响。结果表明,SVM方法可以用于预测P91钢的蠕变-疲劳寿命,且不敏感系数越小,训练样本数据时误差较小,但是模型泛化能力较弱;不敏感系数越大,训练样本数据时误差较大,但是模型泛化能力较强。Based on the creep - fatigue test data of P91 base metal and welding quoted from related literatures, SVM （ Supported Vector Machine） was used to train some of these data with load holding time regarded as input characteristic parameter and creep life or fatigue life regarded as output characteristic parameter. A model can be obtained by training some test data, and other test data was used to verify the model. At the same time the selection of the insensitive parameter was studied. The results show SVM can be used to establish the model to predict the creep-fatigue life of Pgl steel, and the insensitive parameter can influence the prediction ability of the model. When the insensitive is small, the training error is small, but the prediction ability is weak. When the insensitive is large, the training error is large too, but the prediction ability is strong.

关 键 词：P91钢 支持向量机 蠕变-疲劳 寿命预测 

分 类 号：O344.6[理学—固体力学] TG142[理学—力学]