ADAPTIVE PREDICTION OF FOREST FIRE BEHAVIOR ON THE BASIS OF RECURRENT NEURAL NETWORKS
V. I. Kozik, E. S. Nezhevenko, A. S. Feoktistov
Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, pr. Akademika Koptyuga 1, Novosibirsk, 630090 Russia
kozik@iae.nsk.su
Keywords: computer simulation, forest fire, recurrent neural network, data assimilation, Kalman filter
Subsection: ANALYSIS AND SYNTHESIS OF SIGNALS AND IMAGES
Abstract
A method of modeling a dynamic process on the Earth surface, for instance, a forest fire, with the use of a recurrent neural network is proposed. The learning process of the neural network, similar to the process of data assimilation in GIS technologies, is described. A method of acceleration of neural network learning by using the Kalman filtration is proposed. The efficiency of its application is analyzed, and the neural network parameters at which it is reasonable to use the Kalman filter are determined.
|
