Abstract:
Earthquake Prediction is a mainly unsolved problem. A large number of different approaches have been tried and only a small number of attempts were fruitful. A few of these are explained briefly in this thesis. One of the most succesful earthquake prediction sytems in use today is the Canada-Nevada, CN, algorithm. It is discussed and contrasted to the neural networks implemented in the project. For this project the earthquake prediction problem is treated as a time senes prediction problem and neural networks that have been used for ordinary time senes prediction with some success have been applied to the problem. The data used was treated as a two dimensional time series with two variables; the magnitude of the present earthquake, and the time elapsed since the previous earthquake. The neural network architectures implemented were the" multilayer perceptron network with sigmoidal activation function, NADINE, and a mult,ilayer network with chaotic activation function. The results were not succesful because of the complex nature of input data and the earthquake generation process.
