This thesis proposes a novel approach to tackle the problem of Self-adaptation in the presence of Uncertainty in model selection in self-adaptive systems. These systems are designed to adapt to changing environmental conditions without human intervention. However, due to their complexity, building accurate models that can capture the wide range of environmental conditions that the system may encounter is challenging. This uncertainty can lead to suboptimal performance and even system failures. To address this issue, the proposed approach uses two techniques: Bayesian Model Averaging and Many Objective Search. The former is a statistical technique that allows for combining multiple models through a weighted average to obtain a more accurate and robust prediction. This technique can effectively handle the aforementioned Uncertainty. The latter is used to search for a new configuration of the self-adaptive systems that allow them to achieve an equilibrium condition, an optimal adaptation in a wide search space. To complete the adaptation, the proposed approach uses the Non-dominated Sorting Genetic Algorithm III (NSGA-III), a state-of-the-art many objective optimization algorithm that can handle conflicting objectives. A simulator is introduced to evaluate the effectiveness of the proposed approach using a case study from the robotics domain. The results show that the proposed idea is effective in detecting and enforcing equilibrium constraints during execution and improving the adaptation performance of self-adaptive systems.
Questa tesi propone un nuovo approccio per affrontare il problema dell'Auto-adattamento (Self-adaptation) in presenza di Incertezza (Uncertainty) nella selezione del modello nei sistemi auto-adattivi. Questi sistemi sono progettati per adattarsi alle mutevoli condizioni ambientali senza l'intervento umano. Tuttavia, a causa della loro complessità, è difficile costruire modelli accurati in grado di catturare la vasta gamma di condizioni ambientali che il sistema può incontrare. Questa incertezza può portare a prestazioni non ottimali e persino a guasti del sistema. Per affrontare questo problema, l'approccio proposto utilizza due tecniche: Bayesian Model Averaging e Many Objective Search. La prima è una tecnica statistica che consente di combinare, attraverso una media ponderata, più modelli per ottenere una previsione più accurata e robusta: questa tecnica può gestire efficacemente l'Incertezza de qua. La seconda viene utilizzata per cercare una nuova configurazione dei sistemi auto-adattivi, che ci consenta di raggiungere una condizione di equilibrio quale adattamento ottimale in un ampio spazio di ricerca. Per completare l'adattamento, l'approccio proposto utilizza il Non-dominated Sorting Genetic Algorithm III (NSGA-III), algoritmo di ottimizzazione multi-obiettivo all'avanguardia in grado di gestire più obiettivi in conflitto. Viene introdotto un simulatore per valutare l'efficacia dell'approccio proposto utilizzando un esempio del dominio robotico. I risultati mostrano che l'idea proposta è efficace nel rilevare e far rispettare i vincoli di equilibrio durante l'esecuzione e nel migliorare le prestazioni di adattamento dei sistemi auto-adattivi.
Self-adaptation under Uncertainty using Bayesian Model Averaging and Many objective Search
MESSUTI, UMBERTO
2022/2023
Abstract
This thesis proposes a novel approach to tackle the problem of Self-adaptation in the presence of Uncertainty in model selection in self-adaptive systems. These systems are designed to adapt to changing environmental conditions without human intervention. However, due to their complexity, building accurate models that can capture the wide range of environmental conditions that the system may encounter is challenging. This uncertainty can lead to suboptimal performance and even system failures. To address this issue, the proposed approach uses two techniques: Bayesian Model Averaging and Many Objective Search. The former is a statistical technique that allows for combining multiple models through a weighted average to obtain a more accurate and robust prediction. This technique can effectively handle the aforementioned Uncertainty. The latter is used to search for a new configuration of the self-adaptive systems that allow them to achieve an equilibrium condition, an optimal adaptation in a wide search space. To complete the adaptation, the proposed approach uses the Non-dominated Sorting Genetic Algorithm III (NSGA-III), a state-of-the-art many objective optimization algorithm that can handle conflicting objectives. A simulator is introduced to evaluate the effectiveness of the proposed approach using a case study from the robotics domain. The results show that the proposed idea is effective in detecting and enforcing equilibrium constraints during execution and improving the adaptation performance of self-adaptive systems.| File | Dimensione | Formato | |
|---|---|---|---|
|
Messuti_Executive_Summary.pdf
accessibile in internet per tutti
Dimensione
820.77 kB
Formato
Adobe PDF
|
820.77 kB | Adobe PDF | Visualizza/Apri |
|
Messuti_Thesis.pdf
accessibile in internet per tutti
Dimensione
3.21 MB
Formato
Adobe PDF
|
3.21 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/10589/202878