Temperature Feedforward Model Based on Improved Stochastic Configuration Network for Coal Mill

Aiming at the practical needs for feedforward mod-eling of the coal mill outlet temperature setpoint under blended coal operation conditions, an improved stochastic configuration network (ISCN) is introduced for predictive modeling. Firstly, the stochastic configuration of hidden node parameters based on the supervision mechanism leads to uncertainty and cannot guarantee parameter optimality. Consequently, an improved pelican optimization algorithm (IPOA) is introduced to optimize the configuration of hidden node parameters, which not only enhances network learning efficiency but also lightweights the network structure. On the other hand, fully supervised SCN mod-eling demands large amount of high-cost experimental data. To reduce modeling expenses, a semi-supervised learning approach based on small labeled data is employed to categorize unlabeled data and expand the training dataset. Nextly, considering that there are some irrationalities in determining the coal mill outlet temperature setpoint using traditional methods. The ISCN introced in this paper is applied to feedforward model of coal mill outlet temperature setpoint. Finally, the validity and practicality of the method are verified by experiments.