While new paradigms for the construction sector are emerging, such as that of “Nearly Zero Energy Building” and sustainability assessment schemes are progressively evolving and becoming more common, empirical studies show how, very often, the gap between the predicted (design phase) and measured (operation phase) energy performance could be very large. This issue is generally addressed with the term “performance gap” and it can be fundamentally caused by design phase, construction phase, commissioning and operational phase errors. This situation creates a problem of credibility in the building industry and, more in general, in sustainability practices because the errors committed can directly reflect on energy performance and, consequently on the running costs and on global cost optimality. Bridging the gap between simulated and real energy usage requires the presence of embedded control and fault detection applications that have to be adaptive, self-calibrating and conceived to enable a model based interpretation of measured data, for performance benchmarking. Considering these general issues, the original part of the research presented is given by the overall methodology that aims to address some of the fundamental problems present in monitoring and control systems. The methodology used for the research, therefore, seeks to unite these steps through the choice of effective tools, able to interact each other, trying to unify methods and models used in different fields and understand how to make them usable in an effective workflow across life cycle phases. Starting from a linearized lumped parameters model, an optimization algorithm is used to find the optimal operation trajectory that satisfies the constraints while minimizing energy demand. With a Linear Programming (LP), in fact, it is possible to find the best possible solution given the specific conditions (i.e. find a solution that is certainly a global optimum). Further, the optimization model, working on set-point violations, can be, in principle, simply added an additional layer on existing automation systems, without having to replace completely the technology, but simply working on the top of conventional PI/PID controllers. In the data analysis workflow presented, two main levels of model calibration are done; firstly, with monthly data the main lumped parameters are analyzed through regression models and, secondly, using hourly data, an identification technique (ARX model) is used to directly calibrate/built the whole model (suitable for multiple temporal scale of analysis) with the estimated parameters. Parameter identification allows also to analyze the evolution of parameters over time, and, thus, lay the foundations for a machine learning system application which may constantly improve the control system knowledge adapting and learning to building and end-users characteristics. The energy demand reduction, however, is not the only objective to be pursued. Internal environmental quality and energy savings can be conflicting objectives and comfort expectation can cause an “economic rebound effect”, limiting the energy saving potential of efficiency practices and, therefore, compromising the return of investment. The Predicted Mean Vote (PMV) evaluation acquires a crucial role in the choice of the limits to be set in the optimization algorithm. The latter in fact, needs a limit in terms of degree/days and degree/hours of comfort violation that can be chosen starting from a comfort assessment, done through the PMV evaluation. The advantages provided by a Model Predictive Control (MPC) system such as the one presented in this research are also dependent of the type of objective function adopted. Starting from the minimal energy demand, in fact, it is also possible to minimize costs (using dynamic tariffs) and emissions (using hourly CO2 emission time-series and emission factors) or considering a combination of both, using weighting factors. Further, it permits to increase the overall energy efficiency, considering both the load matching and the dispatch flexibility problem.
Sebbene la questione ambientale stia promuovendo un rapido cambiamento nelle politiche energetiche e di sostenibilità, studi empirici mostrano quanto, molto spesso, il divario tra i consumi calcolati in fase di progetto e quelli misurati durante la fase operativa dell’edificio sia notevole. Questo divario viene generalmente indicato con il termine "performance gap” e può essere dovuto ad errori commessi nelle varie fasi di progettazione, costruzione e gestione dell’edificio. Ciò sta generando un vero e proprio problema di credibilità sia in ambito edilizio che, più in generale, sulle pratiche di sostenibilità perché si riflette direttamente sul rendimento energetico e, di conseguenza, sui costi di esercizio. Colmare questo divario richiede quindi la presenza di sistemi di controllo e rilevamento guasti che devono essere in grado di adattarsi e calibrarsi alla variazioni in tempo reale. Considerando questi aspetti generali, la tesi qui presentata si propone di trovare una metodologia in grado di affrontare e risolvere alcuni dei problemi fondamentali che riguardano le fasi di monitoraggio e controllo delle prestazioni dell’edificio. La ricerca, quindi, cerca di unire queste fasi attraverso la scelta di strumenti efficaci, in grado di interagire tra di loro, unificando metodi e modelli utilizzati in diversi campi. Viene quindi sviluppato un modello predittivo anche chiamato Model Predictive Control (MPC) partendo da un modello a parametri concentrati linearizzato e utilizzando l’algoritmo di ottimizzazione, il quale minimizza la domanda di energia soddisfacendo dei vincoli imposti; una programmazione lineare (LP), infatti, permette di trovare la migliore soluzione possibile date delle condizioni specifiche. Vengono poi proposte varie tecniche di analisi dei dati a diversi livelli, al fine di calibrare il modello di cui sopra; i dati mensili vengono analizzati tramite modelli di regressione lineare mentre quelli orari con tecniche di identificazione (modelli ARX). L’identificazione dei parametri permette anche di analizzare l'evoluzione di questi nel tempo, e, in tal modo, pone le basi per una applicazione del sistema di apprendimento automatico (machine learning) che può portare ad un costante miglioramento delle prestazioni del sistema di controllo. Infine, il modello presentato, calcolando il Voto Medio Previsto (PMV), permette anche la valutazione del comfort interno; la minimizzazione dei fabbisogni energetici, infatti, non può prescindere da una valutazione della variazione del comfort dell’utente. Inoltre, la valutazione del PMV acquisisce un ruolo cruciale anche nella scelta dei limiti da imporre all'algoritmo di ottimizzazione. Infine, i vantaggi offerti da un sistema come quello presentato dipendono anche dal tipo di funzione obiettivo adottata; partendo dalla minimizzazione della domanda di energia, infatti, è possibile ridurre al minimo anche i costi (utilizzando tariffe dinamiche) e le emissioni (utilizzando serie temporali e fattori di emissione).
