Studying the method of adaptive prediction of forest fire evolution on the basis of recurrent neural networks

A software system is presented for implementation of a fire model on the basis of a recurrent neural network, which ensures real-time simulation of fire evolution. The quality of traditional learning and learning based on the Kalman filter in experiments performed with the neural network is compared. It is demonstrated that the fire overcomes obstacles in the form of regions consisting of incombustible materials owing to the global character of connections of the neural network simulating the fire.