In this paper a modified fuzzy-expert contol algrithm based on combination grey prediction model is proposed to solve the control problems of the coke oven flue temperature system, which usually has the property of high nonlinearity, large time-delay and multiple disturbance. Based on the analytic expression optimized by golden section optimization algorithm, a three-dimensional fuzzy controller is designed and its result is compensated by expert rules compensation module. Meanwhile, in order to make furture control strateges, combination grey prediction model is applied to predict the flue temperature. It is shown that this algorithem can effectively solve the large time-delay control problem in the flue temperature system. Also it speeds up system response and diminishes steady-state error. Actual operation data verification results show that the algorithm has good control quality and weakens the influence of time delay. This work may give a new method for sovling the problem in the coke oven flue temperature system with large time-delay.