模拟退火优化SVM参数的变压器故障诊断

利用群智能算法优化支持向量机(SVM)参数往往需要引入额外的变量，使变压器故障诊断问题更加复杂，对此提出一种自适应的模拟退火算法优化(ASA)支持向量机参数。通过设计自适应的冷却进度表，使得寻参的过程仅仅依赖于退火速率以及网格搜索粒度，保持了较少的参数设置。在相关数据集上的实验表明，与已提出的粒子群算法(PSO)、遗传算法(GA)相比，ASA算法具有更快的收敛速度以及较好的诊断精度。利用自适应的模拟退火算法能够较好的优化SVM参数并提高变压器故障诊断的精度。

Optimized Parameters by Simulated Annealing of SVM in Power Transformer Diagnosis

Optimizing the parameters of Supported Vector Machine generally introduces extra variables, causing the original problem more complicated. Hence, a Self-adaptive Simulated Annealing Algorithm is proposed. By means of designing a self-adaptive annealing schedules, the course of searching the best parameters only depends on two variables, annealing rate and grid search granularity, keeping relatively fewer parameters setting. As with the experiment shown on related datasets, ASA algorithm acquires the faster speed of convergence and decent diagnosis accuracy. Utilizing ASA acquires better parameters of SVM ,while improving the accuracy of transformer diagnosis.