The invention discloses a closed-loop self-learning demand response system and method. The system comprises a DR server, a DR client and an intelligent electric meter, an electric power company in the DR server distributes the latest electric energy information and service control information to stimulate the DR client to dispatch electric power load equipment, the DR client classifies and grades the input electric energy information and the power of the electric power load equipment by the aid of a multi-input single-output network, continuously learns and determines the use plan of the electric power load equipment based on rules, predicts electric power demand information, collects the information to the DR server to form a closed loop, stimulates the electric power company to collate, clean and mine the demand information and performs future DR planning, dispatching, control and new information generation based on processing results. Network load can be timely and effectively balanced, a user is reminded of switching electric energy, user expenses are decreased, and network stability is improved.