Electric vehicles are a key technology to decarbonize the transport sector where interior permanent magnet synchronous motors (IPMSMs) are the best performer at the heart of the electrical drive system. In order to optimize their operational efficiency, the model-based method associated with parameter identification is widely adopted. However, efficiency optimization and parameter identification in the existing methods are implemented independently by different strategies in a sequential execution manner, which does not produce an optimized systemlevel solution. In this paper, the two methods are combined to deal with a constrained optimization problem in an IPMSM drive. Firstly, the problem is converted into a variational problem based on the variational principle and projection dynamic theory. Then, a unified projection dynamic equation (UPDE) is used to estimate the parameters and determine the solution of optimal current (OC) of the IPMSM. Further, a recursive neural network (RNN) corresponding to the UPDE is developed to implement the developed fast efficiency optimization of the IPMSM drive. The results of simulation experiments show the proposed method is effective to identify motor parameters and determine the OC of the drive system rapidly and accurately. Thus, it can rapidly realize efficiency optimization of an IPMSM drive-system. Because the designed RNN can be easily implemented in the hardware, such as a field-programmable gate array (FPGA) or dedicated neural network chip, the method can achieve instantaneous efficiency optimization of the IPMSM drive system and therefore improve the widespread application of IPMSMs in EVs.

电动汽车是交通部门脱碳的关键技术，其中内置永磁同步电机 (IPMSM) 是电力驱动系统核心的最佳表现。为了优化它们的运行效率，与参数识别相关的基于模型的方法被广泛采用。然而，现有方法中的效率优化和参数识别是通过不同的策略以顺序执行的方式独立实现的，并没有产生优化的系统级解决方案。在本文中，将这两种方法结合起来处理 IPMSM 驱动器中的约束优化问题。首先，根据变分原理和投影动力学理论将问题转化为变分问题。然后，采用统一投影动力学方程(UPDE)对IPMSM的参数进行估计并确定最优电流(OC)的解。此外，还开发了与 UPDE 对应的递归神经网络 (RNN)，以实现已开发的 IPMSM 驱动器的快速效率优化。仿真实验结果表明，该方法能够有效地快速准确地识别电机参数并确定驱动系统的OC。因此，它可以快速实现IPMSM驱动系统的效率优化。由于设计的RNN可以很容易地在硬件中实现，例如现场可编程门阵列（FPGA）或专用神经网络芯片，该方法可以实现IPMSM驱动系统的瞬时效率优化，从而提高IPMSMs在以下领域的广泛应用电动汽车。