The objective of this thesis is to investigate the application of the Virtual Reference Feedback Tuning (VRFT) method for control of nonlinear Single-Input-Single-Output (SISO) systems using regulators with the structure of Recurrent Neural Networks (RNNs). In particular, the Echo State Networks (ESN) and Long Short Term Memory (LSTM) networks are considered. In the first part of the thesis, the ESNs and LSTMs are recalled and their use as regulators is discussed. Then, the VRFT method for the direct nonlinear control design is presented and its application with RNN regulators is specified. Secondly, some advanced control schemes that make use of VRFT with RNN regulators are described. More specifically, it is outlined how an explicit integral action can be embedded in the scheme and how RNN regulators can be used to constrain the control variable. Finally, the notable advantages of the application of anti-windup blocks in these schemes are pointed out. In conclusion, the previous approach is tested to the pH process and to the single-track vehicle model. Furthermore, a comparison with other linear and nonlinear regulators is performed.
L'obiettivo di questa tesi è quello di studiare l'applicazione del Virtual Reference Feedback Tuning (VRFT) al controllo di sistemi Single-Input-Single-Output (SISO) non lineari che utilizzano regolatori con la struttura di Reti Neurali Ricorrenti (RNN). In particolare, vengono prese in considerazione le reti Echo State Networks (ESN) e Long Short Term Memory (LSTM). Nella prima parte della tesi, vengono richiamate le ESNs e LSTMs ed è discusso il loro uso come regolatori. Quindi, viene presentato il metodo VRFT per la sintesi diretta dei regolatori per sistemi non lineari e viene specificata la sua applicazione con regolatori RNN. Successivamente, vengono descritti alcuni schemi di controllo avanzati che utilizzano VRFT con regolatori RNN. Più nello specifico, viene spiegato come un'azione esplicita integrale può essere inclusa nello schema e come i regolatori RNN possono essere utilizzati per vincolare la variabile di controllo. Infine, vengono evidenziati i notevoli vantaggi dell'applicazione di blocchi anti-windup con questi schemi. In conclusione, l'approccio precedente è validato attraverso l'applicazione al processo del pH e al modello di veicoli di tipo single-track. Inoltre, viene svolto un confronto con regolatori lineari e non lineari alternativi.
Direct nonlinear control design : virtual reference feedback tuning with recurrent neural networks
D'AMICO, WILLIAM
2018/2019
Abstract
The objective of this thesis is to investigate the application of the Virtual Reference Feedback Tuning (VRFT) method for control of nonlinear Single-Input-Single-Output (SISO) systems using regulators with the structure of Recurrent Neural Networks (RNNs). In particular, the Echo State Networks (ESN) and Long Short Term Memory (LSTM) networks are considered. In the first part of the thesis, the ESNs and LSTMs are recalled and their use as regulators is discussed. Then, the VRFT method for the direct nonlinear control design is presented and its application with RNN regulators is specified. Secondly, some advanced control schemes that make use of VRFT with RNN regulators are described. More specifically, it is outlined how an explicit integral action can be embedded in the scheme and how RNN regulators can be used to constrain the control variable. Finally, the notable advantages of the application of anti-windup blocks in these schemes are pointed out. In conclusion, the previous approach is tested to the pH process and to the single-track vehicle model. Furthermore, a comparison with other linear and nonlinear regulators is performed.File | Dimensione | Formato | |
---|---|---|---|
2019_12_D'Amico.pdf
accessibile in internet per tutti
Descrizione: Testo della tesi
Dimensione
2.88 MB
Formato
Adobe PDF
|
2.88 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/10589/152590