Effective classification of autonomous vehicle (AV) driving behavior emerges as a critical area for diagnosing AV operation faults, enhancing autonomous driving algorithms, and reducing accident rates. This paper presents the Gramian Angular Field Vision Transformer (GAF-ViT) model, specifically designed for analyzing AV driving behavior. The GAF-ViT model is developed upon a novel integration of three key components: GAF Transformation Module, which transforms multivariate driving behavior representative sequences into multi-channel images; Channel Attention Module, which prioritizes relevant behavioral features to enhance classification effectiveness; and Multi-Channel ViT Module, which employs advanced image recognition techniques to accurately classify the resulting multi-channel driving behavior images. This framework not only facilitates detailed analysis of complex multivariate driving behavioral data but also leverages the capabilities of vision-based pattern recognition methods to uncover subtle driving behavior nuances. Experimental evaluation on the Waymo Open Dataset of trajectories demonstrates that the proposed model outperforms baseline models, achieving state-of-the-art performance. Furthermore, an ablation study effectively validates the efficacy of individual modules within the model.
Very short term load forecasting predicts the load over one hour into the future in five minute steps, and is important in resource dispatch and area generation control. Effective forecasting, however, is difficult in view of noisy real-time data gathering and complicated features of load. This paper presents a method based on multilevel wavelet neural networks with novel pre-filtering. The key idea is to use a data pre-filtering method to detect and eliminate spikes within load, apply the wavelet technique to decompose the load into several frequency components, perform appropriate transformation on each component, and feed it together with other appropriate input to a separate neural network. Numerical testing demonstrates the significant value of data pre-filtering and multilevel wavelet neural networks, and shows that our method provides accurate forecasting.
11 September 2021 in Transportation Research Record: Journal of the Transportation Research Board
As congestion levels increase in cities, it is important to analyze people’s choices of different services provided by transportation network companies (TNCs). Using machine learning techniques in conjunction with large TNC data, this paper focuses on uncovering complex relationships underlying ridesplitting market share. A real-world dataset provided by TNCs in Chicago is used in analyzing ridesourcing trips from November 2018 to December 2019 to understand trends in the city. Aggregated origin–destination trip-level characteristics, such as mean cost, mean time, and travel time reliability, are extracted and combined with origin–destination community-level characteristics. Three tree-based algorithms are then utilized to model the market share of ridesplitting trips. The most significant factors are extracted as well as their marginal effect on ridesplitting behavior, using partial dependency plots for interpretation of the machine learning model results. The results suggest that, overall, community-level factors are as or more important than trip-level characteristics. Additionally, the percentage of White people highly affects ridesplitting market share as well as the percentage of bachelor’s degree holders and households with two people residing in them. Travel time reliability and cost variability are also deemed more important than travel time and cost savings. Finally, the potential impact of taxes, crimes, cultural differences, and comfort is discussed in driving the market share, and suggestions are presented for future research and data collection attempts.
Travel time variability (TTV) is a crucial indicator of transportation network performance, assessing travel time reliability and delays. This study investigates TTV metrics within the context of shared mobility using probe data from transportation network companies (TNCs) in Chicago, Los Angeles, and Dallas–Fort Worth. Eight reliability metrics are analyzed and compared for each origin–destination (OD) pair in the network, including standard deviation (SD), the Planning Time Index (PTI), the Travel Time Index (TTI), the Buffer Index (BI), On-time Measures PR (alpha), and the Misery Index (MI), to evaluate their effectiveness in clustering OD pairs using K-means clustering. The findings confirm that SD, PTI, and MI are particularly effective in measuring travel time reliability and clustering within urban systems. This study identifies the most unbalanced supply–demand OD pairs/regions in each city, noting that low/medium-SD clusters around metropolitan airports indicate stable travel times even in high-demand zones, while high-SD clusters in downtown areas reveal significant traffic demands and unreliability. These patterns become more pronounced in study areas with multiple city centers. This study highlights the need for targeted strategies to enhance travel time reliability, particularly in regions like Dallas–Fort Worth, where public transportation alternatives are limited.
Taxis and Transportation Network Companies (TNCs) are important components of the urban transportation system. An accurate short-term forecast of passenger demand can help operators better allocate taxi or TNC services to achieve supply–demand balance in real time. As a result, drivers can improve the efficiency of passenger pick-ups, thereby reducing traffic congestion and contributing to the overall sustainability of the program. Previous research has proposed sophisticated machine learning and neural-network-based models to predict the short-term demand for taxi or TNC services. However, few of them jointly consider both modes, even though the short-term demand for taxis and TNCs is closely related. By enabling information sharing between the two modes, it is possible to reduce the prediction errors for both. To improve the prediction accuracy for both modes, this study proposes a multi-task learning (MTL) model that jointly predicts the short-term demand for taxis and TNCs. The model adopts a gating mechanism that selectively shares information between the two modes to avoid negative transfer. Additionally, the model captures the second-order spatial dependency of demand by applying a graph convolutional network. To test the effectiveness of the technique, this study uses taxi and TNC demand data from Manhattan, New York, as a case study. The prediction accuracy of single-task learning and multi-task learning models are compared, and the results show that the multi-task learning approach outperforms single-task learning and benchmark models.