Research On Temperature Prediction Of Permanent Magnet Synchronous Motor In Electric Vehicles

With the development of the automobile industry and people's awareness of environmental protection,new energy vehicles account for an increasing proportion of people's travel plans.As the main driving core of electric vehicles and hybrid electric vehicles,permanent magnet synchronous motor(PMSM)is often faced with such dangers as demagnetization of permanent magnet and circuit damage when operating temperature is too high.It may even cause the driving ability of new energy vehicles to decline or even lose driving power.In order to ensure the permanent magnet synchronous motor could be safe and stable operation under complicated working conditions,new energy vehicles of permanent magnet synchronous motor temperature prediction problems studied by many scholars,gradually so as to timely access to the real-time temperature of the motor,and take the corresponding cooling method to guarantee the safety of the permanent magnet synchronous motor,reduce the maintenance cost of the machine.Therefore,the study of temperature prediction of permanent magnet synchronous motor has important research significance in the field of driving protection of new energy vehicles.Aiming at the temperature prediction problem of permanent magnet synchronous motor,the additional feature addition and data sampling processing are firstly completed based on the benchmark data,so as to obtain the sampling data set with a smaller data scale than the datum data.Based on the sampled data set,the prediction results of EMWA,RNN and LSTM networks are reproduced,and to solve the problem that the prediction accuracy of the above three prediction methods is not high,a temperature prediction framework of improved CNN-LSTM network is proposed.The experimental results show that the structure can achieve better prediction performance.Secondly,a prediction model of motor temperature based on PPO-RL is proposed.This model combines the proximal strategy optimization algorithm and the actor-critic network with enhanced learning.Based on the sampled data set,the comparison experiment of the model's prediction of motor temperature is carried out,and finally the prediction effectiveness of the model is verified.Finally,aiming at the problem of prediction deviation in the PPO-RL prediction model,a method of temperature prediction of PPO-RL motor with integrated feature optimization is proposed.This method is used to optimize the temperature sampling data set of PMSM through correlation analysis,and the prediction performance of this method is verified based on the data set after characteristic optimization.The final results show that this method has lower prediction error and better temperature fitting effect,which has better research significance and application value.