Abstract:
With the sewage treatment process becoming more and more complex, the proportion of easy-to-measure and hard-to-measure variables is seriously out of balance such that the traditional supervised soft-sensor modeling cannot meet the actual requirements. Aiming at this problem, this paper proposes a new soft-sensor model, semi-supervised Tri-training MPLS. The labeled data are divided into three independent parts, from which the unlabeled data with high confidence will be selected to improve the prediction ability of the model. In addition, the single-output model is upgraded to the multi-output model to directly predict multiple output variables. Finally, it is shown from the simulation results via the BSM1 platform (Benchmark Simulation Model-1) that the proposed soft-sensor model has good prediction ability on multiple output prediction and satisfactory prediction on single target.