Comparison of adaptive neuro-fuzzy inference system and recurrent neural network in vertical total electron content forecasting

• Autores: Pérez Bello, Dinibel; Natali, María Paula; Meza, Amalia Margarita
• Año de publicación: 2019
• Idioma: inglés
• Tipo de recurso: artículo
• Materia: Astronomía; vTEC; Space weather; Neural network; Forecasting
• Institución: Universidad Nacional de La Plata

Accurate prediction of total electron content (TEC) is important for monitoring the behavior of the ionosphere and indeed a magnitude of interest to understand the properties and behavior of the Sun–Earth System. The conditions of this medium have a direct impact on a growing variety of critical technological infrastructure. This work presents a comparison between two different artificial neural networks (ANNs): an adaptive neuro-fuzzy inference system and nonlinear autoregressive neural network (NAR-NN) applied to TEC. Both ANNs where tested on four different geomagnetic locations on 4 1-week periods having a variety of geomagnetic disturbance levels. The effect of using different training period lengths and the system response for 60 and 30 min sampling rate TEC time series was investigated. NAR-NN shows a slightly better performance, being the higher difference during the greater perturbations. There is also a better response when sampling rates of 30 min are used.
Facultad de Ciencias Astronómicas y Geofísicas
Laboratorio de Meteorología espacial, Atmósfera terrestre, Geodesia, Geodinámica, diseño de Instrumental y Astrometría