ASENN: attention-based selective embedding neural networks for road distress prediction

Journal of Big Data

Research

Zhenyu Xu¹  Qieshi Zhang¹  Hamad AlJassmi²  Babitha Philip²  Luqman Ali³ 

[1] CAS Key Laboratory of Human-Machine Intelligence-Synergy Systems, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, No. 1068 Xueyuan Avenue Shenzhen University Town, Shenzhen, China;Department of Civil and Environmental Engineering, United Arab Emirates University, 15551, Al Ain, United Arab Emirates;Emirates Center for Mobility Research (ECMR), United Arab Emirates University, 15551, Al Ain, United Arab Emirates;Emirates Center for Mobility Research (ECMR), United Arab Emirates University, 15551, Al Ain, United Arab Emirates;AI and Robotics Lab (Air-Lab), United Arab Emirates University, 15551, Al Ain, United Arab Emirates;College of IT, United Arab Emirates University, 15551, Al Ain, United Arab Emirates;

关键词: Pavement deterioration;    Tabular data;    Road distress parameters;    Prediction models;    Deep learning;   

DOI  :  10.1186/s40537-023-00845-x

 received in 2023-06-26, accepted in 2023-10-17,  发布年份 2023

来源: Springer

【 摘 要 】

This study proposes an innovative neural network framework, ASENN (Attention-based Selective Embedding Neural Network), for the prediction of pavement deterioration. Considering the complexity and uncertainty associated with the pavement deterioration process, two fundamental frameworks, SEL (Selective Embedding Layer) and MDAL (Multi-Dropout Attention Layer), are combined to enhance feature abstraction and prediction accuracy. This approach is significant while analyzing the pavement deterioration process due to the high variability of the contributing deterioration factors. These factors, represented as tabular data, undergo filtering, embedding, and fusion stages in the SEL, to extract crucial features for an effective representation of pavement deterioration. Further, multiple attention-weighted combinations of raw data are obtained through the MDAL. Several SELs and MDALs were combined as basic cells and layered to form an ASENN. The experimental results demonstrate that the proposed model outperforms existing tabular models on four road distress parameter datasets corresponding to cracking, deflection, international roughness index, and rutting. The optimal number of cells was determined using different ablation settings. The results also show that the feature learning capabilities of the ASENN model improved as the number of cells increased; however, owing to the limited combination space of feature fields, extreme depths were not preferred. Furthermore, the ablation investigation demonstrated that MDAL can improve performance, particularly on the cracking dataset. Notably, compared with mainstream transformer models, ASENN requires significantly less storage and achieves faster execution speed.

【 授权许可】

CC BY   
© The Author(s) 2023

【 图 表 】

Fig. 6

Fig. 2

Fig. 2

Fig. 6

Fig. 1

Fig. 6

12951_2015_155_Article_IEq6.gif

41512_2023_158_Article_IEq9.gif

Fig. 8

12951_2017_323_Article_IEq1.gif

Fig. 5

Fig. 3

41512_2023_158_Article_IEq2.gif

Fig. 7

41512_2023_158_Article_IEq1.gif

Fig. 1

Fig. 5

12936_2017_2051_Article_IEq86.gif

12936_2017_1932_Article_IEq15.gif

Fig. 1

Fig. 7

Fig. 3

Fig. 6

12951_2015_155_Article_IEq4.gif

12867_2016_60_Article_IEq1.gif

Fig. 4

Fig. 2

Fig. 9

12867_2016_60_Article_IEq2.gif

Fig. 1

Fig. 1

Fig. 2

Fig. 10

Fig. 2

Fig. 2

Fig. 2

12936_2017_1885_Article_IEq1.gif

Fig. 1

Fig. 3

Fig. 6

Fig. 5

Fig. 2

12888_2023_5299_Article_IEq5.gif

12888_2023_5299_Article_IEq4.gif

12888_2023_5299_Article_IEq2.gif

Fig. 6

Fig. 1

Fig. 4

Fig. 2