Base station dormancy and power distribution method based on deep reinforcement learning

The invention provides a base station dormancy and power distribution method based on deep reinforcement learning, which comprises the following steps of: establishing an energy consumption optimization model by taking spectrum efficiency and energy efficiency as optimization targets, and splitting a solving problem of the energy consumption optimization model into two sub-problems, namely a base station dormancy decision-making problem and a power distribution decision-making problem, determining an optimization target and a constraint condition of a base station dormancy decision problem and an optimization target and a constraint condition of a power distribution decision problem, and then respectively solving the base station dormancy decision problem and the power distribution decision problem by using two deep reinforcement learning networks; the method can solve the problems that the rate requirement of a user in the real world is ignored in the prior art, and the deviation of a data set may cause the