การพัฒนาระบบพยากรณ์การเคลื่อนที่ของเมฆโดยใช้ภาพถ่ายผ่านดาวเทียมและเซนเซอร์ทางกายภาพ

Abstract

This independent study is to develop and test a cloud movement prediction model applying an artificial neural network using data from satellite images and physical sensors. The predicted cloud images in Muang, Chiang Mai area for the next hour and three hours are generated from the proposed prediction methods. There are two prediction methods, which are 1) absolute pixel image input, and 2) relative pixel image input. Each method conducts with four scenarios, according to physical sensors and prediction periods. Four scenarios are 1) scenario-1: the prediction with data from satellite images and physical sensors for next hour, 2) scenario-2: the prediction with only data from satellite images for next hour, 3) scenario-3: the prediction with data from satellite images and physical sensors for next three hours, and 4) scenario-4: the prediction with only data from satellite images for next three hours. The results show that the prediction using absolute pixel image is more accurate than using relative pixel image but the absolute pixel image method is much slower than the other for training network and testing. Scenario-1 is the most accurate prediction, the average mean squared error (MSE) is 0.0096. The comparison between next hour and three hours prediction found an hour is more accurate than three hours, the average MSE is 0.0132. Using physical sensors data is more accurate than not using, the average MSE is 0.0180.