Online Diagnosis of Synchronous Generator Rotor Winding Inter-turn Short-circuit Fault Using Nu-Support Vector Regression
Abstract
In order to more accurately and easily diagnose synchronous generator rotor winding inter-turn short-circuit fault, this paper puts forward a novel online diagnosis method based on nu-support vector regression (Nu-SVR). Firstly, the diagnosis method measures and collects sample data in different fault-free operating conditions, including terminal parameters and field current. Then, the diagnosis method uses these sample data to establish a Nu-SVR prediction model of field current, and the parameters of the Nu-SVR prediction model are optimized with particle swarm optimization (PSO) algorithm. Next, a predicted field current will be obtained if the measured terminal parameters are input to the Nu-SVR prediction model. Finally, the diagnosis method compares the measured field current with the corresponding predicted field current, and an inter-turn short-circuit fault is online diagnosed if a relative error does not fall in the specific threshold interval. The micro-synchronous generator dynamic simulation results showed that the accuracy and sensitivity of the Nu-SVR diagnosis method was the best in these former diagnosis methods such as the BP diagnosis method, the BRBP diagnosis method and the PSO-SVR diagnosis method. So the Nu-SVR diagnosis method is an effective online diagnosis method for the synchronous generator rotor windings inter-turn short-circuit fault.