An improved real-coded genetic algorithm(IRCGA) is proposed to solve constrained optimization problems. Firstly, a sorting group selection method is proposed, which has good population diversity and is easy to realize. Then, a direction-based heuristic crossover operator(DBHX) is proposed, which can generate numerous crossover directions. and it also enables a great possibility to generate a crossover direction vecD to guide the chromosomes of participation crossover to move towards the optimal solution direction. Even if the crossover direction is inconsistent with vecD, it is also very closed to the direction vecD to the utmost so as to ensure that there is a great chance to produce better offspring chromosomes. Finally, aiming at the shortcoming that a single mutation operator cannot both take into account the local search ability and the global search ability, a combined mutation method is proposed, which makes the mutation operation not only take into account the local search ability, but also the global search ability. The computing results of ten examples show that the proposed IRCGA has a fast convergence speed, and also verify its effectiveness and feasibility.