Explicit model predictive control (EMPC) maps offline the control laws as a set of regions as a function of bounded uncertain parameters using multi-parametric programming. Then, in online mode, it seeks the best solution within these areas. Unfortunately, the offline solution can be computationally demanding because the number of regions can grow exponentially. Thus, this paper presents the application of a deep neural network (DNN) to learn the EMPC's regions for a photovoltaic-based charging station. The main uncertain parameters in this study are the forecast error of photovoltaic power production and the battery's state of charge. Additionally, the connection or disconnection of an electric vehicle is considered a disruption. The final controller creates the regions at the start of each prediction time or when a disruption occurs, only using the previously created DNN. The obtained solution is validated using data from an e-vehicle charging station installed at the University of Trieste, Italy.
