This thesis focuses on anomaly detection in time series data extracted from Milan Silla2 Waste-to-Energy plant, operated by A2A Ambiente. The OpSys project aims to optimize the plant’s energy management system in compliance with the ISO 50001 standard and to support the preparation of the energy audit required by Legislative Decree 102/2014. To achieve these objectives, real-time Energy Performance Indicators (EnPIs) proposed by Utilitalia are developed. These indicators allow for real-time monitoring of energy consumption across the plant’s various sections and enable the detection of anomalous system behaviors through advanced anomaly detection models. The study introduces the concept of time series and explores unsupervised methods for anomaly detection, ranging from classical techniques like Isolation Forest to more recent neural network-based approaches tailored for this purpose. The adopted methodology is structured into three phases: the application of anomaly detection models to the EnPIs of the plant’s sections, a focused analysis of individual lines and energy-consuming units within each section, and, finally, the proposal of corrective measures followed by an assessment of the resulting benefits. The findings are discussed in relation to three energy performance indicators associated with the plant’s main sections: the steam generator, the flue gas treatment and the thermal cycle. Lastly, the thesis outlines the next steps of the project, which involve expanding the analysis and developing a dashboard to present the results.
Questa tesi affronta il tema del rilevamento delle anomalie nelle serie temporali estratte dall’impianto di termovalorizzazione Milano Silla2, gestito da A2A Ambiente. Il progetto OpSys ha l’obiettivo di ottimizzare il sistema di gestione dell’energia del termovalorizzatore, in conformità con la norma ISO 50001, e di supportare la redazione della diagnosi energetica richiesta dal Decreto Legislativo 102/2014. A tal fine, vengono sviluppati gli indicatori di prestazione energetica (EnPI), proposti da Utilitalia e calcolati in tempo reale. Gli indicatori consentono il monitoraggio continuo dei consumi energetici delle diverse sezioni dell’impianto, nonchè l’identificazione di sistemi devianti dalle normali condizioni di funzionamento attraverso l’applicazione di modelli avanzati per il rilevamento delle anomalie. Dopo un’introduzione al concetto matematico di serie temporali, la tesi esplora metodi non supervisionati per il rilevamento delle anomalie, partendo dalle tecniche classiche di machine learning fino ad arrivare ad approcci più avanzati basati su reti neurali, opportunamente adattati per questo specifico contesto. La metodologia adottata si articola in tre fasi: applicazione di modelli di rilevamento delle anomalie agli EnPI delle sezioni dell’impianto, analisi approfondita delle singole linee e utenze all’interno di ciascuna sezione e, infine, la proposta di misure correttive e la conseguente valutazione dei benefici. I risultati vengono discussi in relazione a tre indicatori di prestazione energetica associati alle principali sezioni dell’impianto: il generatore di vapore, il trattamento dei fumi e il ciclo termico. Infine, la tesi delinea i prossimi passi del progetto, che prevedono l’espansione dell’analisi e lo sviluppo di una dashboard per visualizzare i risultati.
Anomaly detection in time series data: an application to the energy consumption of Silla2 waste-to-energy plant
Pagani, Carlotta
2023/2024
Abstract
This thesis focuses on anomaly detection in time series data extracted from Milan Silla2 Waste-to-Energy plant, operated by A2A Ambiente. The OpSys project aims to optimize the plant’s energy management system in compliance with the ISO 50001 standard and to support the preparation of the energy audit required by Legislative Decree 102/2014. To achieve these objectives, real-time Energy Performance Indicators (EnPIs) proposed by Utilitalia are developed. These indicators allow for real-time monitoring of energy consumption across the plant’s various sections and enable the detection of anomalous system behaviors through advanced anomaly detection models. The study introduces the concept of time series and explores unsupervised methods for anomaly detection, ranging from classical techniques like Isolation Forest to more recent neural network-based approaches tailored for this purpose. The adopted methodology is structured into three phases: the application of anomaly detection models to the EnPIs of the plant’s sections, a focused analysis of individual lines and energy-consuming units within each section, and, finally, the proposal of corrective measures followed by an assessment of the resulting benefits. The findings are discussed in relation to three energy performance indicators associated with the plant’s main sections: the steam generator, the flue gas treatment and the thermal cycle. Lastly, the thesis outlines the next steps of the project, which involve expanding the analysis and developing a dashboard to present the results.File | Dimensione | Formato | |
---|---|---|---|
2025_04_Pagani_Tesi.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
19.3 MB
Formato
Adobe PDF
|
19.3 MB | Adobe PDF | Visualizza/Apri |
2025_04_Pagani_ExecutiveSummary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
5.01 MB
Formato
Adobe PDF
|
5.01 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/10589/234542