Real-time and accuracy are requested in reactive power optimal control of regional power grid. The accuracy of the solution, timeliness of calculation and rapidity of control must satisfy the demand of power grid. Based on multi-thread genetic algorithm, target classification reactive power optimal control of regional power grid is proposed for above requirements. Multi-threaded calculation of sensitivity and specified space sensitivity are adopted for choosing initial population; fuzzy self-adjustment and similarity are used in crossover probability and mutation probability respectively. Improved genetic algorithm which is optimized by gate multi-thread enhances the convergence speed; control target is considered by classification; parallel remote control is considered by gate. Combining with one regional power grid instance, analysis results show the feasibility of proposed method. This method satisfies the real-time requirements of reactive power optimal control of current regional power grid.