In a long-term perspective, the need for an increasing energy utilization efficiency and a reduction of greenhouse gases emission has been leading the modern power system toward an optimized use of primary energy sources. Furthermore, progresses in the frameworks of Internet of Things, big data and cloud computing have been providing new opportunities for innovative applications in many fields, including the energy sector. Against such a background, this work focuses on the development of optimized tools for the operation scheduling of multi-energy systems, oriented at real-time control and profit maximization for the energy service provider. Starting from the analysis of an existent control system installed in a trigeneration plant of a hospital complex in the North-East of Italy, four different optimizers are implemented. Their function is regulating the simultaneous operations of cogenerators, boilers, electric chillers, and absorption chillers by maximizing the hourly profits for the service provider while meeting technical and contractual requirements. The first three optimizers are based on an instantaneous optimization which is respectively carried out via particle swarm optimization, genetic algorithms and mixed integer linear programming (MILP); on the contrary, the last one is based on MILP and performs a predictive rolling-horizon optimization. Predictions are implemented through the elaboration of deep-learning regression models for the forecast of power demands in the short term. At first, these models are developed as multi-layer perceptrons for one hour-ahead predictions, reaching coefficients of determination above 90% on average. After that, the models are extended by sequential computations to carry out day-ahead forecasts. The performance of each optimizer is assessed on four test cases that simulate the seasonal weekly operations of the trigeneration plant of the hospital complex in the North-East of Italy. The MILP predictive optimizer turns out to be the best performing optimized tool among the four developed in terms of total weekly profits for the energy service provider; concurrently, it is also found to require the highest average computational time per optimization process. However, being compliant with real-time quarter-hour regulation, the MILP predictive optimizer is adapted for the field implementation on the hospital complex in the North-East of Italy. Thereby, the phases of the intervention are outlined and described in detail; in parallel, the algorithm is modified with the addition of security procedures aiming at increasing its resilience and continuity of operation. Finally, through an economic assessment of the intervention, the seasonal profits for the energy service provider are estimated to rise by more than 25% thanks to the optimized control, in two different scenarios.
In una prospettiva a lungo termine, la necessità di aumentare l'efficienza energetica e ridurre le emissioni di gas serra sta conducendo il settore della produzione di potenza verso un uso ottimizzato delle fonti di energia primaria. Inoltre, i progressi nei campi dell'internet delle cose, dei big data e del cloud computing stanno creando nuove opportunità per applicazioni innovative in diversi settori, compreso quello energetico. In tale contesto, questo elaborato propone lo sviluppo di strumenti ottimizzati per la gestione di sistemi multi-energetici, orientato al controllo in tempo reale degli impianti e alla massimizzazione dei profitti per il fornitore di servizi. Partendo dall'analisi di un sistema di controllo installato in una centrale di trigenerazione di un ospedale nel Nord-Est Italia, sono implementati quattro diversi ottimizzatori. Il loro scopo è quello di regolare il funzionamento simultaneo di cogeneratori, caldaie, frigoriferi elettrici e ad assorbimento, massimizzando i profitti orari per il fornitore di servizi e rispettando requisiti tecnici e contrattuali. I primi tre ottimizzatori si basano su un'ottimizzazione istantanea, elaborata rispettivamente tramite ottimizzazione a sciame di particelle, algoritmi genetici e programmazione mista lineare intera (MILP); viceversa, l'ultimo è basato su MILP ma effettua un'ottimizzazione predittiva ad orizzonte flottante. La previsione è implementata tramite l'elaborazione di modelli di regressione basati su tecniche di apprendimento profondo per la stima dei fabbisogni energetici nel breve termine. Inizialmente, questi modelli sono sviluppati come percettori multistrato per effettuare previsioni con un'ora di anticipo, raggiungendo coefficienti di determinazione mediamente superiori al 90%. Successivamente, tali modelli sono estesi per elaborare previsioni con un giorno in anticipo tramite calcoli in successione. Le prestazioni di ciascun ottimizzatore sono valutate su quattro settimane tipo che simulano il funzionamento stagionale della centrale di trigenerazione dell'ospedale nel Nord-Est Italia. L'ottimizzatore MILP predittivo risulta essere il sistema di controllo migliore, fra i quattro sviluppati, in termini di profitti totali settimanali per il fornitore di servizi energetici; contemporaneamente, è anche lo strumento che richiede in media un maggior tempo computazionale per i processi di ottimizzazione. Tuttavia, essendo compatibile con una regolazione in tempo reale al quarto orario, l'ottimizzatore MILP predittivo è adattato per l'implementazione sul campo presso il complesso ospedaliero nel Nord-Est Italia. A tale scopo, le fasi dell'intervento vengono delineate e analizzate nel dettaglio; parallelamente, l'algoritmo viene modificato con l'aggiunta di procedure di sicurezza atte ad aumentarne la resilienza e la continuità di funzionamento. Infine, tramite un'analisi economica dell'intervento, viene stimato un aumento dei profitti stagionali per il fornitore di servizi energetici maggiore del 25% grazie al controllo ottimizzato, in due diversi scenari.
Analysis, development and field implementation of an optimized predictive control system for complex trigeneration plants
MALABARBA, RICCARDO
2018/2019
Abstract
In a long-term perspective, the need for an increasing energy utilization efficiency and a reduction of greenhouse gases emission has been leading the modern power system toward an optimized use of primary energy sources. Furthermore, progresses in the frameworks of Internet of Things, big data and cloud computing have been providing new opportunities for innovative applications in many fields, including the energy sector. Against such a background, this work focuses on the development of optimized tools for the operation scheduling of multi-energy systems, oriented at real-time control and profit maximization for the energy service provider. Starting from the analysis of an existent control system installed in a trigeneration plant of a hospital complex in the North-East of Italy, four different optimizers are implemented. Their function is regulating the simultaneous operations of cogenerators, boilers, electric chillers, and absorption chillers by maximizing the hourly profits for the service provider while meeting technical and contractual requirements. The first three optimizers are based on an instantaneous optimization which is respectively carried out via particle swarm optimization, genetic algorithms and mixed integer linear programming (MILP); on the contrary, the last one is based on MILP and performs a predictive rolling-horizon optimization. Predictions are implemented through the elaboration of deep-learning regression models for the forecast of power demands in the short term. At first, these models are developed as multi-layer perceptrons for one hour-ahead predictions, reaching coefficients of determination above 90% on average. After that, the models are extended by sequential computations to carry out day-ahead forecasts. The performance of each optimizer is assessed on four test cases that simulate the seasonal weekly operations of the trigeneration plant of the hospital complex in the North-East of Italy. The MILP predictive optimizer turns out to be the best performing optimized tool among the four developed in terms of total weekly profits for the energy service provider; concurrently, it is also found to require the highest average computational time per optimization process. However, being compliant with real-time quarter-hour regulation, the MILP predictive optimizer is adapted for the field implementation on the hospital complex in the North-East of Italy. Thereby, the phases of the intervention are outlined and described in detail; in parallel, the algorithm is modified with the addition of security procedures aiming at increasing its resilience and continuity of operation. Finally, through an economic assessment of the intervention, the seasonal profits for the energy service provider are estimated to rise by more than 25% thanks to the optimized control, in two different scenarios.| File | Dimensione | Formato | |
|---|---|---|---|
|
2019_10_Malabarba.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
8.79 MB
Formato
Adobe PDF
|
8.79 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/10589/149592