A collaborative control strategy for domestic air conditioners with adaptive access is proposed addressing communication barriers caused by diverse and incompatible device protocols in heterogeneous domestic air conditioners

along with low computational efficiency in real-time regulation. First

the domestic air conditioner information interaction architecture is constructed

and the adaptive access method is proposed on this basis. Then

the deep reinforcement learning multi-conditioner collaborative control strategy is developed

and the soft-max sampling strategy and the prioritized experience replay mechanism are introduced to improve the MAD3QN algorithm

and the SMPER-MAD3QN algorithm is proposed. Finally

a centralized training with decentralized execution is implemented based on SMPERMAD3QN

which allows multiple air conditioners to collaboratively participate in the regulation of the algorithm. The simulation results measured that the packet loss rate of multi-protocol domestic air conditioner information interaction is 0.36%

and the interaction latency is lower than 25ms

which indicates that the adaptive access can significantly shorten the real-time decision-making time and realize the unified management and control of multi-protocol domestic air conditioner. Meanwhile

the proposed algorithm realizes the collaborative participation of multiple air conditioners in demand response（DR）under the premise of guaranteeing the comfort of users

and the algorithm has excellent robustness

which improves the flexibility and reliability of the dispatchable resources on the demand side.