An effective dual-predictor controller mechanism using neural architecture search for optimization of residential energy hub system

The energy hub integrates heating, cooling, and electrical units for residential setups equipped with photovoltaic (PV) systems and energy storage, aiming to minimize energy consumption costs while optimizing the performance of the energy hub system. However, uncertainties such as fluctuations in electricity market prices and variations in solar irradiance can disrupt the stability and robustness of the optimal control of the system. To address this, we propose a dual-predictor controller mechanism, which integrates a recurrent neural network (RNN) sequence predictor based on neural architecture search (NAS). As dataset complexity and scale continue to increase, designing appropriate RNN architectures tailored to different data distributions has become a critical factor in overcoming performance bottlenecks in sequence prediction tasks. More specifically, the neural architecture of RNN is optimized by differentiable architecture search (DARTS), known as an effective NAS method. In this way, RNN has better ability of forecasting disturbances caused by uncertainties within the complex system. Then, we design a robust model predictive controller (RMPC) in the dual-predictor controller framework, and uses linear inequality matrices to ensure the precise planning under uncertain conditions, which further enhances the stability of the optimized control strategy. Simulation results under various scenarios demonstrate that the proposed optimal control strategy is notably effective in reducing energy hub costs.