Izgradnja linearnih regresijskih modelov za napovedovanje toplotnega odziva stavbe

Begelj, Žiga (author), Govekar, Edvard (mentor), Potočnik, Primož (comentor)

V okviru magistrskega dela smo raziskali izdelave modelov za dolgoročno in kratkoročno napoved toplotnega odziva stavbe, ogrevane s toplotno črpalko. V ta namen smo pripravili več množic podatkov, ki so bile sestavljene iz vremenskih podatkov ter pripadajočih podatkov o temperaturah dvižnega voda toplotne črpalke in toplotnega odziva stavbe, pridobljenih s pomočjo TRNSYS simulacije referenčne stavbe. Z uporabo linearne in stopenjske regresije smo zgradili več vrst modelov za dolgoročno in kratkoročno napovedovanje temperature v stavbi ter jih primerjali s hevristično razvitimi modeli. Pri tem smo za dolgoročno napoved uporabili regresijski model, za kratkoročno pa avtoregresijski in avtoregresijski z dodanimi zunanjimi vhodi. Rezultati raziskave so pokazali, da so zaradi narave uporabljenih regresorjev dolgoročni modeli manj točni od kratkoročnih, vendar uporabni za večmesečne napovedi, kratkoročni pa so točnejši, vendar uporabni le za kratek horizont napovedi. V delu predstavljeni dolgoročni modeli za večmesečne napovedi dosegajo RMSE-napake okrog 0,5 °C, najtočnejši kratkoročni model pa za horizont napovedi do 24 ur vnaprej napoveduje temperaturo v stavbi z RMSE-napako pod 0,05 °C. Razviti dolgoročni modeli so uporabni za optimizacijo različnih ogrevalnih sistemov v stavbah, razviti kratkoročni modeli pa so z izkazano visoko točnostjo uporabni pri prediktivnem vodenju ogrevalnih sistemov.

• Language: Slovenian
• Keywords: modeliranje toplotnega odziva stavb linearna regresija stopenjska regresija predikcijski modeli dolgoročni predikcijski modeli kratkoročni predikcijski modeli
• Work type: Master's thesis/paper
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2018
• PID: 20.500.12556/RUL-100615
• Publication date in RUL: 01.04.2018

Secondary language

• Language: English
• Title: Development of Linear Regression Models for Building Heat Response Prediction

Abstract

Within the master’s thesis, we investigated the long-term and short-term modelling of the heat response of the building, heated by a heat pump. For the purpose of the research, we prepared several data sets that consisted of weather data, heat pump temperature data and heat response data of the building obtained through the TRNSYS simulation of the reference building. The prepared data was used for learning and cross-validation of models. On the basis of linear and stepwise regression, we built several types of long-term and short-term models for predicting the temperature in the building and compared them with heuristically developed models. Long-term models are regressive and short-term models are autoregressive and autoregressive with exogenous inputs. Due to the used regressors, long-term models are less accurate than short-term ones, but useful for several month predictions, while hort-term ones are more accurate, but useful only for short prediction horizons. The presented long-term models for several month predictions achieve RMSE error of around 0,5 °C, while the most accurate short-term model for a prediction horizon of up to 24 hours predicts the temperature in the building with a RMSE error below 0,05 °C. Presented long-term models are useful for optimizing various heating systems in buildings, and developed short-term models are due to high accuracy useful in model predictive control of heating systems.

• Keywords: building heat response modelling linear regression stepwise regression prediction models long-term prediction models short-term prediction models