Model-predictive control of mixed-mode buildings with rule extraction

A series of model-predictive control (MPC) techniques have been explored for optimizing control sequences for window operation in mixed-mode (MM) buildings using EnergyPlus, and results for a simplified MM office building have been presented. Initial results for a small office in Boulder, Colorado show the ability to save upwards of 40% of cooling energy through near-optimal night cooling strategies, even in existing facilities. Strategies can be tuned to avoid overcooling the space by introducing heating energy into the objective function used in the MPC process. A complementary statistical technique has been introduced that allows for the “extraction” of logistic decision models from the optimal control results. The process works best when some time-lagged information is present as a predictor variable to ensure that some process memory is preserved. A generalized linear model (GLM) in the form of a multi-logistic regression was able to mimic the general characteristics of the optimizer results, achieving 70–90% of optimizer energy savings, but at a small fraction of the computational expense. Given the simple mathematical formulation of the logistic regression, it would be possible to implement this sort of decision model into modern direct digital control systems to control MM buildings in a near-optimal manner in real time.