This article addresses the issues of low accuracy and time-consuming nature associated with traditional density detection methods in ore grinding classification by proposing an intelligent pulp density detection method. Through mechanistic analysis of the pulp fluid, linear known terms and nonlinear unknown terms are identified. A holistic recognition of pulp density is performed by combining Gaussian process regression with a regularized stochastic configuration algorithm. Additionally, the variance estimated by the mechanistic model is set as the training objective for the data-driven model, enhancing the model's capacity to acquire data information. Meanwhile, a collaborative computing method is employed to apply the adaptive intelligent detection method in the industrial domain, ensuring realtime detection and adaptability of the pulp density detection model. Based on industrial data experimental analysis, the proposed method shows an average absolute error of 7. 13, a root mean square error of 9. 31, a determination coefficient of 99. 51% , and a sample quantity proportion of relative error δ < 1. 0% at 83. 58%. These results are better than those of other comparative algorithms. The effectiveness of the pulp density detection model is enhanced.