Photovoltaic Energy Anomaly Detection using Transformer Based Machine Learning

This study uses the Anomaly Transformer model to find anomalies in photovoltaic energy generation in Malang, Indonesia. The main background of this study is the lack of satellite monitoring in this region and the importance of annual data for electricity generation forecasting. Temperature scattered direct solar radiation, and hourly electricity production are all part of the dataset used which is only available since 2019. Anomalies were detected at 05.00 and 16.00 WIB, indicating instability in the energy supply due to high temperatures in the morning and heavy rain in the afternoon. Detection of these anomalies is important to improve the efficiency and reliability of photovoltaic systems, reduce operational costs, and reduce the risk of system failure. Indonesia has many challenges for photovoltaic energy generation due to its unique location, with many islands located close to the equator. The use of the Anomaly Transformer algorithm improves the accuracy of anomaly detection over conventional methods. This algorithm helps to find complex patterns in very large time series. The results show that the anomaly transformer model can effectively detect anomalous patterns. It offers ideas to improve the stability and efficiency of photovoltaic systems in Malang and other areas with comparable environmental conditions. Improved energy efficiency and environmental sustainability are the results of anomaly pattern detection.