An Online Neural Network Approximator-Based Model-Free Predictive Control Approach for Power Converters

Recently, model predictive control is widely applied in practice. To enhance the robustness of the controlled system, model-free predictive control is singled out as an alternative solution. However, conventional model-free predictive control method based on ultralocal model still relies on inductance parameters during the design process, and capacitance parameters are also necessary for neutral point voltage balancing. The dependence on parameters leads to a reduction in the robustness of the control method. To address this issue, a novel model-free predictive control approach based on an online neural network approximator is proposed, which combines neural network predictor with adaptive filtering theory. Meanwhile, the neutral point voltage prediction is achieved through the implementation of an adaptive solution. Notably, this method is completely independent of system parameters. Hence, in the face of changes in system operating conditions, the input gain can be adaptively adjusted. Furthermore, the stability of the system is proven. Finally, experimental results demonstrate the superior performance of the proposed method in power converter control.