Abstract: To cope with the challenge of frequency control of interconnected power grids brought by the high proportion of renewable energy,the existing studies have proposed that the drawback of conventional lag control of automatic generation control（AGC） units is overcome by the dynamic optimal dispatch of AGC units.However,the dynamic optimal dispatch model of AGC units is a mixed-integer nonlinear programming（MINLP） model,which is not easy to solve.Based on the convex optimization theory,an improved algorithm for the dynamic optimal dispatch model of AGC units is proposed.Firstly,the bilinear term of the original model is relaxed based on the variable substitution and piecewise McCormick envelope technique.Secondly,the original model is transformed into a mixed-integer linear programming（MILP） model that can be efficiently solved by linearizing the power losses.On this basis,a bi-level iterative algorithm is proposed to improve the solving accuracy of the dispatch model by tightening the bilinear term relaxation gap through the inner loop and reducing the linearization error of power loss through the outer loop.Finally,the IEEE 14-bus system and the IEEE 57-bus system are adopted to verify the effectiveness of the proposed improved algorithm.