2. A NTECEDENTES BIBLIOGRÁFICOS Y ESTADO ACTUAL DEL TEMA
2.3 Asesinos en serie y crueldad animal
2.3.1 Características de los maltratadores de animales
This chapter proposes a method for passenger route choice deduction from smart card data. The passenger route choice deduction is important for analyzing passenger service in terms of travel time, which is dependent on route choice. Thanks to the unique data set resulting from conductor checks, the passenger route choice deduction can be validated at the journey specific level. The journey validation is of a higher accuracy than previous validation methods based on train capacity utilization proposed by Kusakabe et al. (2010). Especially when not all passengers travel by smart card, validation on a per journey level is more accurate than validation on an aggregate level over all passengers. To our knowledge, this is the first paper to include validation of this method, and to specifically evaluate different methods for Route Generation and Route Selection. Evaluations are based on a real life data set from Netherlands Railways.
We compared several Route Generation methods based on different network formula- tions of the public transport network. We showed that the network formulation strongly influences the ability of finding all reasonable routes. Although no Route Generation method is known to find all reasonable routes (Fiorenzo-Caralano, 2007), our best method finds routes for 93 % of the journeys in our validation sample. Route Generation methods based on minimum cost searches in the Extended Network that allow penalizing transfers perform significantly better than minimum cost searches in the Basic Network formula- tion. Moreover, although the set of routes is increased by using data from earlier days to learn routes from the conductor checks, a minimum cost based generation method performs almost as well as this learning algorithm.
Through the implementation of 5 route Selection Rules we compare previously made assumptions on passenger route choice including first departure by Frumin and Zhao (2012), longest route by Kusakabe et al. (2010) and latest arrival by Sun et al. (2012), with two additional rules of least transfers and a balancing rule between longest journey and least transfers. We measure performance based on the subset of journeys that were checked by a conductor during their journey. We find that previously assumed selection rules perform poorly in a complex network such as that of Netherlands Railways. Especially the first departure rule performs badly, as was also concluded by Schm¨ocker et al. (2013). However, a single Route Selection Rule weighing travel time and transfers, STA, is able to assign 95% of the journeys in our sample correctly. We find that this method works well on both regular days and in case of disruptions. The resulting data is suitable for analyzing passenger service, as illustrated by our case study, but could also be used for, for instance, calibrating route choice models as done by Schm¨ocker et al. (2013).
2.9 Conclusions and Discussion 39
Currently conductor checks are used to broaden the route set. Alternatively, a learning algorithm could be used to reduce the set of routes. Future research could develop statis- tical arguments that define when to eliminate a route, or where to gather additional data to validate the elimination of a route. Moreover, results of the Route Deduction method could be used to fine-tune the parameter settings of the Extended Network, possibly aim- ing to set them in such a way that, based on a single setting, all possible routes can be found. This would increase the computational speed, reduce the number of candidate routes, and possibly increase the performance of the method.
Chapter 3
Forecasting Passenger Flows during
Disruptions based on Smart Card
Data
This Chapter is an adaptation of Van der Hurk et al. (2013), which was presented at the IAROR Conference in Copenhagen, 2013.
Co-authors: L.G. Kroon, G. Mar´oti and P.H.M. Vervest.
3.1
Introduction
Although timeliness of operations is one of the key performance indicators of public trans- port operators, disruptions causing delays unfortunately do occur in these systems. When they do occur passengers expect quick solutions that minimize their inconvenience. De- tailed data on passenger demand is required to minimize passenger inconvenience. Re- cently introduced automated fare collection systems, such as smart card ticketing, generate data on passenger journeys. These data are often not available in real-time, and therefore forecasts of passenger demand are needed to anticipate passenger demand in disruption management practices. This chapter presents a framework for forecasting passenger flows in case of disruptions using these new data.
Traditionally, demand forecasts in (public) transport are split into two phases: in the first phase, the origin-destination matrix (OD matrix) is estimated for a specific time period, for instance for a day or a (peak) hour. In the second phase, the OD matrix demand is assigned to paths through the public transport network. The assignment to paths aims to estimate the choices of passengers according to a behavioral model. Nguyen
and Pallottino (1988) and Spiess and Florian (1989) introduced the concepts of hyperpaths and travel strategies to describe different observed path choices of passengers. Trozzi et al. (2013) propose a model for assigning passengers to paths in networks with congestion. Anderson et al. (2014) estimate a model for multi-modal route choice using survey data. Schm¨ocker et al. (2013) formulate and calibrate a route choice model for bus passengers based on smart card data. Generally these models are aimed at estimating passenger route choices in undisrupted situations.
Newly available smart card data contains information on the OD matrix, and is suit- able to deduce information on the routes chosen by passengers (Chapter 3. Unlike OD- matrices, it does not aggregate journeys over time but rather contains information about both the departure and arrival time of individual passengers. It is available for all pas- sengers using smart cards, which is generally a much larger number of passengers than the number of participants in a survey. Moreover, smart card data does not suffer from inaccuracies due to errors made by survey participants in their recollection of past jour- neys. Passenger counts per trip, either estimated by conductors, resulting from samplers specifically counting the number of passengers on a specific trip, or from automatic sensors in the vehicle, provide information on the crowdedness and overall demand per trip, but not on the origin and destination of passengers. This information is especially important when aiming to estimate a change in passenger flows due to a disruption. Consequently smart card data provides a unique combination of detailed data per journey for a large set of passengers.
The objective of this chapter is to propose a framework for forecasting demand during disruptions. Specifically, the paper proposes a model to forecast the number of passen- gers per planned path using smart card data, rather than to separate the estimation of OD-matrix and path choice models. These forecasts could be more accurate than tra- ditional methods, as they use the combined information of origin, destination and route choice which can all be derived from smart card data. These forecasts for the number of passengers per planned path will provide the likely location and destination of passengers at the start of a disruption. Thereby these forecasts can assist in estimating the change in passengers’ routes from the start of the disruption without changing route choices of passengers before the start of this disruption. The derived estimates are also aimed at supporting dispatchers in their decision making process during a disruption by providing insight into the passenger inconvenience resulting from the disruption.
The framework consists of three steps (i) analyze the smart card data to derive the
combined OD matrix and path assignment over time, (ii) generate forecasts on the number