Double Echo State Network with Multiple Reservoirs for Time-series Prediction

With the rise of neural networks, Echo state network (ESN) is also added to the queue of the neural network. ESN is a powerful form of reservoir computation, is a new type of recurrent neural network and only need to train the output weights. ESN overcomes the problems of the traditional recursive neural network such as too complicated training algorithm, slow convergence speed and easy to fall into local minimum In order to reflect different learning tasks need different status updating methods and appropriate reservoir are constructed for a given task, an improved leaky integrator echo state network is proposed. The improved model, named Double Echo State Network with Multiple Reservoir(DM-ESN) for Time-Series Prediction. It uses multiple sub-reservoir to build a total reservoir and extends it in depth. Then the batch gradient descent method is utilized to optimize the DM-ESN parameters. Finally, the proposed method is applied to the time-series prediction. Simulation results show that the DM-ESN model can greatly improve the accuracy and stability.