Neural-network-based dynamic model of VAV systems

This study proposes a neural network based nonlinear dynamic modeling method to find the predictive model of VAV systems for implementing better real-time control and optimization of VAV system performance. On studying the generalization theory of feed-forward neural networks, a dynamic VAV air-conditioned zone model is developed to simulate the system behavior under predictive control schemes. The neural model is constructed and trained with Bayesian framework combined with “early stopping” method. The model was validated against real data gathered from an existing VAV system. Validation results show that the model has high accuracy of predicted outputs and good generalization abilities.