Abstract: In order to realize intelligent feeding and precise operation of semi-autogenous grinding process，improve equipment operation efficiency and reduce production cost，a real-time monitoring and intelligent control method based on multi-source fusion of vibration，acoustic and current signals is pro⁃ posed to solve the problems of insufficient accuracy of single sensor，poor adaptability of working condi⁃ tions and control lag in mill load monitoring.The key features are extracted by time domain analysis (RMS， kurtosis，margin factor) and frequency domain analysis (center of gravity frequency，frequency band ener⁃ gy ratio). The harmonic dynamic compensation mechanism and feature dynamic weighted fusion model are innovatively introduced，and the high-precision estimation of filling rate (error <3%) is realized by combin⁃ ing the dual-channel deep learning architecture.The industrial application results show that the system re⁃ sponse delay is less than 100 ms，and the power consumption per ton of ore is reduced by 12.4%. It effec⁃ tively solves the monitoring and control problems of traditional methods under complex working conditions， and provides a complete solution for the intelligentization of the mill.

Key words: mill load monitoring, multi-sensor fusion, dynamic weighting, deep learning, closed?loop control