Model Predictive Direct Torque Control of Permanent Magnet Synchronous Motors

目前，嵌入式磁体永磁同步电机（IPMSM）由于其高功率、高转矩密度和高效率，主要用于电动和混合动力汽车。为了充分利用电力驱动的性能潜力，适当的驱动控制是必不可少的。IPMSM的传统驱动控制目前是基于PI控制器的。相比之下，模型预测控制器（MPC）具有明显更好的性能潜力：MPC利用待控制系统的过程模型。基于该模型，在每个采样步骤中确定最优的未来控制变量。然而，这种在线优化对计算硬件的要求很高。由于创新的方法和现在可用的更强大的计算硬件，MPC已经可以用于IPMSM，但这些方法会导致其他问题，如大电流和转矩波动。该项目的目的是开发一种MPC方法，该方法是专门为控制汽车传动系统中的IPMSM而设计的。因此，要开发的MPC方法是基于直接MPC的。与传统的MPC相比，在线优化过程中只考虑逆变器的可行电压矢量（根据可能的开关状态）。因此，优化的执行非常节省资源，因此实时MPC可以在廉价的计算硬件上实现。由于计算可以很容易地并行化，fpga或多核架构适合实现。MPC将作为直接转矩模型预测控制实现，其中工作点选择直接包含在MPC的固有优化算法中。这种操作点的优化特别有助于减少损失。为了分别减少电流和转矩的波动，对脉冲模式生成进行了重新设计：它可以根据目标函数值对矢量调制进行类似的优化。这导致电流纹波的减少，尽管一个相对较大的控制器周期时间，因此计算工作可以显着减少。通过对目标函数进行自优化加权，优化了运行过程中各工作点的转矩动态。

Currently, permanent magnet synchronous motors with embedded magnets (IPMSM) are mainly used for electric and hybrid vehicles due to their high power and torque densities as well as their high efficiency. In order to utilize the full performance potential of an electric drive, an adequate drive control is essential. Conventional drive controls for IPMSM are currently based on PI controllers. In comparison, Model Predictive Controllers (MPC) have the potential of a significantly better performance: A MPC exploits a process model of the system to be controlled. Based on this model optimal future control variables are determined in every sampling step. However, this online optimization sets high demands on the computation hardware. Due to innovative approaches and nowadays available more powerful computing hardware, MPC can already be used for IPMSM, but these approaches lead to other problems like high current and torque ripples.The aim of this project is to develop an MPC approach which is specifically designed for the control of IPMSM in the drivetrain of automobiles. Therefore the MPC approach to be developed is based on Direct MPC. In contrast to classical MPC, only the feasible voltage vectors of the inverter (according to possible switching states) are taken into account during the online optimization. Thus, the optimization is performed very resource-efficient, so real-time MPC can be implemented on inexpensive computation hardware. Since the calculations can be easily parallelized, FPGAs or a multi-core architecture are qualified for the implementation. The MPC will be implemented as a Direct Torque Model Predictive Control whereby the operating point selection is directly included in the inherent optimization algorithm of the MPC. This operating point optimization contributes inter alia to minimize losses. To reduce the ripples of the current and torque, respectively, the pulse pattern generation is redesigned: It can be optimized similarly to vector modulation, depending on the objective function values. This leads to a reduction of the current ripple in spite of a relatively large controller cycle time so that the computing effort can be significantly reduced thereby. By using a self-optimization for weighting the objective function, the torque dynamics for each operating point is optimized during runtime.