This article deals with the processing of data on the entrances and exits of passengers for railway stations in Moscow and the suburbs. Smart transport cards are used worldwide in transport applications as a payment tool. So, for railways (cities) its usage creates the big and constantly updated collections of transactions data from cards validation equipment. The deployment model for railways in Moscow region allows us to know exactly the starting and ending points of the each route. This detailed information allows us to obtain generalized information on the modes (models) of the actual use of the railway transport. The detected travel patterns could be mapped to the model of the social and economic behavior of residents of the capital region. And vice versa, we can use known artifacts of the behavior of the inhabitants of the region as the search patterns for transport data.The conclusion that mobility is one of the main characteristics and one of the key components of a smart city is a well-known fact.
This paper deals with the processing of data on the check-in (entrances) and check-out (exits) of passengers for railway stations in Moscow and the suburbs. Within the framework of the existing model of moving around suburban and urban railways, each passenger presents (validates) his travel document twice: at the entrance to the railway station before the trip (check-in) and at the exit from the railway station after the end of the trip (check-out). This feature, together with the unique identity of the travel document, allows us to accurately know the starting and ending points of the route. Accordingly, it becomes possible to analyze this displacements data in order to obtain the generalized information on the modes (models) of the actual use of the railway communication. It seems to us that the information obtained during this analysis can be useful for railways for assessing their activities and planning changes, and for city services to assess the ongoing changes in the urban environment and planning future changes.
The aim of the work is to search for patterns of travel of railway passengers and map these patterns to the model of the social and economic behavior of residents of the capital region. It is also possible to consider the inverse problem - the search (the con rmation) of known artifacts of the behavior of the inhabitants of the region in the data on the movements.
As we pointed out above, in the existing system, a railway ticket (travel
document) is presented twice - at the entrance and at the exit. Accordingly, for
each trip, we know the station where the ticket was used at the entrance and
the station where the ticket was used at the exit. In terms of social networks, we
know the pair - check-in and check-out. It means, by the way, a pro table
difference, for example, from information on the validation of travel documents in
the metro or buses. There (in Moscow) we have only information about the
entrance. Accordingly, to obtain information about the starting and ending points
of the trip, it is necessary to use any heuristic algorithm [
To date, detailed information on the passengers' activity at railway stations is practically not used. Perhaps, one of the few applications that could be mentioned here is the calculation of the total numbers of passengers by stations. It seems to us that the above-mentioned detailed data contain much more valuable information that re ects not only the patterns of the use of railway transport in the city but also allows us to identify some other artifacts of city life.
If we can explain the relationship of information obtained on the basis of registration data with some patterns of behavior in the city, then tracking the changes in the data on the stations (which is technically feasible on the part of the railway, for example) will allow us to determine (or predict) some changes in the life of the city. In other words, changes that can be identi ed in the process of constant monitoring of registration data will indicate a change in the processes in the city. Accordingly, data on passages can be used to track changes in patterns of behavior of urban residents (passengers). And vice versa, understanding the relationships will allow us to predict how changes in the city will a ect the use of the railway.
The rest of the paper is organized as follows, In Section 2, we describe related works. In Section 3, we describe railway data processing and discovered links with urban life. 2
Modeling urban behavior by mining geo-tagged data is a popular topic for
research [
For transport data, the models should look slightly di erent. For example,
for urban railways, the routes of passengers are precisely known. Activities,
naturally, are also limited. The paper [
The paper [
In our calculations, we've used the following de nition for the Hillinger coefcient. Let pi(x) be distribution of probability density function (i= 1,2, ... ,N ). The Hillinger coe cient among these variables is:
R =
N
X(Y pi(i))1=N
x i=1
(1)
The value of R is between 0 and 1. The larger the R, the more related the
probability density functions [
In general, for our kind of research, time-dependent analysis of urban
movement patterns [
Data for analysis for 2016/2017 years on suburban and city stations of the Moscow region was provided by the Center for Digital High-Speed Transportation Systems of the Russian University of Transport. The data les contain the following information for each pass (input or output): date and time, type of the event: entrance or exit, a current station, a station where the passenger arrived (if this is an exit), type of the tari (price characteristic): full or preferential, type of the ticket: one-time one-way ticket, one-time round-trip ticket, subscription (reusable ticket, travel card)
What was included in the rst phase of our research? Firstly, these are the usage patterns of the stations. We can assume that there are di erences in how passengers use railway stations. Moscow city (more precisely, workplaces in Moscow) is the center of attraction, and accordingly, we can expect that for the suburban stations (outside the city boundary in Figure 1) there will be a peak at the entrance of passengers in the morning hours (Figure 2).
