A Decreased Extreme Learning Machine with Ridge Parameter for Online Identification of Nonlinear Systems

A recursive method of decreased extreme learning machine (DELM) is proposed for online identification of nonlinear systems. The output weights of ELM can be recursively updated by decreasing the hidden nodes one by one in an efficient manner. Furthermore, a ridge parameter is introduced into the transposed matrix to overcome the singular problem. The simulation results for several benchmark problems demonstrate that the proposed DELM method can reduce the computational complexity efficiently, and maintain the good prediction performance of the model, compared to the traditional ELM algorithm.