การพยากรณ์กำลังผลิตไฟฟ้าจากพลังงานแสงอาทิตย์เพื่อปรับปรุงประสิทธิภาพของโรงไฟฟ้าเสมือน

Abstract

Solar Photovoltaic (Solar PV) power generation is rapidly expanding. However, the preservation of grid stability poses ongoing challenges. Virtual Power Plant (VPP) is essential in addressing these issues by optimizing power generation and distribution. Additionally, short-term forecasting using Deep Learning (DL) can enhance VPP efficiency and reduce grid uncertainty. This thesis proposes a short-term solar power generation forecasting model to enhance the efficiency of VPPs, named “1D CNN-GRU.” The model combines a 1-dimensional Convolutional Neural Network (1D CNN) with a Gated Recurrent Unit (GRU). The 1D CNN extracts key features from time-series data, while the GRU improves short-term prediction accuracy. Additionally, data pre-processing techniques such as feature selection, data smoothing, and data augmentation are employed to enhance the performance of the proposed 1D CNN-GRU model. The proposed model was compared to other models such as CNN, GRU, Long Short-Term Memory (LSTM), and CNN-GRU to evaluate the performance. This comparative analysis utilized data from the hydro-floating solar installation at Sirindhorn Dam in Ubon Ratchathani province, Thailand. The model was tested by testing data from three seasons in Thailand. The results indicate that the proposed model achieved the lowest RMSE in all seasons: 0.025 in winter, 0.050 in summer, and 0.094 in the rainy season, with the shortest training time of 1,038.60 seconds. This indicates that the proposed model outperforms others in terms of efficiency and reduced training time.