Physics-informed neural network-based control of power electronic converters

This chapter introduces a physics-informed neural network (PINN) for the control of grid grid-connected converters by fusing its underlying equations into the training process, thereby reducing the requirement for qualitative training data. First, we provide a comprehensive design guideline for the PINN to circumscribe differential equations and data-driven generalizations in complex systems. Moreover, we cover recent trends in scientific computing that involve the fusion of physics and data-based learning. In comparison with traditional data-driven methods, which either incur a significant computational burden or use overly conservative surrogate models, we explore the PINN’s easy optimization per the design requirements and find it to be significantly superior in terms of computation time and data requirements (trained using only 3001 set points), with an average prediction accuracy of 98.76%. As a result, PINN reveals a new modeling orientation for power electronic converters and is well suited for commercial applications. Finally, its robustness under various grid disturbances has been validated under simulation and experimental conditions.