To closely describe the earliness/tardiness production planning problems in the JIT environment, a nonlinear semi-infinite programming model is proposed. Due to the issues of non-convexivity and having infinitely many constraints, instead of applying traditional optimization ap-proaches, a specially designed genetic algorithm with mutation along the negative gradient direction is developed. The proposed algorithm is a combination of the steepest descent method with the stochastic sampling algorithm. Some numerical results are included to show its potential for industrial applications.