Flexibility prediction, scheduling and control of aggregated TCLs

There should be a constant balance between the demand and supply of electrical power. In Nordic countries, electricity markets have been formulated in such a way so as to facilitate this balance. These markets enable purchases, through bids, for buying and selling the energy (e.g., the day-ahead market) and the reserves (e.g., the frequency containment reserve for normal operation (FCR-N)). Demand response (DR) has received increased attention in recent years since it can efficiently support bidding in these markets. Aggregators, which act as mediators between end-users and the system operator, play an important role here. The aggregator contracts a large number of end-users for DR programs, and plans and controls their heterogeneous thermostatically controlled loads (TCLs), and offers their load flexibility to the markets. Taking into account the small market value of each contributing unit, the cost for the communication and control system enabling the DR service must be kept at a minimum. In this paper, we propose a framework which is adaptable to pre-existing and newly emerging TCLs, with no need for major re-design of the local control loops. We then design a strategy for the aggregator, to predict, schedule and control the aggregated flexibility of the contracted heterogeneous TCLs, in response to the DR signals and in the presence of end-users' behavior uncertainties. In this strategy, we have applied a recurrent neural network (RNN) which learns the aggregated consumption of end-users and predict their aggregated load flexibility. The scheduling and control algorithms are then designed with the aim of participation in FCR-N market. We show that uncertainties in the prediction and scheduling are compensated in the control stage by activating back-up resources. A numerical study on 2000 number of detached houses has been conducted, which shows available 500 kW capacity for participation in the FCR-N market.