Intelligent Pavement Roughness Forecasting Based on a Long Short-Term Memory Model with Attention Mechanism

The international roughness index (IRI) is one of the key indicators of pavement condition during its service life. Accurate IRI can assist transportation agencies in making maintenance decisions, identifying suitable maintenance approaches, and optimizing the financial plan. Although there are models which have been developed for predicting IRI based on artificial neural networks (ANNs), more features could be included and fused for model training to improve the performance. In this study, a long short-term memory (LSTM) model with an attention mechanism which is able to learn time-series related features with high efficiency and quality is developed to better IRI forecasting. The long-term pavement performance (LTPP) database is used for raw data extraction from different climate and traffic conditions. The prediction performance of different models including LSTM-attention (proposed), LSTM, Levenberg-Marquardt backpropagation (LM-b), and back propagation neural network (BPNN) is evaluated and compared with the pavement data from both South Carolina (SC) and Texas. The results show that the proposed model outperforms the other models on accuracy for both SC and Texas pavements, suggesting potential promising applications on the IRI.