In this paper, an improved RBF prediction model was proposed to solve the problems of cumbersome calculation and many influencing factors in the process of coking energy consumption. In the energy consumption prediction model, target flue temperature, flue suction, moisture, volatile matter and coking time were taken as input variables, and coking energy consumption as output variables. Because RBF network had many shortcomings such as poor learning ability and slow convergence speed, an energy consumption prediction model based on improved differential evolution algorithm was proposed. By using the differential evolution algorithm with strong global search ability, the optimal values of the center value, width and output weight of the basis function in RBF network were selected as the center value, width and output weight of RBF neural network. The results show that the improved RBF prediction has high accuracy, stability and training speed, which is of great significance for reducing coking energy consumption, increasing coke output and improving economic benefits of enterprises.