Analysis, modelling and state estimation for large scale electric demand response
The need for additional reserves increases alongside the intermittency of generation and whilst rotating (conventional) generation is replaced, the system’s inertia reduces and balance volatility increases. Conceptually, any regulation measure from the “generation side” has an equivalent countermeasure from the “demand side”. One of the emerging technologies to provide such balancing services is Demand Response (DR). DR is commercially used, mainly via industrial loads combined with small scale diesel and gas generators. However, there is a lot of potential for DR from residential and commercial loads that remains untapped due to implementation costs, lack of technology expertise, load pattern complexity and the need to simultaneously control numerous sources. The main focus of this thesis is to explore the potential of loads, mainly residential and small commercial, to provide DR services and develop methods focused on accuracy for the most challenging services (frequency regulation), whilst aiming for minimal infrastructure and implementation costs. The main points include analysis of common residential and commercial loads for DR services, focusing on thermostatically controlled loads (TCLs). TCLs are thermal loads which operate via thermostats on a duty cycle (on and off state), between two temperature settings in order to maintain an average set temperature. They use electricity as a primary energy source or for their control and pumps. The next part includes analysis and creation of realistic bottom up models to study aggregated behaviour of TCLs during DR actions, as well as the effect of external factors. Afterwards, a distributed State Estimation algorithm is proposed to increase accuracy of aggregated models and track aggregation models from limited information. A new aggregation framework is proposed, specifically designed for heterogeneous populations, whilst being universal for all TCL types. As such, different TCL types can be aggregated together (e.g. cooling and heating). The results of this thesis show that with proper aggregation modelling, state estimation and dynamic updating in time, accuracy of stochastic aggregated models is improved compared to existing frameworks without the need for expensive thermal sensors. This suggests that with relatively limited information the use of residential and commercial TCLs for DR balancing services, is feasible.