Abstract: A mathematical model with a penalty function was developed to solve optimization problems with multiple constraints in the dispatching of reservoir group water supply by using an improved ant colony algorithm, focusing on the method of improvement. This improved algorithm is featured with self-adaptive adjustment in pheromone evaluation coefficient, information quantity, and transition probability, and it can overcome the defects of traditional ant colony algorithm in slow convergence and local-extremum trapping. Its application to the three reservoirs on the Hei River achieved optimal scheduling schemes better than traditional ant colony algorithm. The improved algorithm is better in computational efficiency, optimization results, convergence, and it provides a new approach to optimization of reservoir group water supply dispatching.