Weighting factor design based on Artificial Neural Network for Finite Set MPC operated 3L-NPC converter

Optimum design of the weighting factors for a multi-objective cost function is one of the major challenges of Finite-Set Model Predictive Control (FS-MPC) operated power electronic converters. Especially for multi-level topologies, where multi-objectives must be included in the cost function to ensure a safe operation of the converter, the complexity of the optimization problem is rapidly growing with each new objective included in the cost function. In this paper a new approach for design of the weighting factors for a three level neutral point clamped (NPC) converter using artificial neural network (ANN) is proposed. The ANN is used as a surrogate model of the detailed converter model. In the first step a detailed converter model is simulated for different weighting factor combinations. From the simulations obtained performance metrics (e.g. total harmonic distortion (THD), average switching frequency, DC-link voltage balance) are used to train the ANN. Once the network is trained, it can be used to estimate the performance metrics for any combination of weighting factors. By defining a fitness function using the metrics, weighting factor combinations that optimize the function are found to be very fast. The design is also validated on an experimental set-up, where the measured performance metrics are compared to the ones predicted by the ANN. It is concluded that the results match very well with a difference being below 10%.