These peaks at the entrance (see 1 in Figure 3) through the time t spent on the road should pour into the peaks at the exits from the stations placed in the city (see 2 in Figure 3).
In general, the check-ins between the two peak hours, as well as after the second peak hour, corresponds probably, to the non-obligatory activities. This is the simplest and most obvious pattern. By attenuation of the peaks to the exits, we can determine at which stations those who come to the city leave the railway and transfer to other modes of transport. Of course, for each direction of railroad, these stations will be di erent.
Another possible direction for future research is the analysis of this damping. At least, the primary results show that it is not absolutely stable for the chosen direction. Passengers from time to time change their habits of leaving the train in the morning. This lasts 1-2 days (not for all directions), after which everything returns to the basic scheme.
An additional nding in this connection: the detection of peaks on the exits at the stations where the geo-information system does not show connections with other modes of transport. Why do the passengers go to this station? A possible explanation is the presence of some point of attraction (business center, shopping center, etc.)
According to this simple model, we should see the opposite picture in the
evening. The peaks at the entrances to the "internal" (urban) stations and the
peaks on the exits (with a time gap) at the "external" stations. Findings that
were made here: the picture is not symmetrical. Passengers do not necessarily
leave from the stations they came to (of course, we operate only with quantitative
di erences - there is no information on passengers). Probably, we can propose a
natural explanation for this. There is some mobility upon completion of work.
Outgoing tra c (large peaks) is tied to stations where there are connections to
other modes of transport (where transport accessibility is better).
It corresponds, for example, with results presented in [
Hypotheses that require veri cation: asymmetry in tra c is greater in the warm season (higher mobility) and on Friday (for the same reasons).
The analysis shows the presence of stations without pronounced peaks in the morning and evening hours. At this moment, the reasonable explanation is the conclusion that the stations are not connected with work tra c. For example, for stations outside the city, this is more typical for holiday villages. In the city, it is typical for stations in the former industrial zones (where mass housing construction is only being developed). Another possible explanation for the absence of peaks is the linking of the station to some large transport (interchange) nodes, where there is always a large passenger tra c (so, working migration does not add anything signi cant to the constantly existing tra c).
There were no cases of the presence of one peak (regular) at the entrance and / or exit in the examined dataset. It should be noted that there is no reasonable "urban" explanation for such a hypothetical situation.
Another point related to a tra c, is the con rmation (or refutation) of the above-mentioned found move pattern on the data for the day o (e.g., Sunday, Saturday). Obviously, for stations with predominantly "working" tra c, we should see the absence of peaks in the morning and evening hours on weekends. Single outbursts are possible and connected, most likely, with some mass events held over the weekend.
As per classi cation of stations by tra c patterns, we can follow the model
presented in [
The next moment, which was investigated in the work - is the ratio of onetime tickets versus reusable tickets (travel cards). The idea of this comparison is based on the following fact. Reusable tickets (travel cards) are cheaper. Accordingly, those who travel constantly, will most likely use them. Therefore, a greater percentage of travel cards corresponds to more constant (robust) tra c. Tickets are bought before the trip, accordingly, it would be interesting to compare the ratio of one-time and reusable tickets at the entrance to the stations. Accordingly, stations with deviations from average ratios of ticket types were identi ed. So far, the explanation that is being considered here is the availability of interchange transport nodes near such stations, from where the "random" passengers for the railway arrived.
One-time tickets are of two types - one way and round-trip. The next possible step is to analyze the ratio of such tickets. Also interesting is the question of changing the ratio of one-time tickets and travel cards on the days of the week ( rst of all - comparing working days and days o ). The increase in the number of one-time tickets at the weekends shows that these days, the railway is really "acquiring" new passengers who do not use the railroad for a week.
In addition to the above-mentioned Hillinger coe cient, we also used
methods of analyzing the similarity of time series. As per [
So-called lock-step measure in this classi cation is well-known Euclidean distance. It is de ned as the square root of the sum of the squared di erences between corresponding data points in two time series data. As it is mentioned in the all statistical papers, the main problem of the Euclidean distance is the need to have the same length for time series. It is not a problem in our case, because we can, for example, divide the day into ve-minute intervals and thus construct the same length-of-time time series for the number of inputs and outputs.
So-called elastic measures use dynamic programming to align sequences with
di erent lengths. The typical example is so-called DTW [
In our work, we've successfully used a shape-based similarity measure,
socalled Angular Metric for Shape Similarity (AMSS) [