Addressing the problem of non-convergent parameter identification due to many unknown pa-rameters,a frame of iterative particle swarm optimization was proposed to identify the parameters in three iterative steps.Using experimentally measured motor temperature field data,with the mean square error between the estimated and measured temperatures at each node as the objective function,parallel optimi-zation was transformed into a three-step serial iterative optimization.The number of optimization variables were reduced at each step,the population size was shrunk,and falling into local optima was avoided.The frame was applied to a five-node lumped-parameter thermal network(LPTN)model of a motor with rated power at 70 kW.The identified parameters demonstrate the change of the ordinary thermal resist-ance and capacities with temperature,the change of the motor losses,the eddy-current coefficient of sta-tor resistance and air gap thermal resistance with the motor speed.The identification performance is eval-uated in terms of the maximum error and average error of temperature at the active-winding,end-winding,permanent magnet,stator tooth and stator yoke.Dynamometer experiments conducted under comprehen-sive operating conditions validate that,compared with experimental measurements and traditional lumped parameter models that adopt fixed parameter values,the proposed model gains higher accuracy and better adaptability to different operating conditions compared with the conventional LPTN model with fixed pa-rameters.
维普
万方数据