A neural network approach for fire following earthquake loss estimation

Given the great fire conflagrations following the 1906 San Francisco and the 1923 Great Kanto earthquakes, fire-following earthquake has been widely recognized as a major potential component of earthquake loss. However, because of the complexity of the physical phenomena, any reasonable loss estimate must be based on real time simulations, which are often too complicated and time consuming. This paper describes a neural network approach for fire following earthquake loss estimation. The paper also describes how the backpropagation neural network can be modified to use the conjugate gradient method for speeding up the training process. >