PK
P.K. Krishnakumari
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
10 records found
1
A comprehensive understanding of shippers’ preferences can help transport freight forwarders create targeted transport services and enhance long-term business relationships. Nevertheless, limited research examined the benefit of considering shippers’ preferences in the decision-making of synchromodal transport planning and the collection of relevant data is still not straightforward.
This research proposes an innovative framework to learn shippers’ preferences in synchromodal transport operations and optimize transport services accordingly. A preference learning method is developed to capture shippers' preferences through pairwise comparisons of transport plans. In order to model the underlying complex nonlinear relationships and detect heterogeneity in preferences, artificial neural networks are employed to approximate shippers' utility for a specific plan. Based on the learned preference information, a synchromodal transport planning model with shippers’ preferences (STPM-SP) is proposed, with the objectives of minimizing the total transportation cost and maximizing shippers’ satisfaction. An Adaptive Large Neighborhood Search algorithm is developed for solving this optimization problem. This algorithm takes into account the two different objective functions and searches for Pareto solutions to the planning problem.
A case study is conducted based on the European Rhine-Alpine corridor to demonstrate the feasibility and effectiveness of the proposed methodological framework. Basic discrete choice models, binary logit models, are used as benchmarks for preference learning and the synchromodal transport planning model without preferences (STPM) is used as the benchmark for planning. The results show that the proposed preference learning method has better predictive power than the baseline model, achieving higher accuracy and lower variation. With the consideration of shippers’ preferences, STPM-SP can significantly increase shippers' satisfaction with transport services. Scenarios with different types of preferences are tested and results show that the average of maximum improvements in satisfaction reached 37.76%. This research contributes to learning shippers' preferences in the transport operation process and highlights the importance of incorporating these preferences into the decision-making process of synchromodal transport planning. ...
This research proposes an innovative framework to learn shippers’ preferences in synchromodal transport operations and optimize transport services accordingly. A preference learning method is developed to capture shippers' preferences through pairwise comparisons of transport plans. In order to model the underlying complex nonlinear relationships and detect heterogeneity in preferences, artificial neural networks are employed to approximate shippers' utility for a specific plan. Based on the learned preference information, a synchromodal transport planning model with shippers’ preferences (STPM-SP) is proposed, with the objectives of minimizing the total transportation cost and maximizing shippers’ satisfaction. An Adaptive Large Neighborhood Search algorithm is developed for solving this optimization problem. This algorithm takes into account the two different objective functions and searches for Pareto solutions to the planning problem.
A case study is conducted based on the European Rhine-Alpine corridor to demonstrate the feasibility and effectiveness of the proposed methodological framework. Basic discrete choice models, binary logit models, are used as benchmarks for preference learning and the synchromodal transport planning model without preferences (STPM) is used as the benchmark for planning. The results show that the proposed preference learning method has better predictive power than the baseline model, achieving higher accuracy and lower variation. With the consideration of shippers’ preferences, STPM-SP can significantly increase shippers' satisfaction with transport services. Scenarios with different types of preferences are tested and results show that the average of maximum improvements in satisfaction reached 37.76%. This research contributes to learning shippers' preferences in the transport operation process and highlights the importance of incorporating these preferences into the decision-making process of synchromodal transport planning. ...
A comprehensive understanding of shippers’ preferences can help transport freight forwarders create targeted transport services and enhance long-term business relationships. Nevertheless, limited research examined the benefit of considering shippers’ preferences in the decision-making of synchromodal transport planning and the collection of relevant data is still not straightforward.
This research proposes an innovative framework to learn shippers’ preferences in synchromodal transport operations and optimize transport services accordingly. A preference learning method is developed to capture shippers' preferences through pairwise comparisons of transport plans. In order to model the underlying complex nonlinear relationships and detect heterogeneity in preferences, artificial neural networks are employed to approximate shippers' utility for a specific plan. Based on the learned preference information, a synchromodal transport planning model with shippers’ preferences (STPM-SP) is proposed, with the objectives of minimizing the total transportation cost and maximizing shippers’ satisfaction. An Adaptive Large Neighborhood Search algorithm is developed for solving this optimization problem. This algorithm takes into account the two different objective functions and searches for Pareto solutions to the planning problem.
