Diagonal recurrent neural network based control using adaptive learning rates

A method of choosing learning rates adaptively in controlling a dynamic system using diagonal recurrent neural networks is presented. The architecture of DRNN is a modified model of the fully connected recurrent neural network with one hidden layer. The hidden layer is comprised of self-recurrent neurons, each feeding its output only into itself and not to other neurons in the hidden layer. An unknown plant is identified by a diagonal recurrent neuroidentifier (DRNI), and provides the sensitivity information of the plant to a diagonal recurrent neurocontroller (DRNC). A dynamic backpropagation algorithm (DBP) is used to train both DRNC and DRNI. The DRNN captures the dynamic nature of a system, and, since it is not fully connected, training is much faster than for the case of a fully connected recurrent neural network. For faster learning, the use of an adaptive learning rate that guarantees convergence is developed. The proposed approach is applied to numerical problems. Simulation results are included