A digital twin of urban traffic — traffic model and mobility prediction
ewosoft Smart City — a living, calibrated model of the entire road network that forecasts traffic and lets the city simulate the effects of decisions before they are made in reality.
The challenge
Large cities in Poland and Europe face pressure that cannot be solved by expanding infrastructure alone. The number of vehicles grows faster than road capacity, and the cost of congestion — lost time, burned fuel, emissions and reduced reliability of public transport — burdens both the city budget and households. At the same time, cities collect huge amounts of traffic data that usually sits unused in separate systems.
The subject of this case study — a large regional capital — faced a classic dilemma: keep investing in concrete and asphalt, or first learn to make better use of the existing network. The city transport authority decided to build a traffic digital twin — a living, calibrated model of the whole road network that not only describes the current state but forecasts driver behavior and lets the city simulate the effects of development decisions before they are made in reality.
The transport authority had data from many systems but could not turn it into anticipatory decisions. Planning relied on periodic, manual traffic counts and static signal plans, leading to problems typical of large cities:
- Data in silos — detectors, ANPR cameras, public-transport GPS and ticketing systems worked separately, with no shared network model. Combining them into one picture took weeks of manual analyst work.
- Reactive traffic management — congestion was cleared only after it formed; without a 15–60-minute forecast, control-center operators were always a step behind what was happening on the roads.
- Static signaling — plans set rigidly by time of day, insensitive to events (accidents, mass events, weather). A single incident could trigger a cascade of jams across a whole district.
- Investment decisions without simulation — new tram lines or bus lanes were designed with no way to test the effects before deployment. Wrong assumptions surfaced only after costly construction.
- No hard data for climate policy — difficulty reporting transport emissions and assessing the effect of measures (Vision Zero, CO₂ reduction). The city declared targets but lacked a tool to measure progress.
The scale of the problem was measurable: during peak hours, the average speed on key corridors dropped below 18 km/h, and public-transport delays disrupted timetables and discouraged residents from giving up their cars.
The solution
We deployed ewosoft Smart City — Urban Traffic Digital Twin: a calibrated digital twin of the city’s road network, fed with real-time data and enriched with an AI prediction layer. A single environment combined building the traffic model with its practical use — from congestion forecasting to simulating city-development scenarios. The solution rests on four EWOSOFT Smart City pillars.
Pillar 1 — Data integration and model building (Big Data)
All sources — detectors, ANPR, fleet GPS, FCD, ticketing data — were connected into a shared data layer feeding a hybrid meso/micro-simulation model. The model reproduces origin-destination (OD) matrices, traffic flows and driver behavior, then is calibrated to field measurements. The data layer runs in a streaming architecture, so the model “breathes” with the city — updating on a minute cycle, not once a quarter as with classic counts.
Pillar 2 — Real-time traffic prediction (AI/analytics)
Machine-learning models (time series + graph neural networks that learn relationships between adjacent network segments) forecast volumes, speeds and congestion risk over a 15–60-minute horizon. The system detects anomalies (accident, failure, unusual flow) and automatically recomputes the network impact, telling the operator where congestion is about to form. Forecasts factor in context: day of week, event calendar, weather and seasonality.
Pillar 3 — Adaptive control and scenario simulation (system integration)
The digital twin recommends adaptive signal plans and Transit Signal Priority (TSP). In “what-if” mode, planners test variants — a new bus lane, a street closure, a clean-transport zone, a change in intersection layout — and see the forecast effects for the whole network before anything is built. The same environment thus serves two time horizons: operational (minutes and hours) and planning (months and years).
Pillar 4 — Sustainable mobility and ESG reporting (sustainability)
The model estimates emissions (CO₂, NOₓ, PM) from flows and speeds, supports planning a shift toward public transport and provides hard data for climate policy and Vision Zero. Every investment scenario can be assessed not only for traffic flow but also for environmental footprint — which becomes an argument in EU-funding conversations.
Architecture at a glance
The solution was built in layers so that further modules can be developed independently:
- Data acquisition layer — connectors to detectors, ANPR cameras, fleet GPS, FCD and ticketing systems, with real-time normalization and quality control.
- Model layer — a hybrid meso/micro-simulation engine reproducing the network, OD matrices and road-user behavior.
- Prediction layer (AI) — models forecasting volumes, speeds and congestion risk and detecting anomalies.
- Decision layer — signal and TSP recommendations, a “what-if” simulator and an ESG-reporting module.
- Presentation layer — dashboards for the control center and for planners, with maps, forecasts and scenario comparisons.
A scenario in practice — a bus lane instead of a general lane
The best illustration of the model’s value was a dispute over one of the main inbound corridors to the center. Road engineers proposed widening the roadway with an extra general lane; the public-transport team wanted a bus lane. Instead of settling the debate on intuition, both variants were run through the digital twin.
