Time-Varying Demand Response for an Electro-Thermal Integrated System: An Efficient and Robust Zeroing Neural Network Approach

As a specialized recurrent neural network (RNN), the Zeroing Neural Network (ZNN) has demonstrated efficacy in solving diverse real-time, time-varying problems. This paper proposes a ZNN-based model framework to address the challenge of time-varying demand response within building energy management systems, specifically for thermo-electric integrated systems. The framework encompasses model construction, control design, performance analysis, and robustness testing. Results indicate that the system’s target power effectively adapts to real-time electricity pricing with minimal response latency. The ZNN controller achieves extremely low tracking error, with a maximum observed value of merely −1.38 kW. Furthermore, the system exhibits strong robustness against disturbances in the coefficient of performance (COP) and achieves lower tracking error with increased thermal storage tank heat capacity. The ZNN has significant performance advantages in time-varying demand response scenarios.