Solar photovoltaic (PV) power forecasting is a crucial aspect of efficient energy management in the renewable energy sector. This thesis explores the application of various types of artificial neural networks (ANNs) for predicting PV power output by considering various variables that affect the power output. The proposed ANNs are used to forecast the output power for different PV technologies while considering different prediction horizons. Additionally, the impact of panel ageing is investigated using different machine learning models. To evaluate the performance of the proposed ANNs, real-world PV power data was collected and preprocessed. The preprocessed data was then used to train and test different ANNs, including recurrent neural networks, autoencoders and convolutional neural networks. The experimental results show that the proposed ANNs can accurately pre- dict PV power output, with LSTM demonstrating the best performance for short-term forecasting. Furthermore, the impact of panel ageing on PV power was analyzed using different machine learning models, including linear regression and predictive analysis. The results show that the machine learning models can effectively predict the degradation of PV panel performance over time. To improve the accuracy of predictions, the effects of splitting the dataset into two dis- tinct datasets - sunny and cloudy - is investigated. Furthermore, a separate prediction model is utilized for each of these datasets. The results indicate that clustering the dataset leads to improved prediction accuracy. Overall, this thesis provides a compre- hensive analysis of the application of different ANNs for solar PV power forecasting and the impact of panel ageing on PV power output. The results demonstrate the potential of using machine learning techniques for accurate and reliable solar PV power forecasting.
La previsione dell’energia solare fotovoltaica (PV) è un aspetto cruciale per una gestione efficiente dell’energia nel settore delle energie rinnovabili. Questa tesi esplora l’applicazione di vari tipi di reti neurali artificiali (ANN) per prevedere l’output dell’energia solare PV, considerando diverse variabili che influenzano l’output energetico. Le ANN proposte vengono utilizzate per prevedere l’output energetico per diverse tecnologie PV, considerando diversi orizzonti di previsione. Inoltre, viene analizzato l’impatto dell’invecchiamento dei pannelli utilizzando diversi modelli di apprendimento automatico. Per valutare le prestazioni delle ANN proposte, sono stati raccolti e preelaborati dati reali sull’energia solare PV. I dati preelaborati sono stati quindi utilizzati per addestrare e testare diverse ANN, tra cui reti neurali feedforward, reti neurali ricorrenti e reti neurali convoluzionali. I risultati sperimentali mostrano che le ANN proposte possono prevedere con precisione l’output dell’energia solare PV, con il modello LSTM che offre le migliori prestazioni per la previsione a breve termine. Inoltre, l’impatto dell’invecchiamento dei pannelli sull’output dell’energia solare PV è stato analizzato utilizzando diversi modelli di apprendimento au- tomatico, tra cui la regressione lineare e l’analisi predittiva. I risultati mostrano che i modelli di apprendimento automatico possono prevedere in modo efficace il degrado delle prestazioni dei pannelli PV nel tempo. Per migliorare l’accuratezza delle previsioni, viene studiato l’impatto della suddivisione del dataset in due sottodataset: soleggiato e nuvoloso. Inoltre, viene utilizzato un modello di previsione dedicato per ciascun sottodataset. I risultati indicano che la suddivisione del dataset porta a un miglioramento dell’accuratezza delle previsioni. In generale, questa tesi fornisce un’analisi esaustiva dell’applicazione di diverse ANN per la previsione dell’energia solare PV e dell’impatto dell’invecchiamento dei pannelli sull’output dell’energia PV. I risultati dimostrano il potenziale dell’utilizzo delle tecniche di apprendimento automatico per una previsione accurata e affidabile dell’energia solare PV.
Solar PV power forecasting using machine learning
DHINGRA, SALONI
2022/2023
Abstract
Solar photovoltaic (PV) power forecasting is a crucial aspect of efficient energy management in the renewable energy sector. This thesis explores the application of various types of artificial neural networks (ANNs) for predicting PV power output by considering various variables that affect the power output. The proposed ANNs are used to forecast the output power for different PV technologies while considering different prediction horizons. Additionally, the impact of panel ageing is investigated using different machine learning models. To evaluate the performance of the proposed ANNs, real-world PV power data was collected and preprocessed. The preprocessed data was then used to train and test different ANNs, including recurrent neural networks, autoencoders and convolutional neural networks. The experimental results show that the proposed ANNs can accurately pre- dict PV power output, with LSTM demonstrating the best performance for short-term forecasting. Furthermore, the impact of panel ageing on PV power was analyzed using different machine learning models, including linear regression and predictive analysis. The results show that the machine learning models can effectively predict the degradation of PV panel performance over time. To improve the accuracy of predictions, the effects of splitting the dataset into two dis- tinct datasets - sunny and cloudy - is investigated. Furthermore, a separate prediction model is utilized for each of these datasets. The results indicate that clustering the dataset leads to improved prediction accuracy. Overall, this thesis provides a compre- hensive analysis of the application of different ANNs for solar PV power forecasting and the impact of panel ageing on PV power output. The results demonstrate the potential of using machine learning techniques for accurate and reliable solar PV power forecasting.File | Dimensione | Formato | |
---|---|---|---|
2023_07_Dhingra_Thesis_01.pdf
accessibile in internet per tutti
Descrizione: Thesis
Dimensione
26.91 MB
Formato
Adobe PDF
|
26.91 MB | Adobe PDF | Visualizza/Apri |
2023_07_Dhingra_Executive Summary_02.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
4.92 MB
Formato
Adobe PDF
|
4.92 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/10589/210727