Building data model identification and predictive demand response control

Abstract

Data-based models are conceptual tools for describing the real world entities and the connection among them. There are a lot of studies with usage of data-based structures for forecasting, optimization and fault detection in building energy systems.

The main objectives of this thesis have been devoted to evaluate how data-based modelling can contribute to energy building system identification and demand response control. In overall view, this research work contains the separate sections for identifying energy system of buildings and establishing the optimal demand response control.

In first stage of this research, multi stages prediction algorithm has been developed by means of a novel combination of signal processing technique (wavelet transformation) and dynamic neural network. Different prediction algorithms have been examined for forecasting the energy demand of building with considering random profiles in occupancy, lighting and equipment profiles. Also ability of model in forecasting energy demand has been investigated for different type of building structures and insulation levels.

The model predictive control then has been organized for achieving the optimal energy trading schedule in energy building systems. The system includes building, Photovoltaic (PV) component, heat storage tank and ground source heat pump (GSHP). The heat demand of building contains space heating and domestic hot water demands. Active thermal storage is used for saving energy regarding to dynamic electricity price. Then based on dynamic coefficient of performance (COP) for GSHP, the required electricity of compressor has been achieved. On-site energy generated by PV system has been balanced with electricity demand of system to support trading with grid.

The applied predictor structure improves the time of applying control algorithms for building energy systems signifi-cantly. The applied control methodology and on-site energy generation also save the energy cost as 28% and energy consumption as 17% for passive massive buildings. The results highlight the adaptability of proposed algorithm in prediction and identifying the energy system for different building structures and insulation levels. The comparative results also reveal the priority of the proposed method in aspect of prediction accuracy as compared to neural network. The integration of signal analysis and dynamic neural network is strong alternative for common simulation software in applying and evaluation of different control systems.