Development of a Novel Dynamic Weighted Imputation Algorithm for Soft Sensing with Missing Data

Accurate measurement of grinding particle size in copper ore processing is crucial for production quality and efficiency. Due to the limitations of hardware sensors and high measurement costs, soft sensors technologies are widely used for prediction. However, missing data in collected datasets often hinders system modeling and prediction performance. To address this issue, this study introduces a novel dynamic weighted imputation algorithm based on Stochastic Configuration Network (SCN), named DWI-SCN. The DWI-SCN algorithm starts with training a SCN model using complete data, and then generates multiple-candidate imputed values by integrating weighted similarity and Gaussian distribution, subsequently selecting the optimal value based on prediction error. Finally, the model is retrained using the fully imputed dataset. Experimental results demonstrate that DWI-SCN significantly improves predictive performance in both grinding particle size and other industrial datasets.