9401249 Hittle The objective of this research project is to increase the efficiency of control methods for energy systems in buildings. IN meeting this objective, novel combinations of feedback controllers and neural network learning algorithms will be developed. to achieve the objective, recent developments in reinforcement learning methods for neural networks will be adapted to the HVAC system control problem. Neural networks will not be the sole source of control decisions for such a complex system. Adaptive schemes are notorious for leading to instabilities. to lessen the possibility of introducing instability, adaptive schemes will be integrated with existing, fixed controllers. the adaptive methods will only augment the control actions taken by the fixed controllers and a limit will be enforced on the degree to which the fixed controller's decision can be modified. This will result in a control system with an initial performance at the level provided by the original control designer. Then, with experience, the learning mechanism will be able to increase the performance of the controller beyond its initial ability. Results will certainly be of interest to building engineers, but will also advance the sate-of-the-art in neural networks for control. ***
