Combinatorial optimization is often with the local extreme point in large numbers. It is usually discontinuous,
multidimensional, non-differentiable, constraint conditions, highly nonlinear NP problem. In this paper, according to the
characteristics of combinatorial optimization problem, we put forward the combination optimization of multi-agent differential
evolution algorithm (COMADE) through combining the multi-agent and differential evolution algorithm, in which
we designed the competition behavior and self-learning behavior of agent. Through performance testing of strong connected,
weak connected and overlap connected deceptive function on the COMADE algorithm, the results show that the
COMADE algorithm is effective and practical value.