Artificial Intelligence Powered Sustainability in Solar and Wind Hybrid Energy Systems

Abstract

Increasing global energy demand and environmental concerns drive the shift to sustainable alternatives. Solar and wind energy, with their eco-friendly attributes and abundant availability, emerge as key contenders. However, effectively harnessing renewable energy faces challenges due to variability, intermittency, and the need to adapt energy systems to diverse environments. Engineers and planners face the unpredictability inherent in renewable resources influenced by weather conditions, seasonal changes, and geographical variations. As renewable energy grows in the energy mix, integrating fluctuating sources into the grid becomes complex, necessitating energy storage and grid management solutions. Pakistan aims for 16% solar and wind energy by 2040. Technological advancements, including artificial Intelligence (AI), Machine Learning (ML), and improved weather forecasting, enhance renewable energy predictions. Accurate forecasting is crucial for a stable power supply, requiring sophisticated models. Supply side forecasting, fundamental for energy planning, faces challenges due to the unpredictability of solar and wind. Various forecasting techniques, including Transformer, Long Short Term Memory (LSTM), Support Vector Regression (SVR), and Linear Regression (LR), have been explored to improve the accuracy of renewable energy forecasts. The thesis conducts two case studies of energy forecast in Islamabad. The first study focuses on wind speed prediction using LR, SVR, and LSTM models. LSTM emerges as the most effective, achieving 78% accuracy for a 2-day wind speed forecast. Mean absolute error (MAE) serves as the performance metric. Combining techniques optimize prediction accuracy, facilitating the integration of more renewable energy into the grid. Addressing intermittency, storing excess energy as hydrogen is proposed, 6.76 kg estimated hydrogen production per day using wind energy and Proton Exchange Membrane (PEM). Similarly in the second study, hybrid solar and wind energy systems exhibit similar trends, inspiring the exploration of alternative hybrid solutions. The Transformer model predicts energy production, achieving 90.7% accuracy for solar irradiance and 90.45% for wind speed. Additionally, the analysis of model behavior unveiled that the R2 score exhibited a direct correlation with the look-back period and epochs, while demonstrating an inverse relationship with training data, horizon, and learning rate. xiii In conclusion, the global shift to sustainable energy, driven by rising demand and environmental concerns, face challenges in efficiently harnessing renewable energy due to its variability. As countries like Pakistan aim to integrate more renewables into their energy mix, advancements in technology, particularly artificial intelligence and machine learning, play a critical role in improving the accuracy of energy predictions. Accurate supply side forecasting is essential for effective energy planning. In the context of hybrid systems such as solar and wind, the Transformer model stands out for its significant accuracy in predicting energy production. These advancements represent significant advances toward achieving resilience and sustainability in the energy sector. Furthermore, to relieve the challenges posed by intermittency, storing surplus renewable energy in the form of hydrogen is proposed as a promising and viable solution.