A case study is conducted based on the European Rhine-Alpine corridor to demonstrate the feasibility and effectiveness of the proposed methodological framework. Basic discrete choice models, binary logit models, are used as benchmarks for preference learning and the synchromodal transport planning model without preferences (STPM) is used as the benchmark for planning. The results show that the proposed preference learning method has better predictive power than the baseline model, achieving higher accuracy and lower variation. With the consideration of shippers’ preferences, STPM-SP can significantly increase shippers' satisfaction with transport services. Scenarios with different types of preferences are tested and results show that the average of maximum improvements in satisfaction reached 37.76%. This research contributes to learning shippers' preferences in the transport operation process and highlights the importance of incorporating these preferences into the decision-making process of synchromodal transport planning.
This research proposes an innovative framework to learn shippers’ preferences in synchromodal transport operations and optimize transport services accordingly. A preference learning method is developed to capture shippers' preferences through pairwise comparisons of transport plans. In order to model the underlying complex nonlinear relationships and detect heterogeneity in preferences, artificial neural networks are employed to approximate shippers' utility for a specific plan. Based on the learned preference information, a synchromodal transport planning model with shippers’ preferences (STPM-SP) is proposed, with the objectives of minimizing the total transportation cost and maximizing shippers’ satisfaction. An Adaptive Large Neighborhood Search algorithm is developed for solving this optimization problem. This algorithm takes into account the two different objective functions and searches for Pareto solutions to the planning problem.
A case study is conducted based on the European Rhine-Alpine corridor to demonstrate the feasibility and effectiveness of the proposed methodological framework. Basic discrete choice models, binary logit models, are used as benchmarks for preference learning and the synchromodal transport planning model without preferences (STPM) is used as the benchmark for planning. The results show that the proposed preference learning method has better predictive power than the baseline model, achieving higher accuracy and lower variation. With the consideration of shippers’ preferences, STPM-SP can significantly increase shippers' satisfaction with transport services. Scenarios with different types of preferences are tested and results show that the average of maximum improvements in satisfaction reached 37.76%. This research contributes to learning shippers' preferences in the transport operation process and highlights the importance of incorporating these preferences into the decision-making process of synchromodal transport planning.
To analyze latent multiple specific patterns in the line-based public transport daily delay occurrence, a data-driven explorative analysis of public transport daily delay spatial-temporal distribution pattern is performed based on the k-means clustering algorithm. Firstly, we used aggregated daily delay profile to visualize how the delay is distributed in space and time. And the pattern of daily delay distribution is represented by the image features. Secondly, the image features are extracted by the pre-trained neural network ResNet50, and the output image feature vector are used for implementing unsupervised k-means clustering algorithm. Finally, the k-means clustering results reveal five different daily delay patterns. The distinctive characteristics of these five delay patterns are analyzed and lead to some significant results, which could provide public transport operators with a better understanding of how delays occur on a specific line.
...
To analyze latent multiple specific patterns in the line-based public transport daily delay occurrence, a data-driven explorative analysis of public transport daily delay spatial-temporal distribution pattern is performed based on the k-means clustering algorithm. Firstly, we used aggregated daily delay profile to visualize how the delay is distributed in space and time. And the pattern of daily delay distribution is represented by the image features. Secondly, the image features are extracted by the pre-trained neural network ResNet50, and the output image feature vector are used for implementing unsupervised k-means clustering algorithm. Finally, the k-means clustering results reveal five different daily delay patterns. The distinctive characteristics of these five delay patterns are analyzed and lead to some significant results, which could provide public transport operators with a better understanding of how delays occur on a specific line.
Accurate and trustworthy short-term traffic prediction is crucial in the modern world for the comfort of drivers and decision-makers as it is used to improve the performance of traffic management systems, lessen congestion, increase safety, and shorten journey times. It is possible to discover useful information for network transportation planning, such as forecasting demand, finding bottlenecks, and prioritizing infrastructure improvements, by concentrating on network-wide traffic prediction.
Scholars have developed a variety of methods that can be generally divided into model-based and data-based methods in order to accurately predict network-wide traffic. However, while studies have demonstrated the capability of deep learning methods, particularly convolutional neural networks (CNNs), in predicting traffic states, the complex nonlinear spatial and temporal traffic characteristics, the time-consuming model creation and training, and the unexplained methodology and predictions continue to pose challenges to the task.
This thesis seeks to address these issues by analyzing how deep neural networks identify spatiotemporal traffic patterns for network-wide traffic predictions. To this end, a hybrid CNN-RNN model utilizing a pretrained Inception ResNet v2 feature extractor and a long short-term memory encoder-decoder is constructed to forecast network traffic speeds. A pretrained Inception ResNet v2-based image classifier is built based on the predictions to identify traffic patterns, and Grad-CAM is used to explore how the model identifies them. A freeway network in Amsterdam, Netherlands, is used as a case study.
