聚类加权PSO-LSSVR的模拟电路性能在线评价策略

针对模拟电路性能在线评价中现场数据采集不可避免产生错值和干扰的问题,运用模糊C-均值聚类（FCM）方法,根据数据特征将电路模型分为多个子模型,同时对子模型样本进行加权;利用标准支持向量机（LSSVR）,结合粒子群算法（PSO）对于参数优化的优越性,实现对于错值、干扰参数的有效抑制.考虑到传统的离线评价策略在训练样本组发生变化时,存在模型不能适时调整等问题,引入增减交互更新设计,实现模型的快速在线更新.实验以高效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传统的OTL放大电路的八项技术指标构建训练集,建立PSO-LSS-VR的组合模型进行在线评价.结果表明：该方法能有效处理错值和干扰所带来的回归偏差,性能优于传统LSSVR法、ε-SVR法,评价精度与精密仪器性能评价精度近乎相同,同时具有运算速度优的优势.仿真实验进一步验证了模拟电路性能评价方法PSO-LSSVR的有效性.

New PSO and LSSVM regression based on clustering weighted to evaluate analog circuit performance online

This paper features a novel PSO-LSSVR building on clustering weighted scheme as a via- ble alternative to the analogy circuit which suffers an inherent drawback in performance evaluation online, namely the inevitable production of the wrong values and interference due to on-site date collection. This alternative strategy involves firstly dividing the circuit model into multiple sub-models via fuzzy clustering mean （FCM） according to data features, and weighting the samples of the sub-models simultaneously; secondly, effectively inhibiting the wrong values and disturbance parameters by combining the norm LSS- VR with PSO superior in terms of parameters optimization ; and thirdly, introducing the incremental or re- duced learning interaction to update the model online in response to the traditional offline evaluation mod- el incapable of real-time adjustment to the changing samples. The strategy is validated by the experiment drawing on the college analog circuit experiments, the construction of the training set using the traditional OTL performance eight indexes, obtained via precision instrument evaluation in recent two years, and the realization of evaluation online by developing the PSO-LSSVR model. The results reveal that the proposed method PSO-LSSVR capable of effectively dealing with the regressive deviation caused by the wrong val- ues shows a performance superior to that of the traditional methods, such as LSSVR, and z-SVR, an e- valuation accuracy approximate to that of the precise instrument, and a better operation speed. The simu- lation verifies the reliability and validity of PSO-LSSVR evaluation method.