Indirect prediction of paste conveying parameters based on physical information neural networks

In response to the difficulty of monitoring paste conveying parameters (paste pump outlet pressure, filling flow rate) during the filling process of mining paste, this study develops an indirect prediction methodology employing physics-informed neural networks (PINN).The model and coupling mechanism of the hydraulic system of the paste conveying pump and the paste pipeline conveying system are studied, and the differential equations of the mapping relationship between the internal hydraulic system parameters (pump station pressure, plunger displacement) of the paste conveying pump and the paste conveying parameters (paste pump outlet pressure, flow rate) are established. On the basis of the fourth-order Runge–Kutta algorithm, the numerical solution of the differential equations of the mapping relationship is obtained by MATLAB, and in comparison with the measured data of the transmission site, the correctness of the mapping relationship is verified. The governing differential equations are embedded within a deep neural network framework, with multistage physics-informed constraint terms specifically designed for targeted objectives. This study establishes a PINN model that integrates physical constraints with data-driven approaches, enabling the indirect prediction of paste conveying parameters from hydraulic system characteristics. Results demonstrate that the differential equations effectively characterize the mapping relationships between hydraulic system parameters and paste conveying characteristics. The developed PINN model achieves mean absolute errors of 0.35 MPa (<4%) for pressure prediction and 3.6 m³/h (<5%) for flow rate estimation, demonstrating accuracy sufficient for practical engineering applications. Furthermore, the model successfully utilizes measured data from initial pumping phases to predict subsequent paste transportation parameters throughout continuous operations. The proposed method can provide early warning and guidance for the safety monitoring and operation control of paste filling work.