ABSTRACT The effective control of coke oven flue temperature is discussed in this paper. A predictive control method based on orthogonal neural network for coke oven flue temperature is proposed. Firstly, the orthogonal neural network has good fitting ability, global optimisation ability, generalisation ability and fast training speed, which is suitable for the modelling of coke oven heating process. Then, the nonlinear system of coke oven is transformed into a time-varying linear model by linearisation method, and generalised predictive control algorithm can be introduced. In order to improve the performance of the generalised predictive control algorithm, this paper adopts the stair-like generalised predictive control algorithm. The algorithm cannot only avoid the on-line solution of the inverse matrix, greatly reducing the amount of calculation, but also improve the robustness and anti-interference ability. At the same time, some correction strategies are adopted to reduce the computational complexity of the model. Finally, according to the actual working conditions, three case studies under normal working condition, the change of setting temperature and under random disturbance are performed out. Compared with the comparative control method, the simulation results show that the proposed control method is effective.