International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 09 | Sep 2025
p-ISSN: 2395-0072
www.irjet.net
A Novel Real-Time GIS and Machine Learning Framework for Wind Energy Potential Analysis and Turbine Site Optimisation with an Application to Indonesia Sanat Punj1 1British School Jakarta, Indonesia ---------------------------------------------------------------------***--------------------------------------------------------------------source with no combustion and zero direct emissions Abstract - This study proposes a novel, globally applicable
(Particulate matter, Volatile Organic Compounds (VOC), Greenhouse Gases) [1]. Thus, various studies have been conducted in order to optimize and exploit wind energy to maximize clean energy potential in various geographical areas in countries and provinces [2]. Common methodologies for analyzing Renewable Energy Potential in previous studies given a particular geographical setting include the popular Weibull Parameter Estimation Methods, which is dependent on using the statistical analysis of Wind Speed Data to estimate the shape and scale parameters of the Weibull Distribution [3]; some other common methods employed include heavy computational based statistical analysis methods including but not limited to the use of the Geographic Information System (GIS) data.
framework for evaluating wind energy potential and identifying optimal sites for wind power generation, demonstrated through a case study in Indonesia. The framework integrates past meteorological and geospatial data, machine learning with hyperparameter tuning, and real-time GIS and weather data retrieval via API pipelines. The dataset, Wind Power Generation Data Forecasting, comprises 175,200 hourly observations from January 2017 to December 2022 across four geographically distinct turbine sites. Each record includes ten climatic variables (e.g., wind speed, direction at multiple heights, temperature, humidity) and turbine power output. Power output values were normalized on a 0-1 scale to enable consistent training across sites, where 1 represents maximum achievable power. Seven machine learning models were evaluated: ensemble tree-based (XGBoost, Gradient Boosting, Random Forest), neural-networks (ANN, RNN, GRU), and Support Vector Regression. Hyperparameters were systematically tuned, with model-performance assessed using MSE, RMSE, MAE, MedAE, and R². XGBoost achieved the best generalization, combining the lowest error metrics with the strongest fit, making it the optimal choice for deployment. Real-time GIS-based weather data was retrieved via the Copernicus Climate Data Store (CDSAPI) for fixed coordinate-intervals in Indonesia, filtered using the Natural Earth polygons dataset. The trained model assessed normalized wind energy potential at each coordinate, with values sampled across months and times, then averaged to estimate site-specific potential. Ultimately, 145 evenly spaced onshore coordinates (580 ERA5 grid points) generated 3,480 spatiotemporal instances, enabling a spatial map of Indonesia’s wind energy potential and identifying optimal turbine-placement sites
Machine learning methods for assessing wind energy potential started to gain traction from the early 2000s with the initial use of Artificial Neural Networks (ANN) [4][5] and Support Vector Machines (SVM) [6] for wind forecasting and renewable energy potential estimation. These models were utilized in their specific domains for identifying nonlinear relationships and regression-based linear relationships. These models were able to dynamically adjust to different wind regimes based on training data, unlike certain statistical analytic methods, such as the Weibull, and were thus heavily proposed and utilised in studies during that era. With the increasing sophistication of Artificial Intelligence models as their performance optimizes at a groundbreaking rate over the last 15-20 years, machine learning algorithms such as the ensemble method, decision trees and deep learning methods such as the Recurrent Neural Network (RNN) became popularized improving prediction accuracy and handling large datasets from remote sensing technologies such as the LIDAR [7][8].
Key Words: Wind Energy Potential, Machine Learning, Geographic Information System, Renewable Energy, Wind Turbine, Optimisation, Indonesia
In the Indonesian context, the main challenges to implementing wind turbines are also the variability of conditions and difficulty in identifying optimal sites for the wind turbine installation, as identified by the famous book by Taufal Hidayat, “Wind Power in Indonesia: Potential, Challenges, and Current Technology Overview“ [9]. By integrating GIS and machine learning methods, these areas
1.INTRODUCTION Wind power has become a very popular energy resource recently, given its nature of producing clean and renewable energy. It provides electricity without burning fuel or polluting the air, being a technologically mature
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 12