A bootstrapped automated pipeline for developing model predictive controllers for non-domestic buildings

In this paper, we motivate and investigate an alternative approach to the development of predictive models for the practical implementation of model predictive control in non-domestic buildings. We describe how the process can be ‘bootstrapped’ with a very simple model, the crude nature of which illustrates the robustness of our approach. A predictive model for the controller is refined/adapted to the building in operation while maintaining climate control throughout at all times using closed-loop system identification. To remove the necessity for human intervention, we have used genetic programming to learn the predictive models since this combines a number of what are traditionally sequential search operations into a single step. We report preliminary results of a series of simulation experiments that validate the basic approach, and identify further research needed to develop the proposed methodology. Our approach facilitates the adoption of model predictive control by using commissioning data and refinement of models with data from the occupied building, while maintaining thermal comfort. • Addressed the costs of models for model predictive control (MPC) of buildings. • Proposed an automated pipeline to generate predictive MPC models. • Bootstrapped the pipeline with an easy-to-identify model compatible. • Investigated the iterative model improvement using closed-loop re-identification.