Application of the Auto-LSTM model for solar energy forecasting in photovoltaic power systems

• Grant number: 22/10281-6
• Support Opportunities: Scholarships in Brazil - Master
• Start date: June 01, 2023
• End date: November 30, 2024
• Field of knowledge: Interdisciplinary Subjects
• Principal Investigator: Fernando Ramos Martins
• Grantee: Fernando Vasconde de Arruda
• Host Institution: Instituto do Mar (IMar). Universidade Federal de São Paulo (UNIFESP). Campus Baixada Santista. Santos , SP, Brazil
• Associated research grant: 14/50848-9 - INCT 2014: INCT for Climate Change, AP.PFPMCG.TEM

Abstract

The way we exploit our planet has been discussed for decades, and using our resources more sustainably is one of the main changes. Electric energy is one of the primary resources used by our society, and technological development only increases its use. For decades, we have used fossil fuels emitting greenhouse gases into the atmosphere. The transition to renewable resources is the alternative to change this scenario. Treaties such as Paris and Glasgow and the 2030 agenda encourage countries to develop targets for using renewable sources and neutralize the emission of carbon dioxide in the atmosphere. Brazil, in turn, has a primarily renewable electrical matrix based primarily on the water resource; however, the dry periods showed the need to diversify the energy resources to achieve greater resilience and security in meeting the electrical demand. Solar energy is one of the most promising. Still, some challenges bring uncertainties regarding its intermittence caused by the atmospheric conditions, which drives the development of several studies to predict solar irradiation at different forecasting horizons. This project aims to use a hybrid deep learning methodology (Autoencoder + LSTM) to develop a solar resource prediction model applied to the IEE/USP photovoltaic system and, later, also in small plants in operation in the city of Santos (urban area in the coastal area of São Paulo state). Data observed in meteorological and environmental stations (INMET, CETESB, AERONET) located in metropolitan regions will be used to develop the proposed model. The study will also involve variables produced by operational numerical prediction models available for public access. The project's main contribution is developing a state-of-the-art solar resource prediction model for large urban centers where distributed generation should grow and play an essential role in the national electricity matrix.