While it is expected that the hybrid CNN-RNN model can give comparable performance to the state-of-the-art methods, e.g. the DGCN proposed by Li et al., results indicate that it cannot fully capture the
dynamic characteristics of the traffic, nor can it accurately provide predictions. The image classifier failed to identify the distinct traffic patterns as well, despite Grad-CAM's success in indicating locations with rapid changes of values.
Overall, the findings highlight the influence of inductive bias on deep learning models, and the importance of fine-tuning and model-data compatibility. Although further research is required, the conclusions are still beneficial to make informed decisions when choosing appropriate models for future network-wide traffic speed prediction tasks. ...
Scholars have developed a variety of methods that can be generally divided into model-based and data-based methods in order to accurately predict network-wide traffic. However, while studies have demonstrated the capability of deep learning methods, particularly convolutional neural networks (CNNs), in predicting traffic states, the complex nonlinear spatial and temporal traffic characteristics, the time-consuming model creation and training, and the unexplained methodology and predictions continue to pose challenges to the task.
This thesis seeks to address these issues by analyzing how deep neural networks identify spatiotemporal traffic patterns for network-wide traffic predictions. To this end, a hybrid CNN-RNN model utilizing a pretrained Inception ResNet v2 feature extractor and a long short-term memory encoder-decoder is constructed to forecast network traffic speeds. A pretrained Inception ResNet v2-based image classifier is built based on the predictions to identify traffic patterns, and Grad-CAM is used to explore how the model identifies them. A freeway network in Amsterdam, Netherlands, is used as a case study.
While it is expected that the hybrid CNN-RNN model can give comparable performance to the state-of-the-art methods, e.g. the DGCN proposed by Li et al., results indicate that it cannot fully capture the
dynamic characteristics of the traffic, nor can it accurately provide predictions. The image classifier failed to identify the distinct traffic patterns as well, despite Grad-CAM's success in indicating locations with rapid changes of values.
Overall, the findings highlight the influence of inductive bias on deep learning models, and the importance of fine-tuning and model-data compatibility. Although further research is required, the conclusions are still beneficial to make informed decisions when choosing appropriate models for future network-wide traffic speed prediction tasks. ...
Accurate and trustworthy short-term traffic prediction is crucial in the modern world for the comfort of drivers and decision-makers as it is used to improve the performance of traffic management systems, lessen congestion, increase safety, and shorten journey times. It is possible to discover useful information for network transportation planning, such as forecasting demand, finding bottlenecks, and prioritizing infrastructure improvements, by concentrating on network-wide traffic prediction.
Scholars have developed a variety of methods that can be generally divided into model-based and data-based methods in order to accurately predict network-wide traffic. However, while studies have demonstrated the capability of deep learning methods, particularly convolutional neural networks (CNNs), in predicting traffic states, the complex nonlinear spatial and temporal traffic characteristics, the time-consuming model creation and training, and the unexplained methodology and predictions continue to pose challenges to the task.
This thesis seeks to address these issues by analyzing how deep neural networks identify spatiotemporal traffic patterns for network-wide traffic predictions. To this end, a hybrid CNN-RNN model utilizing a pretrained Inception ResNet v2 feature extractor and a long short-term memory encoder-decoder is constructed to forecast network traffic speeds. A pretrained Inception ResNet v2-based image classifier is built based on the predictions to identify traffic patterns, and Grad-CAM is used to explore how the model identifies them. A freeway network in Amsterdam, Netherlands, is used as a case study.
While it is expected that the hybrid CNN-RNN model can give comparable performance to the state-of-the-art methods, e.g. the DGCN proposed by Li et al., results indicate that it cannot fully capture the
dynamic characteristics of the traffic, nor can it accurately provide predictions. The image classifier failed to identify the distinct traffic patterns as well, despite Grad-CAM's success in indicating locations with rapid changes of values.
Overall, the findings highlight the influence of inductive bias on deep learning models, and the importance of fine-tuning and model-data compatibility. Although further research is required, the conclusions are still beneficial to make informed decisions when choosing appropriate models for future network-wide traffic speed prediction tasks.
