Microwave heating, an alternative heating method, can directly interact with objects to be heated. This method will dramatically improve energy utilization rate, which is clean, energy-saving and emission reduction. According to the nonlinear change of temperature when industrial material is used as microwave heating load, regarding the dimensional and mass parameters in microwave industrial heating processes as research objects, and also based on the functional-linked neural network to extract the deep features of sample data, a cuckoo search algorithm is proposed to optimize the parameters of BP neural network, thus establishing the industrial microwave heating temperature prediction model based on the "data driven" method. Simulation results show the accuracy and instantaneity of the temperature prediction model proposed in this paper.