The grade of concentrate is a crucial indicator reflecting the effectiveness of the flotation process, which differentiates minerals based on their physicochemical properties. The complexity of operational conditions poses significant challenges to the precise prediction of concentrate grade. This paper proposes a method for predicting concentrate grade in the flotation process based on multi-condition integrated learning. Firstly, to address the complexity of operational conditions, an adaptive density peak clustering algorithm based on information entropy index is proposed to intelligently segment the operational conditions of the flotation process, and establish sub-models for concentrate grade prediction under different conditions. Secondly, to address the difficulty of effectively modeling the complex flotation process with a single model, a multi-model ensemble strategy using BP neural network Stacking is proposed, which integrates different condition sub-models to enhance prediction accuracy. Finally, the effectiveness and accuracy of the proposed method are validated using industrial site data.