Wind speed prediction by utilizing geographic information system and machine learning approach: A case study of Karabük province in Türkiye

Abstract

This study analyzed wind speed data for years in Karab & uuml;k province, T & uuml;rkiye, using an Artificial Neural Network (ANN) with a Multilayer Perceptron (MLP) feed-forward network. The Bayesian Regularization algorithm was employed, a well-known training algorithm for Multi-Layer Perceptron (MLP) networks. The study investigated the relationship between wind speed and various meteorological parameters such as month, air temperature, relative humidity, and air pressure. The results obtained from the ANN model provided a reliable methodology for predicting future wind speed values in Karab & uuml;k province. To evaluate the performance of the ANN model, metrics such as Mean Absolute Error (MAE), Average Relative Deviation (ARD), Mean Squared Error (MSE), and R-squared (R2) were utilized. The ANN model demonstrated its efficacy by revealing the highest average wind speeds of 2.7 m/s in Safranbolu province during August, with corresponding MAE, ARD%, MSE, and R2 performance metrics of -0.029, -0.380%, 0.0028, and 0.999, respectively. The maximum measured and predicted Mean Wind Speed (MWS) values were identified in different months across various locations, specifically in August for Eflani, July for both Eskipazar and Karab & uuml;k CC and September for Safranbolu. Notably, the highest recorded MWS was observed at 42.8 m/s in Eskipazar during July, while the lowest MWS was recorded at 16.4 m/s in Eskipazar in October. Besides, by employing Geographic Information System (GIS) analysis, the average wind speeds were ranked for different districts, with Safranbolu, Eflani, Eskipazar, and Karab & uuml;k CC having the highest to lowest wind speeds, respectively.