Predictive demand side management strategies for residential building energy systems
Aachen / E.ON Energy Research Center, RWTH Aachen University (2017) [Book, Dissertation / PhD Thesis]
Page(s): 1 Online-Ressource (xxix, 126 Seiten) : Illustrationen
This thesis presents a generalized methodology, supported by a software framework, for modeling and assessing mathematical programming based predictive demand side management (DSM) strategies that exploit thermal and electrical flexibilities of residential building energy systems (BES) to enhance the integration of renewable energy sources.The modeling and simulation platform is formulated in Python and includes a set of forecasting methods as well as a discrete mixed integer linear programming (MILP) modeling library based on the Gurobi optimizer API. The platform further integrates a non-linear BES simulation model in Dymola/Modelica as a functional mock-up unit (FMU). The evaluation of the forecasting algorithms shows that the lowest forecast error for predicting electrical and space heating demands is provided by SVR and for predicting the weather variables, i.e. temperature and solar irradiation, by ARMA, respectively. The persistence method is selected for predicting the strongly stochastic domestic hot water demand. The introduced scheduling HEMS models for individual buildings consist of a deterministic MILP strategy and a multi-stage stochastic programming (SP) approach that extends the MILP model while incorporating the uncertainties of the electrical and domestic hot water demands. The city district DSM strategies comprise a centralized approach, which serves as a benchmark, as well as distributed formulations based on decomposition techniques. In this work, two distributed DSM approaches are formulated, Dantzig-Wolfe decomposition based column generation (CG) algorithm as well as an integrated Lagrangian decomposition column generation (LRCG). The performance of the scheduling algorithms for individual buildings is evaluated for different BES configurations. The results indicate that predictive HEMS with perfect information enhance the integration of locally generated electrical power from PV units and enables a significant potential of load shifting and cost reduction with respect to a reactive control strategy. Employing point forecasts of weather and energy demand variables within the deterministic scheduling model reduces this potential but further allows for a higher integration of PV power as well as cost reduction compared with a reactive strategy. The multi-stage SP model outperforms the deterministic approach but induces larger modeling and computation effort. The analysis of the DSM strategies for city districts shows that the CG approach provides comparable coordination performance as the centralized model while significantly reducing the computation time. Further, the integrated LRCG approach enables a faster convergence compared with the standard CG formulation.