Experimental design and data analysis and optimization of mechanical condition diagnosis for transformer sets

Abstract The typical power transformer diagnosis approach is imprecise and unstable. A support vector machine classification algorithm proposed, by designing an program that can improve the accuracy speed of energy diagnosis, vibration signals surface twisting in different states are extracted wavelet packet spectrum signal processing method, it verified curve similarity between simulation model measured data greater than 0.98, proving model’s validity. calculation technique online short circuit inductance developed from equivalent model, variation error results less 0.05% when compared to real characteristics. suggested state diagnostic successfully compensates for drawbacks reactance which incapable detecting judging slightly loose or faulty winding. method’s superiority, as well practicability system, demonstrated.