Data-Driven Soft Sensor Modeling Method for Grinding Particle Size Using IRVFLN with Compact Constraints

The quality of grinding particle size is directly related to the concentrate grade and recovery rate in flotation. Through the analysis of the production status and basic production data of the actual grinding process, it is evident that there are issues such as high online detection costs, long lag times, and implementation difficulties associated with grinding particle size. Therefore, this article establishes an incremental random weights neural network with compact constraints (IRCC) grinding particle size prediction model. This model utilizes compact constraints to select high-quality hidden parameters, thereby establishing a highly generalized model. Furthermore, this article proves the universal approximation property of the proposed IRCC. Experimental results on actual grinding data demonstrate that IRCC exhibits excellent performance in predicting grinding particle size.