SOFIA UNIVERSITY EXPERTS PRESENT METHODOLOGIES FOR ROUTE OPTIMIZATION AND ARRIVAL TIME PREDICTION IN THE ON-DEMAND PUBLIC TRANSPORTATION PLATFORM
Within INNOAIR, Sofia University experts presented methodologies to serve the on-demand public transportation platform for route optimization and arrival time prediction, by providing usable algorithms to be implemented in the platform, including:
• methodological approaches to set up the general optimization problem for the platform;
• data mapping needed for a viable solution;
• models for predicting the time of travel between stops.
Additionally, a focus was placed on the origin-destination calculations based on big data of public transport passenger flow – usually obtained through large sample surveys which are costly and done annually at best. We set an alternative methodological approach by using fully anonymized individual cell phone positioning data.
The primary data source consists of a set of anonymized and aggregated mobile device logs generated every time the mobile device interacts with the operator network due to voice calls, text messages, or data sessions, which are stored by the operator for billing purposes.
The data set analyze contains information about trips from the "Manastirski Livadi" district to 619 different destinations in the city of Sofia for a period of 6 months.
Sofia city has a total of 98 districts. „Manastirski Livadi“ district is one of the fastest growing and dense.
In order to create optimal public transport routes and timetables, it is necessary to analyze the potential passenger flow between any two stops (1) and (2) in the service area, taking into account the fact that citizens use many and different modes of travel.
For each starting destination we can define an indicator NTnj(k) which takes the value 1 if we have at least five trips to destination n in the time interval j on day k or the value 0 if we do not have at least five trips in the corresponding time slot.
The NTnj(k) indicator allows us to summarize the passenger flow generated by each starting destination in an OD matrix of trips A of the following type:
For OD matrix of trips, A , we apply k-means clustering that allows any large number of destinations to be reduced to three fully defined main groups:
(1) destinations to which there are trips primarily in the peak hours of 08:00-11:00 and 18:00-21:00;
(2) destinations to which there is a passenger flow throughout the day
(3) destinations where there are a limited number of trips in a very small number of hour slots.
For origin bus stop "Manastirski Livadi" we created 619 rows OD matrix of trips A. The k-means clustering shows the following results:
The Autocorrelation Function (ACF) diagram of one of the time series describing the number of trips to a specific destination from cluster 2 measured on an hourly basis.
The ACF shows that there are significantly large autocorrelations, particularly at every 24th lag.
This indicates that the data have 24th degree seasonal integration. This is typical of all cluster 2 and cluster 3 time series. Such time series can be well modeled by SARIMA (seasonal autoregressive integrated moving average) models.
Based on the approach we may find SARIMA models to predict and model the citizen travel flows to destinations from Cluster 2. We demonstrate such a data model for “Manastirski Livadi“ district in Sofia, Bulgaria. All coefficients are statically significant. An analysis of the model's errors shows that it describes the dynamics in the passenger flow well.
Our work contributed not only to INNOAIR project but also to the European Green deal goals achievement through direct reduction of the CO2 emissions in the city of Sofia. Statistics confirm that the transport sector accounts to almost one fifth of the global CO2 emissions and we as citizens must act in a thoughtful and sustainable manner in our everyday life.