Scholars have developed a variety of methods that can be generally divided into model-based and data-based methods in order to accurately predict network-wide traffic. However, while studies have demonstrated the capability of deep learning methods, particularly convolutional neural networks (CNNs), in predicting traffic states, the complex nonlinear spatial and temporal traffic characteristics, the time-consuming model creation and training, and the unexplained methodology and predictions continue to pose challenges to the task.
This thesis seeks to address these issues by analyzing how deep neural networks identify spatiotemporal traffic patterns for network-wide traffic predictions. To this end, a hybrid CNN-RNN model utilizing a pretrained Inception ResNet v2 feature extractor and a long short-term memory encoder-decoder is constructed to forecast network traffic speeds. A pretrained Inception ResNet v2-based image classifier is built based on the predictions to identify traffic patterns, and Grad-CAM is used to explore how the model identifies them. A freeway network in Amsterdam, Netherlands, is used as a case study.
While it is expected that the hybrid CNN-RNN model can give comparable performance to the state-of-the-art methods, e.g. the DGCN proposed by Li et al., results indicate that it cannot fully capture the
dynamic characteristics of the traffic, nor can it accurately provide predictions. The image classifier failed to identify the distinct traffic patterns as well, despite Grad-CAM's success in indicating locations with rapid changes of values.
Overall, the findings highlight the influence of inductive bias on deep learning models, and the importance of fine-tuning and model-data compatibility. Although further research is required, the conclusions are still beneficial to make informed decisions when choosing appropriate models for future network-wide traffic speed prediction tasks.
With the electrification in freight transportation, fast-charging facilities are crucial to support enroute charging for long-distance freight trips. The goal of this study is to develop an integrated fast-charging facility planning framework to prepare for the increasing enroute freight charging demand in the Netherlands. Based on highway traffic data, the travel temporal and spatial patterns of general traffic flow and freight flow are extracted and analyzed comparatively. The charging demand is derived from freight traffic data, and network evaluation based on graph theory is used to identify traffic nodes with significance in highway networks. A candidate selection method is proposed to obtain potential deployment locations for charging stations and to-go chargers. On this basis, a multi-period bi-objective optimization model with minimum investment cost and maximum demand coverage is proposed to find optimal solutions for charging facility planning. The case study is formulated based on the Amsterdam highway network. The results show that the proposed model can leverage the potential of early investment to increase the final demand coverage in the last planning horizon.
...
With the electrification in freight transportation, fast-charging facilities are crucial to support enroute charging for long-distance freight trips. The goal of this study is to develop an integrated fast-charging facility planning framework to prepare for the increasing enroute freight charging demand in the Netherlands. Based on highway traffic data, the travel temporal and spatial patterns of general traffic flow and freight flow are extracted and analyzed comparatively. The charging demand is derived from freight traffic data, and network evaluation based on graph theory is used to identify traffic nodes with significance in highway networks. A candidate selection method is proposed to obtain potential deployment locations for charging stations and to-go chargers. On this basis, a multi-period bi-objective optimization model with minimum investment cost and maximum demand coverage is proposed to find optimal solutions for charging facility planning. The case study is formulated based on the Amsterdam highway network. The results show that the proposed model can leverage the potential of early investment to increase the final demand coverage in the last planning horizon.
Including service information in a topological comparison of metro networks worldwide
A comparison of 51 metro networks worldwide using GTFS static data
Master thesis
(2022)
-
S. Vijlbrief, O. Cats, P.K. Krishnakumari, S. van Cranenburgh, R.M. Massobrio
Public transport (PT) plays a vital role in commuting billions of travellers in cities all over the world, providing a mode that is both sustainable and accessible. Metro networks are especially apt at this considering their high-capacity and high-speed operation in urban environments. Comparing different metro networks to one another is a suitable manner for transport planners to gain insights into the characteristics of their networks and which areas of improvement exist. In the field of network science, metro networks have been studied extensively in recent decades. While this provided many new insights in the field of network science, the practical relevance for the field of transport science often remained limited. This limited relevance is primarily caused by the lack of realism of the network representations used, not incorporating the actual operation and service that the network provides. As such, this study proposes a comprehensive comparison of metro networks worldwide including service information. This comparison study includes service characteristics in the form of the total travel time indicator for shortest path calculations, which is a combination of the in-vehicle time, waiting time and number of transfers. The median of this total travel time is taken for each network and compared to that of other networks. This metric in turn is contrasted with network- and city-related characteristics in order to explore relations between these factors and to explain the patterns discovered. From this analysis, it is revealed that the travel time increases with network size. The indicator that is especially apt at explaining the differences in total travel time between networks is the number of stations combined with the average direct station distance. The total travel time methodology applied in this study shows significantly different results to other commonly used methods that rely only on in-vehicle time or hops to calculate shortest path travel times. The waiting time turns out to be the main contributor to these significant differences. Future studies can expand on this by considering other network science indicators and looking further into local indicators. In addition, the methodology could be expanded with more detailed transfer information and other PT modes.
