As the ubiquitous electric power internet of things (UEPIoT) evolves and IoT data increases, traditional scheduling modes for load dispatch centers have yielded a variety of challenges such as calculation of real-time optimization, extraction of time-varying characteristics and formulation of coordinated scheduling strategy for capacity optimization of electric heating and cooling loads. In this paper, a deep neural network coordination model for electric heating and cooling loads based on the situation awareness (SA) of thermostatically controlled loads (TCLs) is proposed. First, a sliding window is used to adaptively preprocess the IoT node data with uncertainty. According to personal thermal comfort (PTC) and peak shaving contribution (PSC), a dynamic model for loads is proposed; meanwhile, personalized behavior and consumer psychology are integrated into a flexible regulation model of TCLs. Then, a deep Q-network (DQN)-based approach, using the thermal comfort and electricity cost as the comprehensive reward function, is proposed to solve the sequential decision problem. Finally, the simulation model is designed to support the validity of the deep neural network coordination model for electric heating and cooling loads, by using UEPIoT intelligent dispatching system data. The case study demonstrates that the proposed method can efficiently manage coordination with large-scale electric heating and cooling loads.