DEEP LEARNING FOR SOLAR GHI FORECASTING IN ISLAMABAD

Abstract

The growing threat of global climate change stemming from the huge carbon footprint left behind by fossil fuels has prompted interest in exploring and utilizing renewable energy resources. Solar energy being one of the most abundant sources of green energy has attracted huge research attention over the years. There are numerous technologies available to resource solar energy and convert in others energy forms such as electric and thermal. Solar energy can be converted to electric and thermal energy by establishing solar thermal and solar photovoltaic plants and grids. These plants can facilitate whole cities and industries but the intermittent nature of solar energy which depends on a number of weather variables like cloud covers, humidity, rain, wind speed etc. These factors make it a challenge to manage solar power grids. The electric or thermal power output depends on majorly on solar Global Horizontal Irradiance (GHI). To manage large solar photovoltaic grids and plan on power distribution future forecasts must be made to ensure smooth grid operations and prevent power outages and load shedding. Several statistical, ML and DL techniques exist and have been used for many years for a range forecasting problem including sales management, power management, finance and many more. Many first world countries use these methods to plan for the management of their industries, goods distribution, and infrastructure management. This study uses statistical and Deep Learning methods to forecast solar GHI in the city of Islamabad which not only helps in grid management and power distribution, but also brings attention towards the potential solar power production in Pakistan and its part to play in tackling global climate change. In this study, we present statistical methods namely Seasonal Auto-Regressive Integrated Moving ix Average Exogenous (SARIMAX) and Prophet, and Machine Learning methods such as Artificial Neural Networks (ANN), Convolution Neural Networks (CNN) and Recurrent Neural Networks (RNN). We have used Long Short-Term Memory (LSTM) variant of Recurrent Neural Networks. The selection forecast methods in our study are based on their ability to work with time series data. For our forecasting problem we have used the weather data that had been collected for four years and 9 months with precise instruments stationed in Islamabad. Our forecast problem and the data recorded, both are of time series in nature, and we have used different models with different configurations to see which performs best for our dataset. The performance of every model is studied using different error metric such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE), and Coefficient of Determination (R2).