Artificial neural network controller based on model predictive control

The use of modern control techniques in power converters, such as model predictive control (MPC), has increased in recent years because of its intuitive algorithm and easy implementation. However, when the complexity of the converter topology increases, the computational cost also increases, limiting the switching frequency operation and compromising the high-power density of the converter. To address this challenge, artificial intelligence-based controllers have been introduced, specifically artificial neural networks (ANNs), which are able to learn from the behavior determined by controllers such as the MPC. In this paper, a procedure for building an ANN that has the same features as a MPC but without its disadvantages is proposed, such that the ANN can substitute the corresponding MPC as a controller. To facilitate this explanation, the proposed procedure will be described using a Buck DC-DC converter. However, other DC-DC converters could be used as well. The proposed procedure can be divided into four steps. First, the Buck converter operates under MPC action under different conditions (in simulation); at the same time, the information is collected in a database. This information is used in the second step for the ANN off-line training; once the ANN is fine-tuned according to the desired behavior determined by the MPC, the ANN is used to control the switching devices in the third step. To validate this, a simulation is performed with the Buck converter under ANN action. Finally, the ANN is implemented in a microcontroller and used to regulate the output voltage in a Buck prototype.