Under the background of the national energy strategy of "carbon peaking" and "carbon neutrality", conventional generators driven by fossil energy are gradually replaced by renewable energy units such as wind power and photovoltaics. The microgrid with the characteristics of development and extension can fully promote the large-scale access of such distributed power sources with strong randomness, thus achieving rapid development and construction. At the same time, the development of electric vehicles (EV) is a trend to ensure low-carbon energy. China also regards the development of electric vehicles as a strategic emerging industry. The development of microgrids has also prompted electric vehicles to be widely used in power grid shaving peaks and valleys, and curb power fluctuations. However, when large-scale electric vehicles are connected to the microgrid at the same time, it may also lead to the degradation of the power quality of the islanded microgrid, and even the instability of the entire microgrid. To this end, a multi-microgrid power generation control strategy with electric vehicles based on evolvable model predictive control (MPC) is proposed in this paper. Firstly, based on the multi-microgrid interconnection structure of controller interaction, considering the coupling relationship between generator terminal voltage regulation and system frequency control, a power generation control model with multiple microgrids with electric vehicles is established. Secondly, an adaptive algorithm of controller parameters based on MA-DDPG is designed: the frequency controller takes the real-time frequency offset and EV station output power boundary as the state set, and the adjustable parameter matrix Qx of the MPC controller as the action set, and the frequency deviation is used as the reward function index, and the voltage controller takes the real-time voltage as the state set, the proportional-integral coefficient of the PI controller as the action set, and the voltage offset as the reward function index; so as to realize the adaptive adjustment of the weight parameters of the MPC and the PI controller. Meanwhile, under the architecture of "centralized training and distributed execution", the intelligent agent group can realize the cooperative control between the sub-microgrids according to the real-time operating status information. The simulation results show that, the automatic voltage regulation loop increases the active power disturbance, which puts forward higher requirements for the load frequency controller. Under the load disturbance and wind power disturbance, the microgrid frequency control effect under the learning-based MPC controller is significantly better than that of the traditional controller. When various extreme faults occur in the system, the proposed controller can still control the frequency fluctuation of the microgrid within 0.01 Hz through coordinated control and parameter self-adaptation, the control excellence rate can still reach 100%, and the recovery time is still less than 1 s, the robustness of the multi-microgrid performance is significantly enhanced, and the performance is better than the traditional MPC controller in all aspects. In addition, when the machine learning controller fails, the proposed two-layer controller structure can still ensure that the frequency fluctuation of the microgrid is controlled within 0.01 Hz, and the control excellence rate can reach 100%, which is significantly better than the DDPG controller. The following conclusions can be drawn from the simulation analysis: (1) Compared with PID and fuzzy control, the evolvable MPC controller can transform the frequency control process into solving an optimization problem, and thus well adapt to the stochastic scene in the multi-microgrid system. (2) Compared with the traditional MPC, the DDPG agent can adjust the MPC and PI control parameters according to the real-time operating environment state, so as to better adapt to the complex working conditions where the system parameters and structure change with time. (3) Compared with the DDPG controller, the proposed double-layer protection structure has stronger security and stability. When the machine learning agent fails and cannot output actions normally, the MPC controller can use the preset parameters to complete the frequency control process until the machine learning controller returns to normal. © 2024 China Machine Press. All rights reserved.
