The strain monitoring data of bridge structure health monitoring has strong tendency and randomness, in order to improve the prediction accuracy of the data, the traditional single autoregressive integral moving average model (ARIMA) and BP neural network prediction model are weighted and combined, and the two methods are applied to the prediction of the strain monitoring data recorded by the bridge structural health monitoring system of a river crossing bridge in Jiangxi Province. The results show that when the first mock exam is used alone, the BP neural network has better prediction effect than the ARIMA model.The prediction accuracy of weighted and combined models are better than that of single model, and the difference between the sum of squares error (SSE) of weighted model and combined model with BP neural network model is the largest, up to 50.23% and 49.87% respectively.By comparing the error indexes of the weighted model and the combined model, it is found that the prediction accuracy of the two prediction models is very close.The error envelope range of single prediction model is larger than that of the other two models, and the total error sum of ARIMA model is about 50με, the total error sum of BP neural network model is about 30με, the total error sum of the weighted model is about 21.09με, the total error sum of the combined model is about 20.97με. After analysis, both the weighted prediction model and the combined prediction model can predict the SHM strain of the bridge.