Abstract: The quality of coke has a great effect on the furnace process. In order to solve the problem of large amount of calculation and inspection of coke quality prediction linear method, the coke quality prediction model based on DE-BP neural network is established on the basis of analyzing the factors affecting coke quality, which uses principal component analysis method to determine the parameters of the input vector with the coal and coke ash quality indicators A_d, sulfur S_td, crushing strength M₄₀, abrasion resistance M₁₀ as the output vector prediction. The result of simulation shows that the relative errors of real and estimated values of the indicators are below 4%, overcoming the low prediction precision of BP neural network and the shortcoming of easily trapped in local minimum. It can meet the requirements of production process and has a certain value for coke production.

Key words: BP neural network, coke quality, differential evolution algorithm, predictive model