The Parameter Updating Method Based on Kalman Filter for Online Sequential Extreme Learning Machine

The recursive least-square algorithm in the on-line sequential extreme learning machine (OS-ELM) is well known for its good convergence property and least square error in stationary system. But the performance in unstable system is not so good due to the matrix singularity and forgetting factor trade-off. In this paper, an improved OS-ELM algorithm is introduced based on kalman filter (KOS-ELM) for online parameter updating through considering the modeling error into the equations to avoid the matrix singularity and randomly walk character into the parameters to handle the non-stationary. The regression experiment which are simulated both in stable and unstable condition system demonstrate that compared with the classic overlay ELM and OS-ELM, the proposed method achieves higher accuracy and stability.