In response to the instantaneous temperature changes caused by complex operating conditions in the service state of aero-engine, the temperature sensor exhibits hysteresis due to the thermal inertia of its own materials, and the collected temperature signals are susceptible to noise interference. We conducted pre-processing analysis anddynamic compensation research on the dynamic response measured temperature signal of a certain type of aero-engine using platinum resistance sensors in service. By using the optimized CEEMDAN algorithm to extract and filter out mid to high frequency noise features from measured signals, and based on Hilbert transform to filter out small random noise and reconstruct the final signal, the denoising results are characterized by the correlation coefficient with the theoretical response curve of the sensor. On this basis, the ARX model with parameter optimization of the reconstructed signal was used for overall dynamic error compensation. Comparative analysis was conducted through root mean square dynamic error and time constant calculation, and uncertainty of the reconstructed signal was evaluated. The results show that the optimized CEEMDAN and Hilbert transform can more effectively remove noise and reconstruct the original signal, with a correlation coefficient of 99.9% with the sensor response curve and a relative expanded uncertainty of about 3.3%. The ARX model parameter w is relatively large, the maximum reduction in overall dynamic error after compensation is 71.36%, and the time constant is reduced by 2.76 s.