Based on the just in time idea, a production planning model is established in the steel-making and continuous casting. In the model, with sequence casting constraint, the objective is to minimize the operation conflict time and the deviation time of punctual delivery. Then a hybrid GA combined the backward inferring algorithm with GA is presented to solve the model. As the delivery time, the start time of each charge on each working stage is obtained by the backward inferring algorithm. GA is applied to search the suitable start pouring time, select the reasonable operation time and transportation time, assign the feasible processing unit, and evaluate the optimization production plan. The test with the production data in a steel plant shows that the model and algorithm can draw quickly a high-quality and performable production plan, and the average staying time of the production plan is about 7% shorter than that in a practical process.