Abstract: A dynamic-window-search ant colony optimization (ACO) algorithm, integrated with genetic optimization techniques, is proposed for large-scale multi-stage decision making problems, which are of strong nonlinearity, complex constraints on system states and control inputs, non-analytical system representation, and additive and monotonic objective functions. A subset of the feasible decision set at each stage is dynamically selected for the algorithm by real-coded genetic optimization and is mapped to the nodes of the corresponding layer in a layered construction graph to reduce the size of the search space. Computational complexity analysis and simulation results demonstrate that, in comparison with basic ACO algorithms, the proposed algorithm greatly improves the computational efficiency.