Improved incremental Regularized Extreme Learning Machine Algorithm and its application in two-motor decoupling control

Regularized Extreme Learning Machine (RELM) is an ideal algorithm for regression and classification due to its fast training speed and good generalization performance. However, how to obtain the suitable number of hidden nodes is still a challenging task. In order to solve the problem, a new incremental algorithm based on Cholesky factorization without square root is proposed in this paper, which is called the improved incremental RELM (II-RELM). The method can automatically determine optimal network structure through gradually adding new hidden nodes one by one. It achieves less computational cost and better accuracy through updating output weights. Finally, neural network generalized inverse (NNGI) based on II-RELM is applied to two-motor synchronous decoupling control. Simulation indicates that the proposed algorithm has excellent performance in prediction control. It realizes the decoupling control between velocity and tension.