Kernel extreme learning machine for flatness pattern recognition in cold rolling mill based on particle swarm optimization

In this paper, we propose a kernel extreme learning machine (KELM) flatness recognition model based on particle swarm optimization (PSO). Compared with the extreme learning machine (ELM), the KELM has fewer initial parameters and better recognition performance. Next, the PSO algorithm can serve to optimize the setting parameters of KELM, and finally, the proposed algorithm (briefly the PSO-KELM) is applied to recognize the flatness pattern of cold rolling mill. In particular, PSO-KELM is trained and tested by simulation flatness data, and the test results show that the PSO-KELM dominates backpropagation neural network (BP), ELM, and KELM. Then, the measured data from the shape meter of cold rolling mill is taken as test data, and the test results are reconstructed to the flatness curve by the flatness curve equation. By the fitness of the measured flatness values with the recognition flatness curve, we claim that the PSO-KELM can make accurate recognition in complex situation.