...
Public transport (PT) plays a vital role in commuting billions of travellers in cities all over the world, providing a mode that is both sustainable and accessible. Metro networks are especially apt at this considering their high-capacity and high-speed operation in urban environments. Comparing different metro networks to one another is a suitable manner for transport planners to gain insights into the characteristics of their networks and which areas of improvement exist. In the field of network science, metro networks have been studied extensively in recent decades. While this provided many new insights in the field of network science, the practical relevance for the field of transport science often remained limited. This limited relevance is primarily caused by the lack of realism of the network representations used, not incorporating the actual operation and service that the network provides. As such, this study proposes a comprehensive comparison of metro networks worldwide including service information. This comparison study includes service characteristics in the form of the total travel time indicator for shortest path calculations, which is a combination of the in-vehicle time, waiting time and number of transfers. The median of this total travel time is taken for each network and compared to that of other networks. This metric in turn is contrasted with network- and city-related characteristics in order to explore relations between these factors and to explain the patterns discovered. From this analysis, it is revealed that the travel time increases with network size. The indicator that is especially apt at explaining the differences in total travel time between networks is the number of stations combined with the average direct station distance. The total travel time methodology applied in this study shows significantly different results to other commonly used methods that rely only on in-vehicle time or hops to calculate shortest path travel times. The waiting time turns out to be the main contributor to these significant differences. Future studies can expand on this by considering other network science indicators and looking further into local indicators. In addition, the methodology could be expanded with more detailed transfer information and other PT modes.
Improving the Service of E-bike Sharing by Demand Pattern Analysis
A Data-driven Approach
Master thesis
(2021)
-
Z. Zhang, N. van Oort, S.P. Hoogendoorn, P.K. Krishnakumari, F. Schulte, Max Schalow, Chingiskhan Kazakhstan
E-bike sharing has gradually gained popularity in recent years, while the research in this field is still quite limited. This study applies data-driven methods, mainly demand pattern analysis, to facilitate the development of operational strategies in a cost-effective and operator-friendly way. Demand pattern is analysed in an innovative spatial analytical unit, overlapping circle, which is proven to achieve more beneficial effects than the traditional units (i.e., the administrative units), and hourly clustering is conducted to derive the reallocation strategies by mitigations of imbalance in supply and demand in recurrent hourly clusters. Additionally, this work constructs several indicators to evaluate the strategies in a real-life context, taking both the operator and the users into account. The proposed methodology is applied in a case study, bondi’s e-bike sharing in The Hague with a 4-month time frame from 19-06-2021. There are 5 hourly clusters emerging via agglomerative hierarchical clustering: 1) the first peak hour (16:00-16:59); 2) the second peak hour (17:00-17:59); 3) the first transition hour (18:00-18:59); 4) the second transition hour (19:00-19:59); 5) the off peak (20:00-15:59). The corresponding reallocation strategies are then proposed to alleviate the imbalance in different periods. Additionally, adjustment in the operational areas is suggested by the supply efficiency and trip duration/distance analyses. The results prove that the operational strategies obtained from demand patterns indeed improve the service, with almost 1.5 times ridership, circa 20% decrease in vehicle idle time, compared to the baseline. and a decent monthly net retention rate at around 60%.
...
E-bike sharing has gradually gained popularity in recent years, while the research in this field is still quite limited. This study applies data-driven methods, mainly demand pattern analysis, to facilitate the development of operational strategies in a cost-effective and operator-friendly way. Demand pattern is analysed in an innovative spatial analytical unit, overlapping circle, which is proven to achieve more beneficial effects than the traditional units (i.e., the administrative units), and hourly clustering is conducted to derive the reallocation strategies by mitigations of imbalance in supply and demand in recurrent hourly clusters. Additionally, this work constructs several indicators to evaluate the strategies in a real-life context, taking both the operator and the users into account. The proposed methodology is applied in a case study, bondi’s e-bike sharing in The Hague with a 4-month time frame from 19-06-2021. There are 5 hourly clusters emerging via agglomerative hierarchical clustering: 1) the first peak hour (16:00-16:59); 2) the second peak hour (17:00-17:59); 3) the first transition hour (18:00-18:59); 4) the second transition hour (19:00-19:59); 5) the off peak (20:00-15:59). The corresponding reallocation strategies are then proposed to alleviate the imbalance in different periods. Additionally, adjustment in the operational areas is suggested by the supply efficiency and trip duration/distance analyses. The results prove that the operational strategies obtained from demand patterns indeed improve the service, with almost 1.5 times ridership, circa 20% decrease in vehicle idle time, compared to the baseline. and a decent monthly net retention rate at around 60%.
