The high thermal energy consumption is one of the main drawbacks hampering the commercial implementation of Membrane Distillation (MD) technology. The development of adequate operating strategies can help to reduce these energy requirements. Accordingly, this paper focuses on the optimal management of the array of MD modules composing a commercial-scale MD plant, trying to reduce their thermal energy consumption while ensuring a given water need. For this aim, the array of MD modules is modelled as a Mixed Integer Programming (MIP) system to consider that some modules can be turned on/off depending on the operation specifications. An algorithm based on the Generalized Bender Decomposition (GBD) is then developed for the efficient solution of the problem. This algorithm is incorporated in a Model Predictive Control (MPC) strategy allowing to manage the plant in real time. The effectiveness of the proposed strategy is verified using a practical example. The obtained results are compared with a manual and a previous strategy presented in literature, showing that for a sunny day, around the 65 and 55% of the thermal energy consumed by these methodologies can be saved, which means important thermal energy savings that can be relevant for the industrial implementation of MD technology. (C) 2020 Elsevier Ltd. All rights reserved.