To address issues such as multi-line and multi-bunker coupling constraints，load fluctuations，and energy waste caused by transportation delays in the coal conveying system of thermal power plants，this paper designs an improved Model Predictive Control(MPC)-Long Short-Term Memory(LSTM)energy-saving optimization collaborative control method. Firstly， a dynamic material balance equation for multiple bunkers is established，enabling coordinated scheduling of coal conveying lines and balanced material level distribution through MPC. A multi-objective function for system energy consumption is constructed，incorporating operational energy，startup energy，and no-load energy. Secondly，an LSTM network is employed to predict future coal demand，combined with a belt conveyor speed control strategy，to proactively respond to load changes andeliminate supply-demand mismatches caused by time delays. Experimental results demonstrate that compared to traditionalmaterial level control，the proposed method achieves a 20% energy-saving rate，reduces equipment startup frequency by