Solar Generation Forecasting by Recurrent Neural Networks Optimized by Levenberg-Marquardt Algorithm

• Published in: IECON 2018 - 44th Annual Conference of the IEEE Industrial Electronics Society pp. 276 - 281
• Main Authors: Awan, Shahid M., Khan, Zubair A., Aslam, Muhammad
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-10-2018
• Subjects: Artificial neural networks; Autoregressive processes; Forecasting; Levenberg-Marquardt Algorithm; Numerical models; Optimization Techniques; Predictive models; Recurrent Neural Networks; Short Term Forecasting; Solar Generation; Solar power generation; Weather forecasting
• ISSN: 2577-1647
• DOI: 10.1109/IECON.2018.8591799

Solar photovoltaic systems convert the solar energy into electric power. Many factors affect the solar power generation through solar cells. The factors include atmospheric conditions, trajectory of sun, weather conditions, cloud cover and the physical properties of solar energy plant that converts solar energy to electric power. The output power is mainly influenced by the incoming radiation and the characteristics of solar panel. Accurate and correct knowledge about these factors guarantees a reliable solar generation forecasting model. This paper proposes a solution for solar power generation forecasting by incorporating the effecting parameters with the use of Recurrent Neural Network (RNN)model. The RNN is further optimized by Levenberg-Marquardt Algorithm (LMA)to get better accuracy of forecasts. The obtained results confirm the suitability of proposed approach.