Aiming at the pilot’s mental task, human machine interface flight task was implemented at different levels, mental workload changes in different evaluation indicators during the flight was analyzed, a new improved multiple linear regression model was proposed, and a construction method of pilots’ mental workload change prediction model in human-machine interaction was discovered in this study. The results suggested that the mental workload of pilots during flight operation changed significantly in the evaluation indicators of response time, accuracy rate, NASA_TLX and the SDNN, while the evaluation indicators of RRI Count, Max RRI, Minimum RRI, Mean RRI, and Max/Min did not change significantly. By the improved multiple linear regression model, the individual mental workload with different levels of flight difficulty can be predicted and ranked, with average prediction accuracy of 87.5%. Accordingly, the proposed prediction model in this paper fit well with the measured data, and can accurately reflect the characteristics of flight deck human-machine interaction mental workload change, which can provide a basis for ergonomic evaluation and optimization design of future aircraft flight deck human-machine interaction.