Quantized nonlinear model predictive control for a building

In this paper, the task of quantized nonlinear predictive control is addressed. In such case, values of some inputs can be from a continuous interval while for the others, it is required that the optimized values belong to a countable set of discrete values. Instead of very straightforward a posteriori quantization, an alternative algorithm is developed incorporating the quantization aspects directly into the optimization routine. The newly proposed quaNPC algorithm is tested on an example of building temperature control. The results for a broad range of number of quantization steps show that (unlike the naive a posteriori quantization) the quaNPC is able to maintain the control performance close to the performance of the original continuous-valued nonlinear predictive controller and at the same time it significantly decreases the undesirable oscillations of the discrete-valued input.