A stable, high quality cement burning system producing clinker with low energy consumption is important for cement company. However, in the actual operation of the combustion system, there are contradictory indicators such as electricity consumption, coal consumption and clinker quality that are difficult to be jointly optimized, as well as dynamic and complex working conditions that are difficult to be artificially adjusted to the optimal state. In order to solve the above problems, a multi-objective optimal predictive control model is proposed in this paper. The prediction part of the model consists of a prediction model of cement burning system based on convolutional neural network. The optimization part, on the other hand, combines electricity and coal consumption into one energy target and uses cement clinker quality as a constraint to find the optimal production state. This modeling approach simplifies the problem and avoids the need to solve for the Pareto optimal solution. The control part is then based on the above-mentioned model with a double-cycle coupled cycle optimization to obtain set values of controlled variables that satisfy the changing operating conditions and finally control target to track the dynamic operating conditions. Experimental results show that the method described in this paper reduces the content of free calcium oxide after cement combustion, thus improving the quality of cement clinker, while reducing the fluctuations of the control variables in each production process and improving the stability of the combustion system. There is also a significant reduction in production energy consumption.