Considering practical requirements of a provincial power grid project and natural characteristics of hydro power, fire power and wind power, in the premise of full amount of wind power integrated into the grid, with minimum the power purchase cost and emissions as the objective, taking into account the thermal and hydropower units’ output smoothness constraints and each day hydropower scheduling fixed constraints, a dynamic multi-objective water-fire-wind coordination optimal scheduling model is built. In order to simplify the solution of the built model, this paper proposes a new constraint processing methods. Through the introduction of quasi-opposition-based learning and self-learning mutation, an improved multi-objective "teaching" and "learning" optimization algorithm (MMTLA) is got and applied to solve the model. Taking a provincial power grid as an example, the simulation results show the validity of the built model and the superiority of the proposed algorithm.