System for monitoring and targeting of energy and water conusmption in public buildings

Potrošnja finalne energije u svetu beleži stalan rast. Kontinuirano unapređenje kvaliteta života u razvijenim zemalja uslovilo je postepeno izmeštanje težišta potrošnje energije iz oblasti industrije u oblast zgrada. Činjenice da se broj stanovnika na planeti uvećava i da životni standard u zemljama u razvoju raste ukazuju da će potrošnja energije u zgradama postajati sve prisutnija tema. Cilj ove doktorske disertacije je razvijanje metodologije za praćenje i predviđanje potrošnje energije i vode u javnim zgradama, pri čemu su očekivani rezultati metodologije relativno intuitivni i univerzalni prediktivni modeli. Podaci potrebni za izradu prediktivnih modela su pribavljani u saradnji sa komunalnim i javno komunalnim preduzećima grada Kragujevca, anketiranjem zaposlenih i merenjima sprovedenim u javnim zgradama. Prediktivni modeli su razvijani na podskupu podataka za obuku a testirani na podskupu podataka za test modela. Za potrebe izrade prediktivnih modela, primenjene su i analizirane linearne (prosta linearna regresija i višestruka linearna regresija) i nelinearne metode (stablo odlučivanja i veštačke neuronske mreže). Svaka metoda zahteva poseban pristup odabira parametara koji utiču na ishod predviđanja. Metode su primenjene na studiji slučaja javnih predškolskih ustanova. Preciznost primenjenih metoda je posmatrana za različite vrednosti potrošnje vode, električne i toplotne energije. U tom smislu, modeli su ispoljavali različite prediktivne sposobnosti na celokupnom skupu podataka, i podskupovima podataka koji predstavljaju različite raspone potrošnji. Iako nelinearne metode pokazuju veću prediktivnu preciznost, na kriterijume za odabir najpogodnije metode, pored sveukupne preciznosti, utiču faktori kao što su: broj parametara potrebnih za formiranje modela, preciznost modela u različitim rasponima potrošnji i nivo stručnosti onog koji metode primenjuje.Final energy consumption records constant growth. Continuous improvement of life quality in developed countries caused the building sector to become the most demanding consumer of energy. Having that in mind, and the fact that the world population is growing while the standard of living in developing countries rise, one can assume that tasks dealing with energy consumption in buildings will become even more challenging in the future. The goal of this dissertation is to develop the methodology for monitoring and targeting of water and energy consumption. The expected results of the methodology are relatively simple and precise predictive models. Required data for the model development was collected in cooperation with public utility services in the city of Kragujevac, and throughout interviews and measurements conducted in city public buildings. All the predictive models were developed on the training data set and verified on the test data set. To develop different predictive models and test their precision and ease of use, linear (Simple Linear Regression and Multiple Linear Regression) and nonlinear (Decision Tree and Artificial Neural Networks) methods were applied. All the methods require different approaches to determine input attributes influencing predicting results. The developed methodology was tested on a case study, public kindergarten buildings. The predictive precision of all the methods applied was determined for different categories of heat, electricity, and water consumption. Although nonlinear methods show better predictive precision, the criteria for optimal method selection is rather based on the possibilities for model utilization. Besides overall predictive precision, factors influencing optimal model selection are the number of input parameters, model precision for different categories of consumption, and level of expertise and experience of those utilizing the model