Automated offline detection of disruptions using smart card data
A case study of the metro network of Washington DC
Master thesis
(2020)
-
Faye Jasperse, Oded Cats, Maaike Snelder, Yilin Huang, Panchamy Krishnakumari
Service reliability is one of the most important performance measures to public transport users. Detecting disruptions helps to measure service reliability, which can be used by public transport operators to improve this reliability. In this thesis, a methodology is described to automatically detect disruptions offline, using smart card data. The day-to-day regularity of delays is investigated using hierarchical clustering on a training set, to distinguish between regular and irregular delays. The clustering result is used to create a probabilistic classifier. This classifier is applied to the test set to find days that do not correspond to a regular pattern: irregular days. After that, disruptions are detected within the irregular days. The outcomes of this study can be applied in multiple ways. Locations where disruptions have occurred can be found and the related passenger delay can be calculated. This can help public transport operators to prioritise which locations to focus on to reduce passenger delays. Furthermore, not only public transport networks, but also other networks can benefit from the outcomes of this study. Speed data of road networks could be used to find disruptions that are caused by accidents, instead of regular traffic jams. On top of that, this study could be used as a step towards real-time disruption detection, for both public transport and road networks.
...
Service reliability is one of the most important performance measures to public transport users. Detecting disruptions helps to measure service reliability, which can be used by public transport operators to improve this reliability. In this thesis, a methodology is described to automatically detect disruptions offline, using smart card data. The day-to-day regularity of delays is investigated using hierarchical clustering on a training set, to distinguish between regular and irregular delays. The clustering result is used to create a probabilistic classifier. This classifier is applied to the test set to find days that do not correspond to a regular pattern: irregular days. After that, disruptions are detected within the irregular days. The outcomes of this study can be applied in multiple ways. Locations where disruptions have occurred can be found and the related passenger delay can be calculated. This can help public transport operators to prioritise which locations to focus on to reduce passenger delays. Furthermore, not only public transport networks, but also other networks can benefit from the outcomes of this study. Speed data of road networks could be used to find disruptions that are caused by accidents, instead of regular traffic jams. On top of that, this study could be used as a step towards real-time disruption detection, for both public transport and road networks.
On-demand transit has become a common mode of transport with ride-sourcing companies like Uber, Lyft, Didi transforming the way we move. With the increase in popularity for such services, the supply needs to adapt according to the demand. For this, the demand needs to analyzed to examine if there are recurrent patterns in them; making it predictable and easily manageable. The identified demand patterns can then be used for optimized fleet management. In this paper, we propose three steps for extracting such demand patterns from travel requests (1) constructing the origin-destination zones by spatial clustering (2) calculating the hourly origin-destination matrix for each day, and (3) temporal clustering to extract the dynamic demand patterns. We demonstrate the three step approach on the open-source Didi taxi data. The data is composed of 1 month (November 2016) of travel requests data from a small area in Chengdu, China with approximately 200 000 rides for a single day on average. It can provide insight into the day-to-day regularity and within-day regularity of the demand patterns in Chengdu.
...
On-demand transit has become a common mode of transport with ride-sourcing companies like Uber, Lyft, Didi transforming the way we move. With the increase in popularity for such services, the supply needs to adapt according to the demand. For this, the demand needs to analyzed to examine if there are recurrent patterns in them; making it predictable and easily manageable. The identified demand patterns can then be used for optimized fleet management. In this paper, we propose three steps for extracting such demand patterns from travel requests (1) constructing the origin-destination zones by spatial clustering (2) calculating the hourly origin-destination matrix for each day, and (3) temporal clustering to extract the dynamic demand patterns. We demonstrate the three step approach on the open-source Didi taxi data. The data is composed of 1 month (November 2016) of travel requests data from a small area in Chengdu, China with approximately 200 000 rides for a single day on average. It can provide insight into the day-to-day regularity and within-day regularity of the demand patterns in Chengdu.