During flight, pilots need to receive a large amount of information in a short time and make correct judgments and decisions. The cognitive processes such as perception, judgment and decision-making will be affected by excessive cognitive load and affect flight safety. Firstly, the physiological data of flight cadets during different flight missions were obtained through flight simulation experiments; Then, the characteristics of RESP and ECG signals were extracted by time-domain and frequency-domain analysis, and the indexes that can reflect the level of cognitive load are selected by statistical methods. Finally, in the light of ensemble learning algorithm, combined with support vector machine, k-nearest neighbor, artistic neural network and other methods, a cognitive load evaluation model is established based on multiple physiological signals. Furthermore, it is compared with single models. The results show that the pilot cognitive load evaluation model established in this paper has a high accuracy rate and can better reflect the pilot’s cognitive load level.