Forecasting the Scintillation Index Using Neural Networks
Date of Award
12-1-2023
Degree Name
M.S. in Electro-Optics
Department
Department of Electro-Optics and Photonics
Advisor/Chair
Advisor: Miranda van Iersel
Abstract
This thesis objective is the forecasting of the scintillation index using a machine learning approach. The scintillation index is a measure of the fluctuations of optical wave intensity, also known as scintillation, occurring during the propagation through atmospheric turbulence. The data used for the machine learning-based scintillation index forecasting was obtained during on-going measurements conducted during several years over a 7 km propagation path at the Intelligence Optics Laboratory of the University of Dayton with a commercial scintillometer. Besides the scintillation index and refractive index structure parameter, also meteorological data such as air temperature, wind speed, and relative humidity were measured on both ends of the propagation path. To investigate the influence of seasonal changes on the forecasting of the scintillation index, the data was divided into four subsets corresponding to the four seasons. Necessary data preprocessing steps have been performed, and the data was used to train different machine learning models. The considered models included: bi-directional long short-term memory (Bi-LSTM), convolutional neural network (CNN), K-nearest neighbor (KNN), and random forest (RF). Different Bi-LSTM models were trained by utilizing a single meteorological parameter as an input. Other Bi-LSTM models were trained on different pairs of meteorological parameters (i.e., air temperature and relative humidity, air temperature and wind speed, and relative humidity and wind speed), as well as using all meteorological parameters as inputs. Performance in scintillation index forecasting by different models was compared using a root mean squared error (RMSE). It was found that the Bi-LSTM model trained on all meteorological parameters demonstrated the best performance with RMSE = 1.274 in fall, 2.359 in winter, 4.317 in spring, and 1.700 in summer.
Keywords
Scintillation index prediction, ANN, Meteorological Data, Atmospheric Optics, Machine Learning
Rights Statement
Copyright © 2023, author.
Recommended Citation
Al Ghezi, Mohammed Al Baqer, "Forecasting the Scintillation Index Using Neural Networks" (2023). Graduate Theses and Dissertations. 7335.
https://ecommons.udayton.edu/graduate_theses/7335