The simulation showed that an extra general lane would improve flow only temporarily — within a dozen or so months, newly generated traffic would fill the capacity (induced-demand effect), and the benefit for most travelers would be marginal. The bus-lane-plus-TSP variant, by contrast, shortened bus travel time enough to shift some travelers from cars to public transport, relieving the whole corridor. The decision was made on numbers, not beliefs — and in a few days, not months.
Implementation
The project was delivered in four stages over 11 months:
- 1Data inventory and integration (months 1–3) — connecting sources, unifying formats, building a shared data layer and quality control.
- 2Model building and calibration (months 3–6) — reproducing the network and OD matrices, calibrating to field measurements per FHWA / UK DoT guidelines.
- 3Prediction layer and control (months 6–9) — training AI models, integrating with signaling and piloting adaptive control on selected corridors.
- 4Scenario simulation and production rollout (months 9–11) — launching “what-if” mode, control-center and planner dashboards, and ESG reporting.
How we worked
Before the model was approved for prediction and simulation, it underwent calibration aligned with recognized guidelines (FHWA / UK DoT / WSDoT). Validation is a critical step — without it, no forecast or simulation would have decision-grade credibility.
- GEH < 5 for link flows — target ≥ 85% of links, result 91% of links.
- Predictive flow agreement (±15%) — target ≥ 85% of links, result 88% of links.
- Volume-forecast MAPE (30-min horizon) — target ≤ 15%, result ~10.5%.
- Travel-time agreement vs. field measurement — within ±15% (target met).
Results
The stage delivered a working digital twin — a single source of truth about traffic used both operationally and for planning. Measurable impact 12 months after go-live:
−21% intersection delay
−75% incident response time
+22% public-transport punctuality
- Average travel time (controlled corridors)
- −13%
- Intersection delay
- −21%
- Number of vehicle stops
- −30%
- Queue length at approaches
- −35%
- Congestion-forecast accuracy (30 min)
- ~89%
- Incident response time
- from ~12 min to ~3 min (−75%)
- Public-transport travel time (TSP)
- −18%
- Public-transport punctuality
- +22%
- Transport CO₂ emissions (pilot area)
- −16%
- Public-transport mode share (modal shift)
- +9%
- “What-if” analysis preparation time
- from weeks to days (−80%)
- Resident satisfaction with traffic flow
- +17% (survey)
- From reacting to anticipating — the city moved from reactive to anticipatory management; the digital twin is a single source of truth about traffic, used simultaneously by the control center and planners.
- Lower investment risk — any change in traffic organization or transport investment can be tested in the model first, reducing the risk of costly mistakes.
- Hard data for policy and funding — shorter travel times, more punctual transit and lower emissions translate into quality of life, network operating costs and the city’s credibility in securing sustainable-mobility funding.
- Better public debate — instead of arguing over hunches, the discussion centers on comparable scenarios and their forecast effects.
What's next
The digital twin is a platform that grows with the city. Each further module reuses the same, once-calibrated base — making further development cheaper and faster than building separate systems. Natural next steps:
- Extending the model with micromobility (bikes, scooters) and parking.
- Integration with crisis/event-management systems.
- Forecasting demand for electric-vehicle charging.
- Opening selected data to residents and app developers under an open-data model.
This is a reference (“model”) case study — the subject is an anonymized large city (~650,000 residents, ~1,100 km of roads). The percentage indicators fall within typical ranges reported in the literature and in smart-mobility / traffic-digital-twin deployments: adaptive signaling (ASCT) usually improves travel time by >10% (Los Angeles: −12.7%, stops −31%, delay −21.4%), and 35–43% on selected corridors; prediction / digital twin — ~15% travel-time drop on test corridors (Singapore); model calibration (FHWA / UK DoT / WSDoT) — GEH < 5 for ≥ 85% of links, MAPE ~10.5%; emissions and modal shift — transport CO₂ reduction typically 15–25% (Beijing: −25% at peak). Reference sources: U.S. DOT ITS Deployment Evaluation, FHWA Adaptive Signal Control Technologies, Miovision, MDPI (Applied Sciences, Sensors), EU Urban Mobility Observatory (Aachen), SJSUTST, TRENDS Research.
About EWOSOFT
EWOSOFT Systemy Informatyczne has been on the market since 2000. In Smart City we combine Big Data, simulation models and artificial intelligence into tools that support real city decisions. We work in stable teams, hand over the full code and rights, and build platforms that grow with the city.
Facing a similar challenge?
Let's talk about a dynamic-pricing module, ERP ↔ e-commerce integrations, or a system that grows with your company.