Model Predictive Control of Residential Central Heating using Economic and Weather-based Data
Authors
Pedersen, Christian Schneider ; Poulsen, Andreas Just
Term
4. term
Education
Publication year
2023
Submitted on
2023-06-01
Pages
79
Abstract
Dette speciale undersøger, hvordan modelprædiktiv styring (MPC) kan reducere driftsomkostningerne for centralvarme i en enkeltfamiliebolig ved at udnytte vejr- og elprisprognoser. En modstand-kondensator (RC) ækvivalent model af boligen identificeres fra måledata med en subspace-metode og udtrykkes som et diskret tilstandsrumsystem. Med en Luenberger-observatør og prognoser for elpris og udetemperatur løser MPC løbende et kvadratisk programmeringsproblem for at vælge styrsignaler, under begrænsninger på ændringshastighed og et outputområde afhængigt af udetemperaturen. En dynamisk prediktionshorisont tilpasses den danske elprisplan, og en direkte term indføres for at inkorporere elpris i reference-sporingsvægte, med værdien fundet systematisk via simulation. Controlleren simuleres og sammenlignes med en baseline-MPC uden prisvægte; for perioden 2022–2023 opnås en samlet omkostningsreduktion på 7,3 %, mens besparelserne i andre simulerede perioder generelt er mindre, sandsynligvis pga. manglende prisenormalisering. Tilgangen adresserer begrænset sensorinformation ved at behandle forsyningstemperaturen som input og fastholder et komfortområde.
This thesis examines how model predictive control (MPC) can lower the operating cost of central heating in a single-family residence by leveraging weather and electricity price forecasts. An equivalent resistor–capacitor (RC) building model is fitted to measured data using a subspace identification method and expressed as a discrete state-space system. With a Luenberger observer and forecasts for electricity prices and ambient temperature, the MPC continually solves a quadratic programming problem to select control inputs, subject to constraints on rate of change and an output range that depends on outdoor temperature. A dynamic prediction horizon aligns with the Danish electricity price schedule, and a direct term is introduced to incorporate price signals into reference-tracking weights, with its value determined systematically by simulation. The controller is simulated and compared to a baseline MPC without price weights; for 2022–2023 it achieves a 7.3% reduction in total operating cost, while savings in other simulated periods are generally smaller, likely due to missing price normalization. The approach addresses limited sensor availability by treating supply temperature as the input and preserving a defined comfort region.
[This summary has been generated with the help of AI directly from the project (PDF)]
Keywords
Documents