基于ELAD和RNN的电动车用电动机运行效率快速优化关键技术研究

The motor that drives the electric vehicle(EV) should has high efficiency during operation , which is very important for save production and operation cost of the EV, and for energy saving and cut emissions in transportation field. The steady-state time of motor is short and its parameters are easily changed when the EV is running. Because of that fact, for optimizing efficiency of motor of the EV, it is an essential principle that the real-time efficiency optimization method should combine with accurate parameters estimation technology, and the method must has faster optimization speed. For this goal, firstly, we model the loss model of the motor in this project, which is relevant to current of motor, and analyze and verify its correctness by simulation and experiment. Secondly, we propose enhanced least absolute deviation method (ELAD) that has good robustness to measurement noise, and analyze and proof its robustness, then we try to estimate motor parameters through it. Thirdly, we will derive the most simple expressions of the ELAD according to optimization theory, the expressions should be easy to implement and compute when estimating motor parameters, and we try to propose a recurrent neural networks(RNN) algorithm that is very easy to implement by electronic circuit to implement the ELAD method, and we will realize it by using FPGA. Finally, we model real-time efficiency optimization of motor as a optimization problem with constraint conditions, and derive the equations satisfied by optimization current (d/q axis current) by Karush- Kuhn-Tucker condition, and study a method solving the equations by the RNN.

电动车用电动机的高效运行，对降低电动车生产和运营成本以及实现交通领域节能减排具有非常重要的意义。电动车用电动机运行时具用稳态过程时间短和参数易变化的特点，对其进行实时效率优化，优化方法必须能实时精确地估计电动机相关参数和具有快的优化速度。为了达到该目的，本项目首先建模电动机的电可控损耗模型，仿真和实验验证其正确性；其次提出对测量噪声具有强鲁棒能力的增强型最小绝对偏差方法（ELAD），分析证明其鲁棒性，并用该方法估计电动机的参数；接着利用优化理论推导ELAD方法估计电动机参数的易于计算和实现的表达式，将其转化为易于用递归神经网络（RNN）求解的一阶微分方程组，并以FPGA为核心芯片开发相应的硬件求解器；最后建模电动机的实时效率优化为基于约束的优化问题，利用Karush-Kuhn-Tucker条件导出实现效率优化的d/q轴电流所满足的方程组，并研究用RNN求解该方程组的方法。

将增强的最小绝对偏差方法（ELAD）、优化理论和硬件实现的递归神经网络（RNN）相结合，以此对车载电动机进行快速和精确的效率优化，对车载电驱动系统的实时高效运行，特别是对电动车的节能减排和推广使用具有十分重要的意义，该技术有很好的应用前景。课题组围绕项目计划书的研究内容展开了较为全面深入的研究工作，取得了较好的研究成果。在综台分析内装式永磁同步电动机(IPMSM)）的各种损耗，建立电动机的电可控损耗模型；利用IPMSM的离散数学模型，推导出用于估计电机参数的回归方程；分析各种参数估计原理，并用MATLAB编程实现各种相应方法的参数估计，对比分析各种方法的优缺点；深入研究基于优化理论和ELAD方法电机参数辨识原理，并利用MATLAB实现的仿真；深八研究基于优化理论和投影动态理论的IPMSM优化电流的实时求解方法，和在FPGA设计用于实现IPMSM优化电流求解和电机参数辨识的递归神经网络，仿真实现了电流的求解和参数辨识；建立基RT-LAB实验设备的电动汽模拟电驱动系统。根据项目任务书，课题组精心安排研究成员，创造性地开展了工作，取得了较好的研究成果和人才培养工作。四年来，发表或录用直接相关科研文章共11篇,sci检索2篇，EI检索6篇。主要发表在Applied Energy，Energies，华中科技大学学报（自然科学版）、四川大学学院（自然科学版）、The World Congress on Intelligent Control and Automation,The IEEE International Conference on Signal Processing,Communications and Computing等同内外重要核心期刊和国际会议上。申请或授权专利1项。培养青年教师3名，硕士研究4名，其中安林杰获国家励志奖学金。按照项同计划书的研究计划要点，本项曰的研究内容均己完成，基本达到预期曰标，后续研究工作仍将进行。