The increasing integration of renewable energy sources into power grids introduces new challenges for voltage regulation, due to their intermittent nature and reduced inertia compared to traditional systems. This thesis proposes an innovative approach for Secondary Voltage Regulation (SVR) using Model Predictive Control (MPC), leveraging data-driven models to enhance grid stability and reliability. Three model identification techniques are analyzed and compared: DMDc (Dynamic Mode Decomposition with Control), SINDYc (Sparse Identification of Nonlinear Dynamical Systems with Control), and FNNs (Feedforward Neural Networks). The comparison was conducted through simulations in DIgSILENT PowerFactory on an IEEE 14-bus system, evaluating prediction error, robustness to noise, and generalization ability. Results show that SINDYc provides an interpretable and accurate model, while FNNs offer the best generalization at the cost of reduced model transparency. DMDc struggled to capture nonlinear dynamics effectively. Finally, the MPC based on the identified models successfully maintained voltage within acceptable limits, improving system stability.
L'integrazione crescente delle fonti rinnovabili nella rete elettrica introduce nuove sfide per la regolazione della tensione, a causa della loro natura intermittente e della ridotta inerzia rispetto ai sistemi tradizionali. Questa tesi propone un approccio innovativo per il controllo secondario della tensione (SVR) attraverso l'uso del Model Predictive Control (MPC), sfruttando modelli data-driven per migliorare la stabilità e l'affidabilità della rete. Tre metodi di identificazione dei modelli sono stati analizzati e confrontati: DMDc (Dynamic Mode Decomposition with Control), SINDYc (Sparse Identification of Nonlinear Dynamical Systems with Control) e FNNs (Feedforward Neural Networks). Il confronto è stato effettuato attraverso simulazioni in DIgSILENT PowerFactory su un sistema IEEE a 14 bus, valutando l'errore di predizione, la robustezza ai disturbi e la capacità di generalizzazione. I risultati mostrano che SINDYc fornisce un modello interpretabile e accurato, mentre FNNs garantisce la migliore generalizzazione a costo di una minore trasparenza del modello. DMDc, invece, ha mostrato limitazioni nell'adattarsi a dinamiche non lineari. Infine, l’MPC basato sui modelli stimati ha dimostrato la capacità di mantenere la tensione entro limiti accettabili, migliorando la stabilità del sistema.
Data-driven MPC for Secondary Voltage Control
Giampaglia, Martina
2024/2025
Abstract
The increasing integration of renewable energy sources into power grids introduces new challenges for voltage regulation, due to their intermittent nature and reduced inertia compared to traditional systems. This thesis proposes an innovative approach for Secondary Voltage Regulation (SVR) using Model Predictive Control (MPC), leveraging data-driven models to enhance grid stability and reliability. Three model identification techniques are analyzed and compared: DMDc (Dynamic Mode Decomposition with Control), SINDYc (Sparse Identification of Nonlinear Dynamical Systems with Control), and FNNs (Feedforward Neural Networks). The comparison was conducted through simulations in DIgSILENT PowerFactory on an IEEE 14-bus system, evaluating prediction error, robustness to noise, and generalization ability. Results show that SINDYc provides an interpretable and accurate model, while FNNs offer the best generalization at the cost of reduced model transparency. DMDc struggled to capture nonlinear dynamics effectively. Finally, the MPC based on the identified models successfully maintained voltage within acceptable limits, improving system stability.File | Dimensione | Formato | |
---|---|---|---|
2025_4_Giampaglia_Tesi.pdf
accessibile in internet per tutti
Descrizione: Tesi
Dimensione
2.63 MB
Formato
Adobe PDF
|
2.63 MB | Adobe PDF | Visualizza/Apri |
2025_4_Giampaglia_Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
1.03 MB
Formato
Adobe PDF
|
1.03 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/10589/235575