Neural-Network- Based Multiple Stage Optimal Control of Variable-Air- Volume Systems

Optimal control schemes were developed for the efficient operation of energy-storage variable-air-volume (VAV) systems to deal with multiple time-scale of air-conditioning system processes and building operation schedules. A multi-layer forward neural network acted as the optimal feedback controller to optimize thermal comfort, energy consumption and operation cost. The neural controller was trained with predictive optimization algorithm to approximate the optimal feedback solution. Other neural networks were used for predicting time-varied parameters such as temperature and thermal load. The optimization methodology was used for a heat-storage VAV system by considering time-of-day price structure for electrical energy. Simulation results show that the developed optimization methodology can compensate uncertain plant nonlinearities and can be used to determine energy efficient operating strategies for VAV systems without violating the thermal comfort in buildings