Explainable energy consumption and speed prediction in sustainable cities using deep learning

作     者：Abd El-Latif, Eman I. El-dosuky, Mohamed 

作者机构：Faculty of Science Benha University Benha Egypt Computer Science Department Arab East Colleges Riyadh Saudi Arabia Computer Science Department Faculty of Computers and Information Mansoura University Mansoura Egypt 

出 版 物：《Neural Computing and Applications》 (Neural Comput. Appl.)

年 卷 期：2025年第37卷第8期

页      面：6233-6249页

核心收录：

学科分类：08[工学] 0824[工学-船舶与海洋工程] 0823[工学-交通运输工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Open access funding provided by The Science  Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB) 

主　　题：Sustainable city 

摘      要：The idea of sustainable cities has drawn a lot of attention due to the quick expansion of metropolitan areas as well as the growing problems brought on by resource scarcity and climate change. Cities that prioritize sustainable practices are those that minimize their negative effects on the environment, maximize resource efficiency, and improve the standard of living for their citizens. Therefore, for sustainable cities, this paper uses the vehicle energy dataset (VED) to estimate travel times and calculate vehicle energy consumption. The dataset contains 12,609,170 road elevation tracks, 12,203,044 speed limit tracks, and 12,281,719 speed limit records with direction. An open-source routing engine called Valhalla is utilized to do a variety of tasks, including finding paths, matching maps, and creating maneuvers based on paths. The three primary stages of the suggested model are data pre-processing, feature extraction, and result interpretation. In the data pre-processing stage, null values are first eliminated and data normalization is implemented. Then, three techniques known as the gated recurrent unit (GRU), recurrent neural network (RNN), and long short-term memory (LSTM) are used to optimize the model. Finally, the results are interpreted through the use of SHAP (SHapley Additive explanations) in explainable artificial intelligence (XAI) techniques. The LSTM model yields the best prediction results, achieving 15.2662 RMSE, 11.7266 MAE, and 0.6696 R2 at 8 batch size, according to the evaluation results. Additional experiments are carried out in batch sizes of 8, 16, 32, and *** lowest metrics are produced by batch sizes of 64, while the best metrics are produced by batch sizes of 8. © The Author(s) 2025.