As to the problem of falling into the local optimum in standard artificial bee colony, it is proposed to introduce an adaptive factor which can expand the search of the swarm and use the Cauchy distribution to improve the universality of colony search. This improved algorithm named adaptive Cauchy mutation artificial bee colony (ACMABC). Then the ACMABC is analyzed in theory by using the theory of random process to prove the convergence of the algorithm. Finally, this modified method is applied to the optimization of the parameters of wind power short-term prediction model, compared with standard statistic strategy, an illustration with higher precision is given.