The value of grid impedance is not always known a priori and cannot always be considered negligible. However, it is important to determine it as it influences several aspects of the system, including the performance of control systems and the efficiency of power exchange. The objective of this study is to test the effectiveness of methods based on the use of neural networks trained using voltage and current measurements obtained through the simulation of a generic system connected to the electrical grid (modeled as an L-R series). Two distinct approaches are used, both based on feedforward neural networks, but differing in the nature of the input data: the first approach uses time-domain measurement periods in response to single-frequency perturbations, while the second uses the ratio of the frequency spectra of voltage and current in response to a broadband signal (PRBS). For both approaches, the procedure involves running numerous simulations for different pairs of R and L values, followed by transferring the data from MATLAB to Python, where the data is split into training and testing sets to train the network and verify the effectiveness of the results. Only feedforward network solutions have been investigated in detail; therefore, it could be interesting to compare the results obtained using different architectures.
Il valore dell'impedenza di rete non è sempre noto a priori, e non sempre può essere considerato trascurabile. Tuttavia è importante conoscerlo visto che influenza numerosi aspetti del sistema tra cui le prestazioni dei sistemi di controllo e l'efficienza nello scambio di potenza. l'obiettivo di questo studio è testare l'efficacia di metodi basati sull'utilizzo di reti neurali addestrate sfruttando le misure di tensione e corrente ottenute attraverso la simulazione di un generico sistema connesso alla rete elettrica (modellata come serie L-R). Si utilizzano due approcci distinti, entrambi basati su reti neurali feedforward, ma che differiscono nella natura dei dati usati come input: il primo utilizza dei periodi di misure nel tempo in risposta a perturbazioni a singola frequenza. mentre il secondo il rapporto degli spettri in frequenza di tensione e corrente in risposta ad un segnale a banda larga (PRBS). Per entrambi gli approcci la procedura consiste nel lancio di numerose simulazioni per diverse coppie di valori R,L, seguite dal trasferimento dei dati da matlab in python, dove vengono suddivisi in dati di train e test per poi effettuare l'addestramento della rete e verificare l'efficacia del risultato. Sono state investigate nel dettaglio le sole soluzioni con reti feedforward, potrebbe quindi essere interessante comparare i risultati ottenuti utilizzando architetture diverse.
Stima dell'impedenza di rete attraverso metodi basati su reti neurali
ALBERTI, PIETRO
2024/2025
Abstract
The value of grid impedance is not always known a priori and cannot always be considered negligible. However, it is important to determine it as it influences several aspects of the system, including the performance of control systems and the efficiency of power exchange. The objective of this study is to test the effectiveness of methods based on the use of neural networks trained using voltage and current measurements obtained through the simulation of a generic system connected to the electrical grid (modeled as an L-R series). Two distinct approaches are used, both based on feedforward neural networks, but differing in the nature of the input data: the first approach uses time-domain measurement periods in response to single-frequency perturbations, while the second uses the ratio of the frequency spectra of voltage and current in response to a broadband signal (PRBS). For both approaches, the procedure involves running numerous simulations for different pairs of R and L values, followed by transferring the data from MATLAB to Python, where the data is split into training and testing sets to train the network and verify the effectiveness of the results. Only feedforward network solutions have been investigated in detail; therefore, it could be interesting to compare the results obtained using different architectures.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_alberti_pietro_dtg.pdf
accesso aperto
Dimensione
5.1 MB
Formato
Adobe PDF
|
5.1 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/85289