基于支持向量机的烧结能耗及性能指标预测模型  

SVR-based Predictive Models of Energy Consumption and Performance Criteria for Sintering

作　　者：王俊凯[1] 乔非[1] 祝军[1] 倪嘉呈 

机构地区：[1]同济大学电子与信息工程学院,上海201804

出　　处：《同济大学学报（自然科学版）》2014年第8期1256-1260,共5页Journal of Tongji University:Natural Science

基　　金：国家自然科学基金(61273046;61034004);安徽省钢铁产业技术创新规划研究(09020203014)

摘　　要：针对烧结过程中能耗和性能指标预测方法精度不高、训练时间长的问题,首先,在总结当前预测建模方法的基础上,将回归型支持向量机(support vector machine for regression,SVR)引入烧结生产系统,分析了2种建模模式;然后,给出基于SVR预测建模一般流程;最后,以某大型钢铁企业为例进行验证,并与传统的多元线性回归、反向传播(back propagation,BP)神经网络、径向基函数(radical basis function,RBF)网络和极限学习机(extreme learning machine,ELM)等预测方法在相同模式内和不同模式间进行比较.结果表明,SVR方法可快速获得理想的预测结果,在预测精度和时间效率上具有优势.As to the unsatisfactory accuracy and long training time in current predictive methods of energy consumption and other performance criteria for sintering process,firstly,based on summary of existing predictive methods,support vector machine for regression （SVR） was introduced into sintering production system,and two modeling modes were proposed.Then,the general procedures of predictive modeling based on SVR were given.After that,the proposed method was verified in a scenario derived from a large-scale iron and steel enterprise,compared with other predictive methods like traditional multiple linear regression,BP neural network,RBF network and extreme learning machine within the same mode and between different modes.The result shows that SVR method can achieve satisfied predictive results rapidly,which have advantages in prediction accuracy and time efficiency over other methods.

关 键 词：烧结 能耗 性能指标 预测模型 回归型支持向量机 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]