基于近端优化的永磁同步电机温度预测方法

A Temperature Prediction Method of PMSM Based on Proximal Optimization

为了精准有效地实现永磁同步电机的温度预测,提出了一种基于近端策略优化(PPO)算法和强化学习(RL)网络的永磁同步电机温度预测模型,即PPO-RL模型,利用PPO算法定义模型训练的损失目标函数,选择Nadam算法作为模型优化器,通过强化学习的Actor-Critic框架最小化损失目标函数,进而完成模型的迭代训练。采用Kaggle公开的永磁同步电机测量数据集进行试验,结果表明,与指数加权移动平均法、循环神经网络和长短期记忆网络相比,PPO-RL模型具有更高的预测精度和可靠性。

For accurate and effective temperature prediction of Permanent Magnet Synchronous Machines(PMSM),this paper proposes a temperature prediction model based on Proximal Policy Optimization(PPO)and Reinforcement Learning(RL)network,the so-called PPO-RL model.This model defines the loss objective function of model training by PPO algorithm,selects Nadam algorithm as an optimizer,and minimizes the objective function through the actor-critic framework of reinforcement learning to complete model iterative training.Experiments are conducted on the data set of PMSM published by Kaggle,the results show that compared with the exponential weighted moving average method,Recurrent Neural Network(RNN)and Long Short Term Memory(LSTM),the PPO-RL model has higher prediction accuracy and reliability.