Perception and decision-making for demand response based on dynamic classification of consumers.

• "Operator-Aggregator-Consumers" joint framework is constructed and the aggregator's optimal trading strategy is analyzed. • To deal with market uncertainty, deep reinforcement learning is used to optimize selection and reward of the consumers. • The SOMs helps to realize dynamic consumer classification, improving the accuracy of response forecast and bid construction. Demand response plays a significant role in improving electricity market efficiency and keeping power system stability. Load aggregators can provide reliable resource for power balance and auxiliary services by integrating dispersed responsive loads. In order to meet the aggregators' requirement of timeliness and accuracy in the real-time demand response market, a method of perception and decision-making for demand response based on dynamic classification of consumer is proposed. The price elasticity of electricity demand is calculated based on continuously updated trading experience and applied as a classification criterion. Consumers are dynamically classified by self-organizing maps algorithm to perceive consumers' responsive ability. Furthermore, the interaction model of aggregator and consumers in market environment is constructed, and deep reinforcement learning is applied to solve the trading strategy in the uncertain market with incomplete information. Simulations show that the results of trading partner and reward price gained by the proposed method are appropriate. Therefore, the deviation between response quantity and the bid-winning volume is effectively controlled, and the transaction revenue of the aggregator is improved. [ABSTRACT FROM AUTHOR]