Ján Drgoňa will give a seminar on "Model Predictive Control with Applications in Building Thermal Comfort Control" on Thursday, July 6, 2017 at 11:00 at IAM in room no. 641.

FB-IAM:Event

Abstract: This thesis deals with applications of model predictive control (MPC) on the building climate control problems. Many studies have proved that building sector can significantly benefit from replacing the current practice rule-based controllers (RBC) for more advanced control strategies like MPC. Despite this intensive research, the application of the MPC in practice is still in its early stages. This is mainly because the MPC requires an accurate controller model of the building envelope and its heating, ventilation and air conditioning (HVAC) systems. However, the necessary level of the model complexity to obtain a good MPC performance remains a priori unknown, and no systematic method is available. This thesis introduces such systematical investigation of the required controller model complexity necessary to obtain the optimal control performance for a given building. Moreover, the optimization-based control algorithms, like MPC, impose increased hardware and software requirements, together with more complicated tuning and error handling capabilities required from the commissioning staff. This problem is tackled in this thesis, by two ways. First, it is shown how the explicit solutions can be synthesized even for the MPC formulations taking into account uncertainties in the weather predictions. The main bottleneck of this approach, however, are its limitations only to the problems of modest complexity. This drawback is further eliminated by introducing a versatile framework for synthesis of simple, yet well-performing control strategies that mimic the behaviour of optimization-based controllers, also for the large scale multiple-input-multiple-output (MIMO) control problems which are common in the building sector. The idea is based on devising simplified control laws learned from MPC by exploiting the powers of multivariate regression algorithms and dimensionality reduction techniques. The main advantage of the proposed methods stems from their easy implementation even on low-level hardware without the need for advanced software libraries.