摘要
应用相关向量机对珊瑚混凝土小样本数据进行回归和预测,与BP神经网络和最小二乘支持向量机的比较发现:相关向量机具有良好的泛化能力,避免了神经网络易出现的过拟合现象,不仅预测准确率不低于最小二乘支持向量机,而且能输出预测值的置信区间;利用相关向量机回归构建解释体系,分析配合比中各组分的重要性、单因子效应及两因素间交互作用,仿真结果与实际结果一致。相关向量机是混凝土性能预测的可靠工具,可为配合比设计和优化提供指导。
Relevance vector machine is introduced to regress and predict the compressive strength of coral concrete on a small sample, comparing with back propagation of neural network and least square support vector machine,it is found that relevance vector machine owns benign generalization ability so that over fitting which occurs in neural network could be avoided,not only the prediction accuracy of relevance vector machine is no worse than least square support vector machine,but also it can provide the confidence interval of predicted value;single-factor importance,single-factor effects and two-factor interaction are analyzed through the using of explanatory system estab-lished by relevance vector regression,results of simulation tests are coincide with that of actual test.It is indicated that relevance vector ma-chine is reliable tool for predicting performance of concrete,it can provide a guidance to the design and optimization of mixture.
出处
《混凝土》
CAS
北大核心
2016年第7期1-6,共6页
Concrete
基金
国家自然科学基金资助项目(41472287)
关键词
相关向量机
珊瑚混凝土
抗压强度
置信区间
解释体系
relevance vector machine
coral concrete
compressive strength
confidence interval
explanatory system
作者简介
李仲欣（1989-），男，在读博士研究生，研究方向为土木工程材料。联系地址：天津市塘沽区河北路一号（300450）联系电话：15122801736
