TY - JOUR A1 - Kontes, Georgios A1 - Giannakis, Georgios A1 - Sánchez, Víctor A1 - de Agustin-Camacho, Pablo A1 - Romero-Amorrortu, Ander A1 - Panagiotidou, Natalia A1 - Rovas, Dimitrios A1 - Steiger, Simone A1 - Mutschler, Christopher A1 - Gruen, Gunnar T1 - Simulation-Based Evaluation and Optimization of Control Strategies in Buildings N2 - Over the last several years, a great amount of research work has been focused on the development of model predictive control techniques for the indoor climate control of buildings, but, despite the promising results, this technology is still not adopted by the industry. One of the main reasons for this is the increased cost associated with the development and calibration (or identification) of mathematical models of special structure used for predicting future states of the building. We propose a methodology to overcome this obstacle by replacing these hand-engineered mathematical models with a thermal simulation model of the building developed using detailed thermal simulation engines such as EnergyPlus. As designing better controllers requires interacting with the simulation model, a central part of our methodology is the control improvement (or optimisation) module, facilitating two simulation-based control improvement methodologies: one based in multi-criteria decision analysis methods and the other based on state-space identification of dynamical systems using Gaussian process models and reinforcement learning. We evaluate the proposed methodology in a set of simulation-based experiments using the thermal simulation model of a real building located in Portugal. Our results indicate that the proposed methodology could be a viable alternative to model predictive control-based supervisory control in buildings. KW - model predictive control in buildings KW - reinforcement learning KW - data-driven control KW - simulation model KW - multi-criteria decision analysis Y1 - 2018 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:92-opus4-4053 N1 - Funding: 1. H2020-EeB-2015: Modelling Optimization of Energy Efficiency in Buildings for Urban Sustainability (MOEEBIUS, #680517) 2. H2020-EeB5-2015: Optimised Energy Efficient Design Platform for Refurbishment at District Level (OptEEmAL, #680676) 3. Federal Ministry of Education and Research of Germany in the framework of Machine Learning Forum (grant number 01IS17071) VL - 11 IS - 12 PB - MDPI AG CY - Basel, Switzerland ER - TY - CHAP A1 - Kontes, Georgios A1 - Sánchez, Víctor A1 - de Agustin-Camacho, Pablo A1 - Tellado, Borja A1 - Giannakis, Georgios A1 - Steiger, Simone A1 - Gruen, Gunnar A1 - Rovas, Dimitrios T1 - Data-driven approach for robust supervisory control in buildings N2 - Model Predictive Control (MPC) for building energy management and indoor climate regulation has been an active research topic in the last decade. A plethora of research work has been published, while impressive results in real-world and simulated applications have been reported in relevant projects and publications. Despite the fact that MPC has evolved into a mature technology with tangible expected benefits, it has not found its way into the market. There are very few companies that include some simple aspects of MPC in their offerings – and still these implementations are far from being fully-fledged MPC solutions. This reluctance of companies to include MPC in their product lines cannot be attributed to any conservatism on the side of the industry; some of the mechanics of MPC are incompatible with the specific needs of building owners and operators. The biggest obstacle here is that MPC requires a model of the building, usually of a very specific linear or bi-linear structure. Model development and calibration induce the largest monetary costs in MPC application and in many cases can make the entire investment unattractive. Another important aspect is the inability of building managers to evaluate the performance of MPC and detect any errors in the setup in case of poor control configurations. Unlike traditional knowledge- based control in buildings where a controller consists of a set of (simple) rules, in MPC an optimization algorithm constantly designs new control strategies for the near future (usually in the form of time- series of setpoints for the building actuators), taking into account the available model of the building and forecasts. If poor control actions are generated, a large amount of information needs to be back- tracked – a task significantly more complex than examining the execution of simple control rules. Taking into account all the above, in the present work we define a methodology that borrows some of the ideas that enable the great potential of MPC, while at the same time satisfies the requirements of building owners on control robustness, explainable control decisions and low cost. To achieve this, in the heart of the methodology lies a data-driven approach for predicting the future thermal state of the building. So, instead of using purpose-built models for every study building, we utilize Gaussian Processes (GPs) to construct regression models based on historical data from the building. A Gaussian Process is a collection of random variables, any finite number of which has a joint Gaussian distribution and is completely specified by its mean and covariance functions. This means that the prediction of a GP for a specific future state is not a point estimate, but a distribution that reflects the predictive confidence of the model. This way, the optimization algorithm can avoid regions with high uncertainty, thus leading to more robust control. The second ingredient of our solution is that the optimized control strategies are not necessarily time- series of setpoints to be applied to the actuators; they can be optimized parameters of the knowledge- based controllers already running in the building. For example, if the supply water temperature of a radiant system is regulated using a heating curve with the external temperature as input, then the optimized controller of the proposed methodology would be a new heating curve, adapted e.g. to the weather predictions of the following day. This feature allows building managers to interpret the control actions applied to the building and increases their acceptability of the solution. The third aspect of the methodology is that instead of re-designing the control frequently (e.g. every hour) like in the classical MPC, we design and apply controllers for the entire day. This reduces significantly the computational complexity associated with classical online MPC, and makes our solution mode robust to connectivity problems or downtimes. In the work presented here, we evaluate the proposed approach at simulation level, using an EnergyPlus model of a real office building in Portugal. The building is equipped with a set of Variable Refrigerant Flow (VRF) systems for heating and for cooling and our solution optimizes the day-to-day operational cost (under a time-of-use tariff) without compromising occupant comfort, by controlling the heating setpoints in selected offices. At the same time, we provide evidence on the quality of the GP predictions on the problem at hand, as well as on the scalability of the proposed approach. On the other hand, two strong requirements of the methodology (similar to the requirements of classical system identification theory) became apparent from the experiments. First, high-quality historical data need to be available from the real building in order to train the GP models. And second, the available historical data may not capture the full model dynamics as e.g. the control setpoints of all offices might be set to a specific value and thus lack in input excitation. In this case, the proposed solution will be unable to discover optimized control strategies “far” from the training data. KW - Data-driven control Y1 - 2018 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:92-opus4-4061 N1 - Funding: 1. H2020-EeB-2015: Modelling Optimization of Energy Efficiency in Buildings for Urban Sustainability (MOEEBIUS, #680517) 2. H2020-EeB5-2015: Optimised Energy Efficient Design Platform for Refurbishment at District Level (OptEEmAL, #680676) 3. Federal Ministry of Education and Research of Germany in the framework of Machine Learning Forum (grant number 01IS17071) CY - Nuremberg, Germany ER -