The European Commission decisions promote using clean transport methods, and one of which is cycling. In order to ensure the successful functioning of an integrated urban transport system cycling has to be one of its components, which naturally fits into the overall intermodal transport system. Due to limited funding it is important to understand which bike path network would be more efficient and usable for municipality and cyclists. The multi agent-based (M AS/ABM ) bicycle path network and exploitation simulator (VeloRouter) allows paths occupancy simulation, but quality of the results depends on the simulator right training.
The European Commission Decision C 6776 established the specific European
Transport Specific Programme implementing Horizon 2020, Part 11 "Smart, green and
integrated transport" [
Therefore, it is important to understand which bike path network would be more efficient and ensure the greatest possible population transfer to bicycles. Cyclists are interested in justified selection of the route to avoid the problems for travellers related with terrain, quality of route and occupancy during the travel time. So, intelligent route planning is important task.
The idea for VeloRouter originated in FP7-ICT-2011-7 FUPOL project No. 287119
(2011-2015) “Future Policy Modelling” [
In conformity with the requirements Skopje Bicycle Inter-Modality Simulator
(http://www.fupol.eu/en/news/skopje-bicycle-inter-modality-simulator) [
Based on the analysis above and achieved results, it can be concluded t hat cycling route design and planning products mainly offer the capability of publishing cycling routes, but do not provide functionality necessary for municipalities to design suitable bicycle path network, which is critical for sustainable urban transport scheme, where component of green transport is growing. 2 VeloRouter - Simulator for Intelligent and Justified Decision Making The multi agent-based bicycle path network and exploitation simulator (VeloRouter) (see Figure 1) is designed in the Repast Symph ony environment and uses OpenStreetMap spatial data.
The technology has dual applicability as it is adapted to both the needs of municipalities and cyclists. Each agent is a cyclist or a group of cyclists that move on a chosen route considering route occupancy, traffic restrictions and the quality of the route.
VeloRouter provides municipalities with bicycle path discussion on the web and geofencing opportunities by receiving feedback from cyclists.
VeloRouter user authentication and authorisation is ensured using social networks: Facebook, Linked-in, Twitter, ResearchGate etc., given that cyclists are avid users of social technologies, whereas municipalities have the option to use individual registration options.
VeloRouter has both opportunities for municipality un cyclists. The municipality is interested in some basic question: Is the offered cycling route map satisfactory? This is recognized by summarizing potential comments and statistics analysis. Behind basic the second question is: Which potential cycling route sections should be built first?
The cyclists have the opportunities not only to send a message to the municipality, but also to publish his routes designed for public discussion. The cyclists want to know what the occupancy of a route will be in certain meteorological conditions on a specific date, as well as if the route is suitable for the travellers group i.e. terrain etc.? MAS/ABM based occupancy simulator can give the answers on the questions mentioned and provides monitoring of bicycle path network development scenarios and changes management. 3
How to Train the Simulator for Justified Decision Making There are two ways of route planning in VeloRouter: with and without occupancy simulation, which can be switched off to reduce waiting time for data processing. Capability of Occupancy simulation determines simulator teaching or training to ensure confidence of the occupancy forecast.
More or less training or teaching of the simulator is substituted by tuning by other
researchers however not always it is the same. Anyway tuning and teaching of the
simulator is still open question therefore amount of real publications is few. The
research related with teaching of simulators are done by Kux [
Input data: • • • • • •
Number of cyclists; Type of day (working day, holiday, all days); Start time (hh:mm interval); Weather probability (summer: April – September, winter: October - March), o Summer day with precipitation (no, not likely, likely, yes), o Summer day without precipitation (no, not likely, likely, yes), o Winter day with precipitation (no, not likely, likely, yes), o Winter day without precipitation (no, not likely, likely, yes); Route (start and end points); Weather conditions (s pecific date and hour), o Temperature (C), o Snow (mm), o Rain (mm), o Atmospheric pressure (hPa), o Wind,
Speed (m/s),
Direction (meteorological degrees).
Output data: • Route occupancy per minute (data structure), o Route section (ID), o Day, o Minute, o Occupancy; • Route section occupancy in total over days (total) (data structure), o Route section (ID), o Day, o Occupancy.
Occupancy simulation algorithm is the following (see Figure 2). The basis is occupancy data stored in Travels and Intentions data Base ( ). Occupancy data contains the number of cyclists, which is considered the smallest entity. The entity represents a specific cyclist who uses the roads. For each such cyclist an agent has to be created, which, based on other attributes and input data, will participate in t raffic at a specific time and day. The data containing all occupancy entries is read from ( ).
