Aiming at the wavelength drift problem caused by the influence of ambient temperature on fiber Bragg grating (FBG) strain sensor, the online prediction algorithm that combined particle swarm optimization (PSO) with sliding window extreme learning machine (SWELM) is proposed for temperature compensation. The PSO algorithm was used to optimize the sliding window and the number of neurons in the hidden layer of the SWELM network, which improved the prediction accuracy of the model, and the minimum root mean square error of the model prediction could reach 0.06pm. PSO-SWELM realized online update and wavelength drift prediction of strain sensor data, and differential calculation of real-time measurement data and prediction data completed temperature compensation. PSO-SWELM was compared with SWELM, and the results show that the accuracy of the proposed algorithm is improved by an average of 11.04%, and has good temperature compensation effect.