In the modern power system, the intermittency of distributed renewable energy and user-side load fluctuations challenge grid stability. Using IoT data, the distribution regulation center directs user resources to engage in demand response through the network. However, deep coupling between the power and communication networks leads to issues like communication delays and packet loss, increasing regulation costs and affecting robustness. To address these issues, this paper proposes a demand response autonomous decision-making method considering regulation command communication delay and reliability. Initially, we develop a cloud-edge-end collaborative framework for autonomous demand response, considering the coupling of power and communication networks. We then define an optimization problem to minimize the weighted sum of regulation costs and grid losses, considering user-side resource outputs and grid constraints. Utilizing Information Gap Decision Theory (IGDT), we address uncertainties in renewable outputs like wind and PV, forming a robust optimization model. We introduce a deep actor-critic (DAC)-based method that incorporates communication delay and packet loss impacts, building a multi-objective Markov decision process (MDP) to optimize and adjust power outputs. This dual cooperative DAC algorithm iteratively refines decisions, enhancing algorithm convergence. Simulation results confirm the method significantly reduces regulation costs and grid losses, improving economic efficiency and robustness.