Multimodal Reisemønsteranalyse

Background Creating good and dynamic mobility systems requires good statistics on where people go and their travel habits. There are no general, fully automatic and cost-effective solutions for this today. Traditionally, public transport companies have received information about boardings from the ticketing system and the driver. As mobile ticketing and open systems have become more widespread, registrations from ticketing systems have also become increasingly unreliable. Many companies have in recent years installed Automatic Passenger Counting (APC) systems in their vehicles. These detect fairly accurate numbers of boardings and alightings at each stop. Travel patterns can be represented by a so-called origin-destination (OD) matrix, which specifies where passengers travel from and to from each and every stop. But none of the above-mentioned solutions provide this detailed information. Knowing about transitions and the use of different mobility modes can also not be detected with current solutions. E.g, APC does not provide data on whether the passenger uses other mobility services such as city ??bikes combined with public transport. Thus, they provide no end-to-end information for multimodal journeys. The objective of the MultiRA project is to calculate the elements in the OD matrix for journeys which might include transitions and different modes of transport. The solution is based on collection and analysis of sensor data from electronic devices, as well as data from other sources that may be available, such as APC data and real-time data. Statistical and machine learning algorithms will be used to calculate and predict travel patterns and passenger flow with high accuracy for the entire multimodal transport system. Achieved results The developed solution is based in part on the collection and anonymisation of WiFi packages from mobile devices. In recent years, there has been a trend towards increased randomization of addresses for devices not connected to a WiFi access point, for privacy reasons. Thus, it turns out that it is much more challenging now to use WiFi packets sent from mobile devices as the only source to estimate OD matrices. The project nevertheless concludes that such data alone can give good results for other use cases than just calculating the OD matrix, such as load statistics and load changes over time. The project's approach to using WiFi packages together with other data such as APC data proved to give good results in calculating travel patterns, and the project has therefore focused much on this. The project has done a good job of streamlining the data collection process. A new, robust WiFi data collector which delivers data in a compressed format has been developed, which also reduces mobile data usage from the vehicles. In total, approx. 5.4 billion WiFi events have been collected. Processes and algorithms have been created to analyze data using sources like APC, GNSS and WiFi, in addition to the static route network data. With these algorithms, one can generate a very precise OD matrix. A lot of work has gone into generalizing these algorithms and making them more robust against erroneous input data. Processed data is stored in a database and made available on an API. A GUI, using the API, has been created for the presentation of finished analyzed data. Travel patterns based on each location are displayed here. By choosing a stop, you get the number of boarding passengers and alighting passengers, and the total number on board, for each stop. The interaction with the stop in the map shows where the alighting passengers came from, where they travel to, or both. An initial algorithm for estimating two-hop trips has been developed: First, single-hop trips are detected. Then two-hop trips are estimated based on the distance and time interval between the alighting and boarding stops of the detected single-hop trip, as well as the direction of the two single-hop trips. Based on the estimated journeys, end-to-end OD matrices for the entire route network, including transitions, can be estimated. The project has provided an opportunity to follow how communication devices such as mobile phones are improving privacy by making it more difficult to track devices. This has many positive consequences, but some less fortunate. Increased insight into how to use mobile devices to obtain useful information and statistics related to mobility, while at the same time fully ensuring privacy, within and ever-changing technology framework has been valuable. This competence can also be utilized in domains other than mobility. The technical solution is further developed in the Innovation Norway-supported project Digital Destinasjon, in which the technology is improved to include free flow travel patterns in an area (in the project it is Geilo) to measure the flow of people and tourists without being tied to fixed locations as in a transport network.

Prosjektet har vist at løsningen kan brukes til å generere en antatt svært korrekt O/D-matrise. Med tilgang til denne kan brukerne av løsningen oppnå de mål virkninger og effekter som er beskrevet i søknaden.

Prosjektets overordnete mål er å utvikle og pilotere et system som frembringer ny og forbedret statistikk over folks reisemønster. Systemet baserer seg på innsamling av Wi-Fi trafikkdata ved hjelp av Wi-Fi monitorer installert i busser, på ferjer og ved stativer for bysykler, i tillegg til informasjon fra andre kilder som måtte være tilgjengelig som APC data og sanntidsdata. Statistiske algoritmer og maskinlæringsalgoritmer vil bli benyttet for å beregne og prediktere reisemønster og passasjerflyt. En storskala pilot som skal implementeres i Stavanger vil være sentral i prosjektet, og prosjektet vil adressere flere forskningsutfordringer. 1) Innsamling, anonymisering og kvalitetssikring av store datamengder må gjøres på en effektiv måte og tilpasset integrering med data fra andre kilder og videre statistisk prosessering. 2) Statistiske algoritmer må utvikles for å beregne statistikk for reiser med overganger, der ulike typer kollektivtransport benyttes, og der kollektivtrafikk benyttes sammen med tilbud om leie av bysykler. 3) Maskinlæringsalgoritmer vil benyttes for å prediktere reisemønster basert på trening med innsamlete data. 4) Applikasjoner som benytter seg av statistikken skal utvikles. 5) Hvordan eierskap til den type data og statistikk som dette prosjektet generer kan inngå i ulike aktørers forretningsmodell vil utredes. Informasjonen som frembringes av prosjektet vil ha stor nytteverdi for å kunne tilpasse transportløsninger til tilgjengelig kapasitet og etterspørsel gjennom optimalisering av rutenett, forbedrete informasjonstjenester for passasjerer og utvikling av fremtidsrettete transporttjenester.