A New Improved Echo State Network with Multiple Output Layers for Time Series Prediction

Through investigating the traditional echo state network, their output structure has only one output layer using the same output weight learning method, such that the networked prediction results is not always reliable. Therefore, a new echo state network with multiple output layers (MOL-ESN) in parallel configuration is proposed for time series prediction in this paper. For the output structure of MOL-ESN, multiple output layers with different output weight learning methods are built. Considering the multiple output layers are introduced, the computing burden of training the MOL-ESN will be also increased, and thus, on the premise of ensuring the stable network output, the prediction performance of the MOL-ESN need to be improved. Finally, the effectiveness of the proposed network is verified by a simulation example.