The international actions carried out to reduce atmospheric emissions, energy independence and plant efficiency, constitute the basis of energy policy for each industrialized state. In order for it to be possible to reduce, and finally eliminate completely the production of energy from fossil sources, a series of objectives and targets have been set at European and national level. The increase of RES plants and their progressive replacement of traditional production plants, determine a series of challenges for the national electricity transmission system which finds itself having to face operating conditions for which it was not initially designed. There are numerous actions that take place on a daily basis on the electrical transmission system and, it is essential to have real-time visibility of the main electrical quantities both at the HV network nodes and at its interface points with the MV networks. An accurate forecast of the energy levels that characterize the electricity system is therefore a fundamental requirement for correct operational planning. In the present thesis work, the methods tested are proposed to make an estimate of the reactive power, at the nodes of the transmission network, using neural networks that use as input, the active power relative to the node under examination and other types of variables. The neural networks thus trained could be subsequently used in different ways, including, for example, the forecast of reactive power, which can be carried out if an accurate forecast of active power is available. It is therefore a new approach, known to the author, never tested in the scientific field, aimed at verifying whether the neural networks, trained using as input information on the past of active power and other non-electrical quantities, are able to make estimates reactive power with sufficient accuracy to be used as an input for other grid calculations.
Le azioni internazionali effettuate per la riduzione delle emissioni in atmosfera, l’indipendenza energetica e l’efficienza degli impianti, costituiscono, per ogni Stato industrializzato, le basi della politica energetica. Affinché sia possibile ridurre, ed infine, eliminare completamente la produzione di energia da fonti fossili, una serie di obiettivi e target sono stati fissati a livello europeo e nazionale. L’incremento di impianti FER e la loro progressiva sostituzione agli impianti di produzione tradizionali, determinano una serie di sfide per il sistema elettrico di trasmissione nazionale che, si ritrova a dover affrontare condizioni di funzionamento per le quali non era stato inizialmente pensato. Numerose sono le azioni che giornalmente si svolgono sul sistema elettrico di trasmissione e, fondamentale risulta essere, avere una visibilità in tempo reale delle principali grandezze elettriche sia ai nodi di rete AT che ai suoi punti di interfaccia con le reti MT. Una previsione accurata dei livelli energetici che caratterizzano il sistema elettrico, è dunque un requisito fondamentale per una corretta pianificazione operativa. Nel presente lavoro di tesi, i metodi sperimentati si propongono di effettuare una stima della potenza reattiva, ai nodi della rete di trasmissione, impiegando reti neurali che utilizzano come input, la potenza attiva relativa al nodo in esame e altre tipologie di variabili. Le reti neurali così addestrate, potrebbero essere successivamente utilizzate in diversi modi, fra i quali, ad esempio, il forecast della potenza reattiva, effettuabile se si ha a disposizione un forecast accurato di potenza attiva. Si tratta quindi di un nuovo approccio, a conoscenza dell’autore, mai testato in ambito scientifico, atto a verificare se le reti neurali, addestrate utilizzando in input informazioni sul passato della potenza attiva ed altre grandezze non elettriche, siano in grado di effettuare stime di potenza reattiva con un’accuratezza sufficiente per essere utilizzate come input per altri calcoli di rete.
ANALISI DI CORRELAZIONE E RETI NEURALI ARTIFICIALI PER LA STIMA DELLA POTENZA REATTIVA AI NODI DELLA RETE DI TRASMISSIONE TERNA
D'AMICO, GIUSEPPE
2019/2020
Abstract
The international actions carried out to reduce atmospheric emissions, energy independence and plant efficiency, constitute the basis of energy policy for each industrialized state. In order for it to be possible to reduce, and finally eliminate completely the production of energy from fossil sources, a series of objectives and targets have been set at European and national level. The increase of RES plants and their progressive replacement of traditional production plants, determine a series of challenges for the national electricity transmission system which finds itself having to face operating conditions for which it was not initially designed. There are numerous actions that take place on a daily basis on the electrical transmission system and, it is essential to have real-time visibility of the main electrical quantities both at the HV network nodes and at its interface points with the MV networks. An accurate forecast of the energy levels that characterize the electricity system is therefore a fundamental requirement for correct operational planning. In the present thesis work, the methods tested are proposed to make an estimate of the reactive power, at the nodes of the transmission network, using neural networks that use as input, the active power relative to the node under examination and other types of variables. The neural networks thus trained could be subsequently used in different ways, including, for example, the forecast of reactive power, which can be carried out if an accurate forecast of active power is available. It is therefore a new approach, known to the author, never tested in the scientific field, aimed at verifying whether the neural networks, trained using as input information on the past of active power and other non-electrical quantities, are able to make estimates reactive power with sufficient accuracy to be used as an input for other grid calculations.I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/83139