集成混合随机森林与KPCA的磨矿粒度随机增量建模OA

Grinding particle size is a key indicator for measuring the production quality of the grinding process,which directly affects the stability of the separation process,concentrate grade and metal recovery rate.However,due to the high cost,long detection time,and easy blockage of online particle size analyzers,it is difficult to perform accurate measurement of grinding particle size.Meanwhile,the broad learning system-based modeling methods face two main limitations:in-sufficient feature extraction capability and random allocation of node parameters in the enhancement layer.By integrating hybrid random forests(HRFs)with kernel principal component analysis(KPCA),a novel improved random incremental learning(RIL)soft measurement method called HRF-KPCA-RIL is proposed in this paper.Firstly,in the feature mapping layer,a hybrid forest group based on random forests and completely random forests is employed to replace the traditional neuron group for carrying out high-dimensional feature vector mapping.Secondly,the KPCA is utilized for feature extrac-tion from the high-dimensional mapping matrix,and the mutual information method is further applied to filter potential features and eliminate redundant information.Then,in the enhancement layer,high-quality nodes for soft measurement modeling are generated using the Greville iterative method-based global constraints to alleviate randomness issues.Final-ly,experimental studies based on real-valued function,benchmark datasets,and grinding particle size dataset validate the superiority and effectiveness of the proposed HRF-KPCA-RIL model.