• Overview
• Visitors
• People
• Links

• Groups
• Publications
• Software

• Courses
• SA/MA
• Lab

• News
• Seminars
• Jobs

Randomized MPC for Building Control

Overview

An increase in energy use efficiency is a key to reducing our dependency on fossil fuels and limiting of worldwide CO2 emissions. Approximately 40% of the global used energy is consumed in buildings, of which roughly half is used for Heating, Ventilation and Air Conditioning (HVAC). At the same time, most investments in building energy efficiency can be expected to pay back through reduced energy bills.

Improved HVAC control is particularly interesting because of the relatively low cost of improved control solutions as compared to other refurbishments in buildings, the comparatively short (10-15 years) lifetime of HVAC systems, and, last but not least, thanks to advances in sensor networks and information technologies that open up entirely new opportunities for advanced building control. Challenges relate to the uniqueness of many buildings, the possible complexity of their technical systems, and the building industry’s fragmentation.

A large scale simulation study carried out with participation of IfA showed model predictive control (MPC) incorporating weather and occupancy forecasts to have the potential of significant energy savings in many building set-ups. This year we started together with industrial partners (Siemens, Gruner) a follow-up project to implement an MPC controller in a representative office building of Actelion Pharmaceuticals Ltd in Basel. The thesis work can be expected to contribute to this project.

Project Description

As we know from everyday experience, weather predictions are subject to uncertainty. In the simulation study it was shown that a MPC that takes into account this uncertainty outperforms a MPC that assumes perfect weather predictions (predictions = realizations). The currently available stochastic MPC approach assumes that the distribution of the uncertainty is Gaussian. This is however not always valid with regard to weather and occupancy predictions. Recently a new MPC paradigm called randomized MPC (RMPC) has emerged that does not require the disturbance to be of a certain distribution but only that a sufficient large set of samples from the stochastic process can be generated.

The goal of this thesis is first to analyze the uncertainty in the weather predictions and then to compare the performances of a RMPC, a MPC assuming normally distributed disturbances and a MPC that assumes perfect weather predictions by means of simulation using already available models of the building. Since the approaches differ in several aspects, the thesis comprises the understanding of both paradigms and elaborating sensible comparisions before actually computing quantitative results. Finally the RMPC could be tried out on the Actelion building.

In case of a good thesis outcome, participation in a paper is likely.