In order to reduce fuel costs and achieve scientific fuel control in coal-fired power generation enterprises, a bi-level fuel cost optimization model that comprehensively considers the reliability of power generation and the economy of production is proposed in this paper. The upper layer is a mixed integer programming model for power coal purchasing and inventory, while the lower layer is a nonlinear optimization model for mixed coal combustion. The bi-level model is the alternating iterative coordination optimization to realize the dynamic decision-making of fuel control. Then, an improved gray wolf optimization algorithm combining chaos mapping initialization and Gaussian mutation is proposed for the high dimensional and multiple constraints of the optimization model. Finally, a simulated operation of a coal-fired power plant is used to verify the results. The results show that the proposed model can reduce the coal cost of coal fired power plant by 7.80%, which proves the effectiveness and feasibility of the proposed model and solution algorithm.