Conventional model predictive direct speed control (MP-DSC) has received extensive attention because of its easy implementation, fast dynamic response, and simple structure. However, weighting factor and parameter mismatch are two key factors restricting the control performance of MP-DSC. In this article, a model-free hybrid parallel predictive speed control (MF-HPPSC) based on ultralocal model is proposed to solve the above-mentioned problems. First, the multiobjective parallel predictive speed control method realizes synchronous control of speed, torque, and flux linkage without weighting factor. The original excellent dynamic response capability is guaranteed, and the steady-state performance is simultaneously enhanced. Moreover, the linear extended state observer (LESO) designed by the unique ultralocal model makes the calculation process of future state prediction without any machine parameters, and the robustness of the system is significantly improved. Finally, the simulation and experiment comparison among the conventional MP-DSC, conventional model-free control, and the proposed MF-HPPSC verified the superiority of the proposed method.
The finite control set model predictive torque control (FCS-MPTC) selects the optimal voltage vector (VV) by the composite cost function composed of torque and flux error, which makes it have a faster dynamic response than conventional control methods. However, the prediction state error caused by machine parameter mismatch and the difficulty in setting the weight factor in the composite cost function seriously restrict the popularization and application of FCS-MPTC. In this paper, a model-free parallel predictive torque control (MF-PPTC) based on an ultra-local (UL) model is proposed to solve above problems. The UL model replaces the machine mathematical model without any machine parameters and only uses the input and output of the system, which greatly improves the robustness of the control system. The nonlinear extended state observer proposed for the unknown part of the system has fast convergence and improves the dynamic performance of the system. In addition, the conventional parallel predictive control structure is optimized to reduce the dynamic adjustment process during the selection of optimal voltage vector. Simulation and experimental comparison between the conventional PPTC and the proposed MF-PPTC verified the superiority of the proposed method.
Model predictive torque control with duty cycle control (MPTC-DCC) is widely used in motor drive systems because of its low torque ripple and good steady-state performance. However, the selection of the optimal voltage vector and the calculation of the duration are extremely dependent on the accuracy of the motor parameters. In view of this situation, A modified MPTC-DCC is proposed in this paper. According to the variation of error between the measured value and the predicted value, the motor parameters are calculated in real-time. Meanwhile, Model reference adaptive control (MRAC) is adopted in the speed loop to eliminate the disturbance caused by the ripple of real-time update parameters, through which the disturbance caused by parameter mismatch is suppressed effectively. The simulation and experiment are carried out on MATLAB / Simulink software and dSPACE experimental platform, which corroborate the principle analysis and the correctness of the method.
With the rapid development of lithium-ion batteries, predicting battery life is critical to the safe operation of devices such as electric ships, electric vehicles, and energy storage systems. Given the complexity of the internal aging mechanism of batteries, their aging process exhibits prominent nonlinear characteristics. Knee point, as a distinctive sign of this nonlinear aging process, plays a crucial role in predicting the battery’s lifetime. In this paper, the cycle life and cycle to the knee point of the battery are firstly predicted using the time dimension and space dimension features of the early external characteristics of the battery, respectively. Then, to capture the aging characteristics of batteries more comprehensively, we innovatively propose a joint prediction method of battery cycle life and knee point. Knee point features are incorporated into the battery cycle life prediction model in this method to fully account for the nonlinear aging characteristics of batteries. The experimental validation results show that the TECAN model, which combines time series features and knee point information, performs well, with a root mean square error (RMSE) of 106 cycles and a mean absolute percentage error (MAPE) of only 12%.