When all occupancy data entries are processed, load data is saved to the ( ). To estimate the load of each route, initially the attributes of each cyclist are calculated. After it is determined whether he participates in traffic, the following conditions are checked: • Is it an appropriate season to be cycling? • Is it an appropriate day (working day / holiday) for the specific cyclist to be cycling? • Is the weather appropriate for the specific cyclist to be cycling? • Is road load adequate for the specific cyclist to be cycling?
The cycling route is calculated for each occupancy entry during each simulation session. If one occupancy entry consists of multiple cyclists, then all take on e route, but the time can vary.
The route is recalculated each time to be up-to-date based on the version of each route calculation algorithm, as for example road restrictions can change on a daily basis.
Weather data are retrieved from the WeatherForecast service, which sends forecast information for the day by the hour, therefore if cycling takes longer than an hour it is possible to simulate whether the trip has to be interrupted.
The MAS/ABM model verification does not rise problems, however validation results
of each use case depends on the capacity of ( ), therefore it can be done during later
stages of the project introduction.
3.2 Occupancy Simulator Training
Occupancy Simulation model is trained and automatically tuned by initial data from
Travels and Intentions data Base ( ). There context-steering training approach is
used [
Related with context capturing [
Lower credibility data ( ) are Cyclists ( ) intended route plans that depend on season and meteorological conditions. It is just intention set that can be realised, but may also not occur. Higher credibility data are the one collected by municipality staff ( ) , interviews in mass events and marker places. However highest credibility data are collected automated by filtering semantic webs ( ). Although it is possible if question is not politicized, that initiates activities of "opinion army" to propagate the "correct" opinion. Also in this case, of course, it is possible to detect artificially generated opinions and exclude them from tot al assessment similar to DDoS [13] attacks, however it would be too expensive for cycling path planning.
In conformity with Credibility Assessment Rate ( ), where = 〈 , , 〉 (1) and - credibility of semantic search data, - credibility of municipality data, - credibility of Cyclists data, initial data credibility all the time is assessed.
Travels and Intentions data Base ( ) stored data quality ( ) depends on volume of the data i.e. from total number of respondent opinions that are related with specific planning region. If planning region includes several thousand cyclists, but only several persons have shared their opinion, then modeling result credibility level cannot be the highest.
Collected data credibility is not constant, but is changing in time. It is influenced by both construction of new bike paths and socio -economical factors. In fact, usability is associated with data deterioration (∆ ).
When data is getting outdated, its significance in the procedure of Occupancy simulation decreases. Data significance is regularly recalculated = ( , ∆ ) . It means that Occupancy simulation forecast in each session can have different credibility as the influence of the set ( ) is different, where ( ) = ( , , , ) (2) and - cyclists data, - municipality data, - semantic search data, - significance of stored data.
Training performance of Occupancy Simulation model is proportional to ( ) context data quantity and credibility, but usability ( ) of Occupancy Simulation model (see Figure 2) can be specified as: ( ) = −1 + ∆ (3) where = → , and is limited by computing resources capacity, but ∆ - reflects training 1, time or interval between ( ) updates.
Simulator training quality is determined by the simulator validation results that are based on modeling data statistics comparison with output data of real functioning bicycle route planning and exploiting system. Comparison is done using Kolmogorov Smirnov testing [14]. It can be noted that this validation type is useful only if ( ) is comparable with number of potential cyclists in planning region . 4
Conclusions
VeloRouter is one of the first open source simulators that takes into account municipality tasks and is cyclist-friendly.
Agent-based simulation supports relatively precise load prediction so unpleasant incidents during a trip can be limited. However main MAS/ABM application bottleneck is necessity to know location of each agent on the map asking for significant calculation resources if agent amount or the size of planning region are growing. However using Cloud and High Performance Computing (HPC) solutions [15] it is possible to remove bottleneck without simulation algorithm changes.
VeloRouter enables continuous occupancy assessment on the planned routes , which are determined by the amount and quality of context captured data in PostgreSQL database. Such a way Occupancy Simulator is trained to increase the assessment credibility.
The main tasks for further development are the use of semantic analytics and visualization results in a manner that corresponds to the user’s perception, a s well as their virtualisation and VeloRouter integration into a municipalities information systems 13. Rajkumar, Nene, M . J.: A Survey on Latest DoS Attacks: Classification and Defence M echanisms. International Journal of Innovative Research in Computer and Communication Engineering, Vol.1, Issue 8, pp.1847-1860 (2013) 14. Corder, G. W., Foreman, D. I.: Nonparametric Statistics: A Step -by -Step Approach.
ISBN 978-1118840313. Wiley (2014) 15. Dongarra, J.I., Van der Steen, A.J: High Performance Computing Systems: Status and Outlook. Acta Numerica. doi:10.1017/S09624929. pp. 1-91. Cambridge University Press (2012)