文摘
In heating ventilating and air conditioning (HVAC) systems, typically two variables (air temperature and air humidity) have to be controlled via several (at least two) actuators. Some of the components show nonlinear behaviour. Therefore, HVAC systems belong to the class of nonlinear multi-input-multi-output systems. A well suited approach to control this class of systems is model predictive control, since the time constants of HVAC systems are high (typically in the range of tens or hundreds of seconds) offering enough time to perform the required online optimization. In order to apply linear predictive control methods, while taking into account the nonlinearities of the plant, a modeling concept based on a physical plant model and a neuro-fuzzy model is proposed. The neuro-fuzzy model is obtained via the so called local linear model tree (LoLiMoT) algorithm. The generation of a linear state space representation from the neuro-fuzzy model is demonstrated. This linear state space model can then be used in a predicitive control scheme, where the linear model is updated each sampling instant from the neuro-fuzzy model. This technique allows the application of standard linear predictive control while taking into account the nonlinearities of the plant. Simulation and measurement results obtained from an industrial test plant are presented.
