Optimizing Nominal Current Output for Aeronautical Ground Lighting Using Machine Learning and Meteorological Data

Although there have been numerous studies on visibility prediction, there have been insignificant studies conducted to predict nominal current output based on visibility. Therefore, this study focuses on optimizing nominal current output at Subang Airport by employing artificial intelligence and meteorological data. The research leverages daily meteorological data to enhance visibility prediction and address aeronautical ground lighting issues emphasizing on the runway edge light. The methodology involves a three-step modeling approach with Bayesian optimization. First, Gaussian Process Regression was utilized to predict visibility, incorporating various meteorological parameters. Second, a correction filter refines the predictions, integrating models such as Regression Trees, Support Vector Machines, Ensemble of Trees, Neural Networks, and Gaussian Process Regression. Finally, prediction of nominal output current was conducted using error squared, generated from the correction filter, and time. Various machine learning models, including Decision Trees, Discriminant Analysis, Naïve Bayes Classifiers, Support Vector Machines, Nearest Neighbor Classifiers, Ensemble Classifiers, and Neural Network Classifiers were evaluated to determine the most effective model. Cross-fold validation with a 5-fold split ensures the reliability and precision of the machine learning algorithms. Performance metrics such as Mean Absolute Error, Mean Squared Error, Root Mean Squared Error, and R-squared were used to evaluate the models. Results highlight the stacked model of Gaussian Process Regression, Gaussian Process Regression, and Nearest Neighbor Classifiers as the most accurate, achieving a 96.2 % accuracy in predicting and improving nominal output current. In conclusion, this study has introduced a novel approach to predicting and improving nominal output current for runway edge light utilizing limited historical meteorological data.