Solar Energy Forecast for Integration of Grid and Balancing Power Using Profound Learning

Abstract

The rapid and unexpected advancements in solar photovoltaic (PV) technology pose a future challenge for power sector experts responsible for managing the distribution of electricity, given the technology’s direct reliance on atmospheric and weather conditions. Therefore, the development of reliable predictive models for short-term solar PV generation forecasts becomes critically important to maintain a stable power supply and ensure seamless grid operations. With the evolution of deep learning and its intricate models, its application in this domain offers a more efficient means of achieving precise forecasts. As a result, the proposed system undergoes the following stages: a) Collecting data from the Sky Images and Photovoltaic Power Generation Dataset (SKIPDD) hosted on a GitHub repository, which contains one-minute intervals of 64x64 sky images and concurrent PV power generation data. b) Enhancing the PV input data through processes such as geometric correction, ortho rectification, pan sharpening, block adjustment, and histogram equalization. c) Extracting PV-related features from these images using an Autoencoder. d) forecasting using integration of CNNbased Bi-LSTM. Experimental evaluation states that the proposed system (ACNN-BiLSTM) outperforms better on various measures (accuracy:0.95, MSE:0.08, MAE: 0.02).

Keywords

• bi-lstm
• cnn
• deep learning
• forecast model
• grid
• photovoltaic power
• solar power