There are abundant mixed-integer nonlinear programming problems in actual engineering and management. For this, a filter hybrid co-evolutionary algorithm is proposed, which utilizes filter technology to deal with constraints instead of penalty function. Hybrid coding and heterogeneous population co-evolution composed of differential evolution and genetic algorithm are proposed for solving mixed-integer. The introduction of the average entropy and Logistic chaos initialization population increases the algorithm's robustness. To improve the search capability, the strategy of cooperative composed of adaptive scaling factor strategy and elite exchange learning strategy is presented, which couples with the population co-evolution into the filter hybrid collaborative evolutionary algorithm. The reactive power optimization results of an actual 30-bus power system show that the proposed algorithm possesses global search ability and effectiveness.