In long downhill sections, it is difficult to control speed using a single braking method for heavy haul trains. Research on the matching strategy of air brake and electric brake can improve the safety and stability of train passage. To address the issues of inaccuracy in traditional two-stage empirical models and the complexity of fluid dynamics calculations during cyclic braking, an equivalent model of the brake cylinder considering factors such as historical relief time and altitude changes was constructed for quick solutions. A Q-learning-based optimization strategy for matching electric brake during cyclic braking was proposed. Experiments verified the accuracy of the equivalent model and the feasibility of the optimization strategy, enabling the optimization of strategies for different braking performances.