一种用于支持向量回归的动态工作集选择方法

A Method of Dynamic Working Set Selection for Support Vector Regression

在航天器控制计算机的软硬件协同设计过程中,功耗建模是一个必不可少的步骤.基于工作集选择的支持向量回归方法是将功耗经验数据训练为功耗模型的有效手段.现有的工作集选择方法利用的是固定的一阶或二阶信息,没有考虑回归参数的影响,导致收敛时间较长.针对这一不足,提出了根据回归参数调整工作集选择策略的动态工作集选择方法 DWSS,减少了算法收敛的迭代次数.在数值试验部分,对此方法进行了验证,结果表明新的方法具有更快的收敛速度.

In the Hardware/Software Co-design process of the control computer on a spacecraft,power modeling is an essential step.The Support Vector Regression（SVR）algorithm based on working set selection（WSS）is an effective method to resolve this problem.Existing WSS methods using fixed first order or second order information regardless of the influence of regression parameters lead to a long iterative time.To break through the limit of these methods,in this paper,a new WSS method named DWSS is proposed.Numerical experiments show that DWSS is faster than existing methods.