Building simulation models in control systems for energy efficiency
MARENZI, GIORGIA
Abstract
While new paradigms for the construction sector are emerging, such as that of “Nearly Zero Energy Building” and sustainability assessment schemes are progressively evolving and becoming more common, empirical studies show how, very often, the gap between the predicted (design phase) and measured (operation phase) energy performance could be very large. This issue is generally addressed with the term “performance gap” and it can be fundamentally caused by design phase, construction phase, commissioning and operational phase errors. This situation creates a problem of credibility in the building industry and, more in general, in sustainability practices because the errors committed can directly reflect on energy performance and, consequently on the running costs and on global cost optimality. Bridging the gap between simulated and real energy usage requires the presence of embedded control and fault detection applications that have to be adaptive, self-calibrating and conceived to enable a model based interpretation of measured data, for performance benchmarking. Considering these general issues, the original part of the research presented is given by the overall methodology that aims to address some of the fundamental problems present in monitoring and control systems. The methodology used for the research, therefore, seeks to unite these steps through the choice of effective tools, able to interact each other, trying to unify methods and models used in different fields and understand how to make them usable in an effective workflow across life cycle phases. Starting from a linearized lumped parameters model, an optimization algorithm is used to find the optimal operation trajectory that satisfies the constraints while minimizing energy demand. With a Linear Programming (LP), in fact, it is possible to find the best possible solution given the specific conditions (i.e. find a solution that is certainly a global optimum). Further, the optimization model, working on set-point violations, can be, in principle, simply added an additional layer on existing automation systems, without having to replace completely the technology, but simply working on the top of conventional PI/PID controllers. In the data analysis workflow presented, two main levels of model calibration are done; firstly, with monthly data the main lumped parameters are analyzed through regression models and, secondly, using hourly data, an identification technique (ARX model) is used to directly calibrate/built the whole model (suitable for multiple temporal scale of analysis) with the estimated parameters. Parameter identification allows also to analyze the evolution of parameters over time, and, thus, lay the foundations for a machine learning system application which may constantly improve the control system knowledge adapting and learning to building and end-users characteristics. The energy demand reduction, however, is not the only objective to be pursued. Internal environmental quality and energy savings can be conflicting objectives and comfort expectation can cause an “economic rebound effect”, limiting the energy saving potential of efficiency practices and, therefore, compromising the return of investment. The Predicted Mean Vote (PMV) evaluation acquires a crucial role in the choice of the limits to be set in the optimization algorithm. The latter in fact, needs a limit in terms of degree/days and degree/hours of comfort violation that can be chosen starting from a comfort assessment, done through the PMV evaluation. The advantages provided by a Model Predictive Control (MPC) system such as the one presented in this research are also dependent of the type of objective function adopted. Starting from the minimal energy demand, in fact, it is also possible to minimize costs (using dynamic tariffs) and emissions (using hourly CO2 emission time-series and emission factors) or considering a combination of both, using weighting factors. Further, it permits to increase the overall energy efficiency, considering both the load matching and the dispatch flexibility problem.File | Dimensione | Formato | |
---|---|---|---|
2017_03_PhD_Marenzi.pdf
accessibile in internet per tutti
Descrizione: Tesi PhD Completa
Dimensione
3.71 MB
Formato
Adobe PDF
|
3.71 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/132577