Parameter optimization strategy of MMC capacitor voltage sliding mode observer based on BP neural network

Submodule capacitor voltages of Modular Multilevel Converter (MMC) can realize self-balancing, which is based on the full-rank switching matrix principle. The self-balancing does not need voltage sensors. However, the effectiveness of votlage balancing is highly dependent on the reliability of the trigger pulses. Trigger pulse errors can lead to unbalanced submodule capacitor voltages. However, due to the absence of sensors, the capacitor voltage is in an unknown state. In order to monitor the capacitor voltage in real time, this paper adopts a capacitor voltage strategy based on Sliding Mode Observer (SMO) method combined with BP neural network. The main purpose of the sliding mode observer is to observe capacitor voltage, while the BP neural network optimizes the parameters in the sliding mode observer to make the observation accuracy more accurate. This scheme estimates the capacitor voltages by acquiring the arm currents and switching signals and utilizing the sliding mode observer. The BP neural network adaptively adjusts the sliding mode gain to achieve precise observation of the capacitor voltages. Finally, an MMC system is built on the MATLAB/Simulink platform to verify the feasibility of the proposed BP neural network-based SMO strategy for MMC capacitor voltage observation.