Time series forecasting using transformer or solar power prediction.

Abstract

Since its inception by Vaswani et al. [31] in 2017, the Transformer model has revolutionized various fields ,includin gnatura llanguag eprocessing ,compute rvision ,audi odat aprocessing, etc. The aim of this project is to investigate the potential of Transformers for solar power production prediction. This task, though challenging, has very interesting practical applications, including grid management, proactive maintenance planning, fault detection, fault prediction, etc. Our main goal of this Master project is to design a transformer model and optimize its parameters for solar power production forecasting and fault prediction tasks. The key features of our proposed Transformer model include relying on a encoder only architecture, integrating static positional encoding, and incorporating the Kolmogorov-Arnold Network as a feed-forward part. Experimental results indicate that our Transformer model achieves Outstanding performance in both anomaly detection and forecasting tasks. Specifically, the Transformer model excels in capturing long-range dependencies in time series data, leading to more accurate and reliable predictions. It significantl youtperforms the LSTM model in several key metrics, demonstrating its superior ability to handle the complexities inherent in solar power production data. This superior performance underscores the transformative potential of the Transformer model in enhancing the accuracy and efficiency of solar power prediction, making it a vital tool for improving grid management, optimizing maintenance schedules, and advancing fault detection mechanisms.