J. Drgoňa, D. Picard, M. Kvasnica, and L. Helsen have shown how to design well-performing approximations of optimization-based controllers applicable on embedded hardware by using advanced machine learning algorithms. The method presented in the paper titled “Approximate model predictive building control via machine learning” and published on ScienceDirect.com is based on two multivariate regression algorithms, namely deep time delay neural networks (TDNN) and regression trees (RT). The highlights:

  • The construction of approximate model predictive control laws (MPC) is presented.
  • Easy implementation of advanced control strategies suitable for low-level hardware.
  • Multivariate regression and dimensionality reduction algorithms are used.
  • Simulation case study employing temperature control in a six-zone building.
  • Simplified control laws retain most of the performance of the original MPC.