Penerapan Jaringan Saraf Tiruan / JST (Backpropagation) untuk Prakiraan Cuaca di Bandar Udara Radin Inten II Lampung

Adi Saputra, Sri Ratna Sulistiyanti, Roniyus Marjunus, Yanti Yuliant, Junaidi Junaidi, Arif Surtono

Jurnal Teori dan Aplikasi Fisika · 20 February 2023

In one sentence. Backpropagation neural network forecasts daily weather at Radin Inten II airport, Lampung, from three years of station data, reaching 74.36% best test accuracy.

Abstract

Weather prediction is needed in planning daily life, one of which is to make decisions. The success of a weather prediction will have an impact on decision making in various fields, including agriculture and aviation. In the field of aviation, weather prediction is important to determine the time, location, direction of motion, altitude and plan the movement of aircraft to take into account operational disturbances that can be caused if the weather is bad and also to consider in determining flight routes or determining in carrying additional fuel if in an emergency. In the case of the aircraft having to return due to unfavorable weather conditions. Therefore the need for a good weather prediction method so as to reduce losses and damage. In this case the author tries to focus on the maximum parameters in the development of weather forecasting information design based on Artificial Neural Networks / Backpropagation by adding input data of rainfall, temperature, humidity, sunlight, air pressure, wind direction and wind speed. This research was conducted in the area of Radin Inten II Airport, Lampung. The material used in this study is in the form of daily data on meteorological conditions in the Radin Inten II Lampung Airport area from the Radin Inten II Meteorological Station for the last 3 years, from 2017 to 2019. This data is needed as input data for the algorithm that will be used in study. Based on the research results, the best training accuracy is 100% on the artificial neural network architecture with levenberg-marquardt training function parameters (trainlm) and scaled conjugate gradient (trainscg), binary sigmoid and bipolar sigmoid activation functions, and the number of neurons 20, 40, 60, 80, and 100. Meanwhile, the best test accuracy is 74,359% on the artificial neural network architecture with the training function parameters gradient descent wit momentum and adaptive learning rate (trainingdx) and binary sigmoid activation function (logsig) and the number of neurons 20 and 80.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.