Development and First Testing of a Framework for Predictive Energy Control of Underground Stations

Building Energy Management Systems consist of hardware and software components. The hardware set-up of BEMS is typically made up of a set of computers in charge of building control and sensoractuator networks. The software side of BEMS is usually made up of a number of functional layers that implement standard management functionalities. This paper will present an application of Model based Predictive Control (MPC) targeted to energy management of the “Passeig de Gracia” metro station in Barcelona. This approach uses the predictions of future building status, obtained by means of a set of Bayesian Networks, in order to determine the optimal control policies. First the predictive Bayesian Networks were developed through the following steps: structural learning based on a simulated dataset; improvement of the network’s topology through enhanced datasets derived from the previous one; final refinement and validation based on experimental data collected through a pervasive wireless monitoring network. Then those networks were integrated within a control framework, including control algorithms, a Dymola TM based virtual model of the station to simulate its evolvement and, on top of them, a user graphic interface to manage the system. The results about energy savings estimation determined by the application of model based predictive control to the station’s mechanical ventilation showed that as much as 35% can be saved on average.