A Python-Based Toolbox for Model Predictive Control Applied to Buildings

摘要

The use of Model Predictive Control (MPC) in Building Management Systems (BMS) has proven to out-perform the traditional Rule-Based Controllers (RBC). These optimal controllers are able to minimize the energy use within building, by taking into account the weather forecast and occupancy profiles, while guaranteeing thermal comfort in the building. To this end, they anticipate the dynamic behaviour based on a mathematical model of the system. However, these MPC strategies are still not widely used in practice because a substantial engineering effort is needed to identify a tailored model for each building and Heat Ventilation and Air Conditioning (HVAC) system. Different procedures already exist to obtain these controller models: white-, grey-, and black-box modelling methods are used for this end. It is hard to determine which approach is the best to be used based on the literature, and the best choice may even depend on the particular case considered (availability of building plans, Building Information Models (BIM), HVAC technical sheets, measurement data). Nevertheless, the vast majority of researchers prefer the grey-box option. In this paper a Python-based toolbox, named Fast Simulations (FastSim), that automates the process of setting up and assessing MPC algorithms for their application in buildings, is presented. It provides a modular, extensible and scalable framework thanks to its block-based architecture. In this layout, each of the blocks represents a feature of the controller, such as state-estimation, weather forecast or optimization. Moreover, the interactions between blocks occur through standardized signals facilitating the inclusion of new add-ons to the framework. The approach is tested and verified by simulations using a grey-box model as the controller model and a detailed Modelica model as the emulator. A time-varying Kalman filter is applied to estimate the unmeasured states of the controller model. FastSim is developed and used in a research environment, however this automated process will also facilitate the implementation of MPC for different building systems, both in virtual and real life.