Improved ill-posed echo state network and its application to blast furnace gas amount forecast

Blast furnace gas is very important in its running process. It is difficult to predict accurately. In the paper, a new method is proposed to deal with the problem of ill-posed echo state network (ESN). ESN could achieve very high precision in time series prediction and overcome many issues encountered in using traditional artificial neural networks. In order to achieve better predicting result in an ill-posed system, L-curve method is introduced to eliminate the effect of ill-pose. Simulation results further illustrate the effectiveness of the design method.