Due to the limits on market requirements, material conditions, and production situations in manufacturing process, conventional optimization approaches are difficult to obtain optimal economical and technical indices with physical constraints. To optimize several conflicting objects such as production rate, economic benefits, and gas emission, a hybrid-model-based intelligent optimization method that consists of an improved genetic algorithm and derived deep learning is put forward in this paper. Integration of the hybrid model has made modeling and optimizing an indivisible whole, in which the fitness of the genetic algorithm comes from deep neural networks by weighted sum of the output variables that correspond to the input solutions. The recurrent neural network (RNN) with disposition-gated recurrent unit (dGRU) is applied to capture the dynamics of blast furnace by training the model over datasets recorded in the production scene. Meanwhile, the self-adaptive population genetic algorithm (SAPGA) with a varied population size depending on the fitness distribution is used to locate the optimal solutions under current working conditions. The hybrid intelligent optimization model, validated by both numerical tests and practical data, has been running in an ironmaking plant for one year. It has proved to be successful in meeting industry demands by optimizing multiproduction indices simultaneously.