多核自适应迭代LSSVR的模拟电路性能评价策略

Strategy for evaluation of analog circuit performance based on adaptive and iterative LSSVR with multi RBF kernel tuning

针对模拟电路的性能评价问题,运用标准支持向量回归机,结合自适应技术的优越性,利用迭代算法改善传统最小二乘支持向量回归的支持向量稀疏性问题,提高训练响应速度。同时采用多径向基核函数以实现核宽度在线调整的灵活度,进一步提高支持向量数目确定的精简性。给出了基于多核自适应迭代最小二乘支持向量回归法的设计思想及构造步骤。实验以高校模拟电路实验为依托,采用近两年内由精密仪器设备测评所得的小功率放大器的8项技术指标构建训练集,进行多核自适应迭代最小二乘支持向量回归评价。实验表明,所提出的方法性能优于传统最小二乘支持向量回归法及ε-SVR法,与精密仪器性能评价结果较为接近,且运算速度优。

Aiming at the issue of analog circuit performance evaluation, a novel strategy by combining the superiority of adaptive technique with support vector regression machine is proposed in this paper. The iterative algorithm is used to adaptively decide the number of support vectors, the sparse property of support vectors is preserved, and the training speed is improved further. Regarding to reduce the number of support vectors, simultaneously, multi RBF is used to interfuse more flexibility to the kernel in line such as the bandwidths. The design idea and the construction steps based on adaptive iterative least square support vector regression（AILSSVR） with multi RBF kernel tuning （MK） are introduced. The ex- periment is supported by the college analog electronic experiments, eight indexes of low power amplifier are used, the evaluation of least square support vector regression with multi kernel is carded out in two years. The experiment indicates that the evaluation performance and the proposed testing speed are superior to that of the traditional LSSVR and ε-SVR.