Sports big data-driven combined with time series analytics greatly improves the prediction of competitive status fluctuations in elite athletes. In this study, an athletic state prediction framework incorporating Informer-based time series analysis is constructed based on multi-source sports big data. The model is utilized to learn the physiological data characteristics of athletes in different competitive states. The prediction performance of the model is specifically analyzed by combining the root mean square error and other evaluation indexes. Athletic state case prediction of 30 elite athletes is realized by deep learning feature data. The physiological data collected in the study show that athletes in the best competitive state generally have body temperatures between 36.5°C and 38°C, and other indicators are maintained in a relatively normal range. The model in this paper achieves good prediction results in different athletic states, with F1 values above 0.94 and prediction errors between 0 and 0.2. The model has a small error value in performing physiological data prediction, such as the absolute error of body temperature is between 0~0.5℃, which realizes the accurate prediction of elite athletes’ competitive state.
