Assessing the Technical and Economic Viability of Onshore and Offshore Wind Energy in Pakistan Through a Data-Driven Machine Learning and Deep Learning Approach

Abstract

An accurate estimation of wind energy productivity is crucial for the success of energy transition strategies in developing countries such as Pakistan, for which the deployment of renewables is essential. This study investigates the use of machine learning and deep learning techniques to improve wind farm producibility assessments, tailored to the Pakistani context. SCADA data from a wind turbine in Türkiye were used to train and validate five predictive models. Among these, Random Forest proved most reliable, attaining a coefficient of determination of 0.97 on the testing dataset. The trained model was then employed to simulate the annual production of a 5 × 5 wind farm at two representative sites in Pakistan—one onshore and one offshore—that had been selected using ERA5 reanalysis data. In comparison with conventional estimates based on the theoretical power curve, the machine learning-based approach resulted in net energy predictions up to 20% lower. This is attributable to real-world effects such as wake and grid losses. The onshore site yielded an LCOE of 0.059 USD/kWh, closely aligning with the IRENA’s 2024 national average of approximately 0.06 USD/kWh, thereby confirming the reliability of the estimates. In contrast, the offshore site exhibited an LCOE of 0.120 USD/kWh, thus underscoring the need for incentives to support offshore development in Pakistan’s renewable energy strategy.