ARCHIVES

This paper proposed a state prediction method for real-time prediction of the operation of a clinker rotary kiln, the most important equipment for cement production. Although many different results have been derived in the past to address this issue, they have not been satisfactory due to installation costs, service life, etc. We first selected the main motor current of the rotary kiln as training data for state prediction and first order denoising using wavelet threshold denoising method. And we performed the second order noise extraction once more using the improved local mean decomposition (LMD) method. Then, we derived an incremental extreme learning machine (IELM) algorithm to enable the extreme learning machine (ELM) algorithm to be applied online. To overcome the “data saturation” drawback present in the IELM, we proposed a limited incremental ELM (LIELM) algorithm that limits the total number of training samples. Then, LIELM predicted the main motor current state change of the rotary kiln. Finally, we compared the prediction accuracy of the proposed algorithm with the standard ELM. This algorithm was able to successfully overcome the phenomenon of "data saturation" and thus better approximate the time-varying nonlinear plant prediction.