Cyber-Physical Systems (CPS) are integrated systems that combine digital and physical components, applied in key sectors such as manufacturing, transportation, healthcare, and energy. A crucial challenge in managing CPS is accurately modeling their internal, often nonlinear, dynamics. Stochastic Hybrid Automaton (SHA) is employed to model such CPS, providing a comprehensive model for capturing both discrete state change and continuous dynamics. One approach to learning these SHA is the $L^{\ast}_{\text{SHA}}$ algorithm. It learns the structure of SHA by fitting a predefined set of Ordinary Differential Equations (ODEs) to observed data. While powerful, the method faces limitations when dealing with unknown systems, particularly due to its reliance on a predefined set of ODEs for fitting. This thesis overcomes this limitation by introducing Sparse Identification of Nonlinear Dynamics (SINDy). SINDy enables the automatic derivation of parsimonious and interpretable models from observational data, enhancing the understanding and control of CPS. The integration of SINDy with $L^{\ast}_{\text{SHA}}$ demonstrates improved modeling and control of CPS, offering a robust framework for managing the complexity of these systems. These advanced models offer significant improvements in the control and management of CPS compared to superficial data analysis, opening new perspectives for optimizing these complex integrated systems. Finally, the thesis discusses the importance of clean data for precise model learning and proposes future developments, including the integration of advanced machine learning techniques, such as Artificial Neural Networks (ANNs) and Reinforcement Learning (RL), with SINDy aiming to create a more robust framework.
I "Cyber-Physical Systems" (sistemi ciber-fisici, CPS) sono sistemi integrati che combinano componenti digitali e fisici, applicati in settori chiave come la produzione, il trasporto, la sanità e l'energia. Una sfida cruciale nella gestione dei CPS è la modellazione accurata delle loro dinamiche interne, spesso non lineari. Per modellare tali CPS vengono impiegati gli "Stochastic Hybrid Automata" (automi ibridi stocastici, SHA), che forniscono un modello completo per catturare sia i cambiamenti di stato discreti che le dinamiche continue. Un approccio per apprendere questi SHA è l'algoritmo $L^{\ast}_{\text{SHA}}$, che apprende la struttura degli SHA adattando un insieme predefinito di "Ordinary Differential Equations" (equazioni differenziali ordinarie, ODEs) ai dati osservati. Pur essendo potente, il metodo presenta limitazioni quando si trattano sistemi sconosciuti, in particolare a causa della sua dipendenza da un insieme predefinito di ODE per il fitting. Per superare questa limitazione, questa tesi introduce la "Sparse Identification of Nonlinear Dynamics" (identificazione sparsa delle dinamiche non lineari, SINDy). SINDy consente la derivazione automatica di modelli parsimoniosi e interpretabili dai dati osservati, migliorando la comprensione e il controllo dei CPS. L'integrazione di SINDy con $L^{\ast}_{\text{SHA}}$ dimostra un miglioramento nella modellazione e nel controllo dei CPS, offrendo un solido quadro per gestire la complessità di questi sistemi. Questi modelli avanzati offrono significativi miglioramenti nel controllo e nella gestione dei CPS rispetto a un'analisi superficiale dei dati, aprendo nuove prospettive per l'ottimizzazione di questi complessi sistemi integrati. Infine, la tesi discute l'importanza dei dati puliti per un apprendimento preciso dei modelli e propone sviluppi futuri, inclusa l'integrazione di tecniche avanzate di machine learning, come "Artificial Neural Networks" (reti neurali artificiali, ANNs) e "Reinforcement Learning" (RL) con SINDy, al fine di creare un framework più robusto.
Improved learning of automata models for cyber-physical systems
Leone, Simone
2023/2024
Abstract
Cyber-Physical Systems (CPS) are integrated systems that combine digital and physical components, applied in key sectors such as manufacturing, transportation, healthcare, and energy. A crucial challenge in managing CPS is accurately modeling their internal, often nonlinear, dynamics. Stochastic Hybrid Automaton (SHA) is employed to model such CPS, providing a comprehensive model for capturing both discrete state change and continuous dynamics. One approach to learning these SHA is the $L^{\ast}_{\text{SHA}}$ algorithm. It learns the structure of SHA by fitting a predefined set of Ordinary Differential Equations (ODEs) to observed data. While powerful, the method faces limitations when dealing with unknown systems, particularly due to its reliance on a predefined set of ODEs for fitting. This thesis overcomes this limitation by introducing Sparse Identification of Nonlinear Dynamics (SINDy). SINDy enables the automatic derivation of parsimonious and interpretable models from observational data, enhancing the understanding and control of CPS. The integration of SINDy with $L^{\ast}_{\text{SHA}}$ demonstrates improved modeling and control of CPS, offering a robust framework for managing the complexity of these systems. These advanced models offer significant improvements in the control and management of CPS compared to superficial data analysis, opening new perspectives for optimizing these complex integrated systems. Finally, the thesis discusses the importance of clean data for precise model learning and proposes future developments, including the integration of advanced machine learning techniques, such as Artificial Neural Networks (ANNs) and Reinforcement Learning (RL), with SINDy aiming to create a more robust framework.File | Dimensione | Formato | |
---|---|---|---|
2024_09_Leone_Tesi.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Testo della tesi
Dimensione
3.02 MB
Formato
Adobe PDF
|
3.02 MB | Adobe PDF | Visualizza/Apri |
2024_09_Leone_Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: Testo dell'Executive Summary
Dimensione
1.15 MB
Formato
Adobe PDF
|
1.15 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/10589/226949