Backpropagation through Time for Learning of Interconnected Neural Networks -- Identification of Complex Systems

Neural networks are mainly employed to model complex systems behavior. This work aims at broadening area of their applications to input-output dynamic complex systems of cascade structure. Each element of the complex system is modeled by a multi-input, multi-output recurrent neural network. A model of the whole system is obtained by composing all neural networks into one global network. Main contribution of this work is generalization of Back propagation Through Time method to complex systems modeled by interconnected neural networks. Appropriate algorithm is provided and numerical simulations are performed.