Building load demand response method based on deep reinforcement learning in multi-agent game environment

The invention discloses a building load demand response method based on deep reinforcement learning in a multi-agent game environment, and the method comprises the following steps: firstly, collecting the load data of a building in a power system, building a load model, and building a Markov game process model for the demand response of the building; secondly, designing a reward function according to the stability requirement of the power system, and defining Nash equilibrium; thirdly, an intelligent agent comprising a strategy network and a value network is established for each building, the strategy network outputs a load action sequence, and the value network evaluates the load adjustable potential of the building. And finally, training a value network and a strategy network by using the load data of the building until all agents converge to Nash equilibrium. According to the method, the game process of a plurality of buildings under the demand response system is simulated, and a reward function is designe