The complex characteristics of coal transmission systems in thermal power plants, such as non-linearity, time variation and large time lags, make traditional control methods less effective and create problems of excessive coal transmission and power production losses. In order to optimise the system performance while reducing energy consumption, an energy saving optimisation model is developed using RBF neural networks around three variables: coal flow rate, belt running speed and system power, and the established energy saving optimisation model is used to build a Smith predictive fuzzy adaptive PID controller. The method has been simulated by MATLAB to achieve automatic online adjustment of the control parameters of the coal conveying system. The speed of the belt conveyor can be adjusted in real time according to the demand of coal combustion, making the control of the coal conveying system more efficient.