The new power system necessitates enhanced transmission and processing capacity of the communication network due to the frequent two-way interaction between the power grid and thermostatically controlled loads (TCLs). However, existing load regulation methods often assume that the communication system is in an ideal state, which means that instantaneous control of large-scale TCLs can be achieved. In practice, however, communication delays and data processing energy consumption cannot be avoided. In order to reduce the overall system regulation delay and energy consumption, this paper proposes a TCLs regulation method based on hybrid communication and cloud-edge-end collaboration, which aims to support the synergistic interaction between TCL resources and the power grid. Firstly, a hybrid communication model is used at the end layer, using the Gossip algorithm to aggregate information about the parameters of the regional TCLs. Then, each communication, caching and computation model in the cloud-edge-end collaboration architecture is analyzed. And the data volume offloading and computational power allocation are solved using the Hybrid Whale Grey Wolf Optimization (HWGWO) algorithm. Finally, the performance of the regulation method is analyzed and compared in terms of transmission delay and processing energy consumption using an arithmetic example. The results show that the hybrid communication model reduces the average communication link distance by 86.64% compared to the centralized communication and reduces the communication delay by 31.25% compared to the distributed communication. The proposed cloud-edge-end collaboration architecture reduces latency by 61.32%, 71.93%, and 52.80% compared with the single-layer architecture, respectively. The energy consumption meets the constraints, which can fully utilize the demand side TCLs regulation potential and is conducive to enhancing the stability of power grid regulation.