Research on Frost Damage Prediction of Railway Subgrade Based on ARIMA-BP Combo Model
Railway roadbed settlement data are susceptible to various complex factors.In order to reduce the cost of manual overhaul and improve the prediction accuracy,a combination of ARIMA-BP prediction model was proposed.First of all,in-depth analysis was done for seasonal permafrost settlement change rules and its data characteristics respectively,to construct the roadbed freezing prediction model based on ARIMA(differential autoregressive sliding average)model and BP neural network model.The analysis found that a single ARIMA model has a strong prediction performance for short-term linear settlement changes,but performs poorly in long-term nonlinear changes in the settlement prediction.BP neural network according to error inverse updated model weights,which deeply excavated the long-term nonlinear change trend of the sequence.Therefore,a combined ARIMA-BP model was proposed,which used BP neural network to complete the initial prediction and fit the residual sequence of the original sequence,and then adopted ARIMA model for residual prediction,and combined the two prediction results to obtain the combined prediction results.Experiments show that,the RMSE of the proposed combined model is 76.9%lower than that of ARIMA prediction method,58.6%lower than that of BP neural network prediction method,and 45.9%lower than that of the combination prediction method based on the optimal weights of ARIMA-BP,which proves that the method can better express the trend of changes in the settlement of roadbed,and play a role of early warning for the safety of railway traffic.
railwayroadbed settlementARIMA modelBP neural networkcombined model
NETL
NSTL
万方数据
