What is behind our platform?

We make complex scientific paradigms accessible for non-technical users. We provide faster and credible outputs thanks to our next-generation programming language and micro-service architecture.

Technical & Scientific Background

Time, cost and accuracy of traditional urban research and simulation-based approaches are problematic for decision makers

City modeling costs a lot and takes months or even years to develop which contradicts with the rapidly urban changing landscape. GeoTwin’s human-centred urban analytics platform dramatically reduces the complex activity & agent-based modeling and simulation process.

GeoTwin’s platform called 'Human-Centred Urban Intelligence Platform' is powered by GeoTwin's proprietary population synthesizer & agent-based simulator and advanced visualization enablers.

Our cloud-based research platform is an industry first and dramatically reduces time and cost for analysts, researchers and decision makers – instead of contracting with a specialist to do a million-dollar plus study that takes months or years, users can run their own simulations, quickly change parameters or hypotheses, and get results in a matter of hours.

Why Simulation ?

Many of today’s problems are too complicated for simple structural models and cannot be handled with just machine learning or statistical techniques.

With our software, you can explore any number of different future scenarios to identify strategies and tactics that work under a wide range of possible outcomes.

Capture emergent phenomena - large - scale behaviour that impossible to predict using traditional methods as they result from the local interactions between agents as well as the transportation supply.

We start by creating a digital twin

We create a completely synthetic population for a defined area, using anonymous and aggregated data. This is a virtual population that is statistically representative of the real one.

We apply activity-based travel modeling to generate a sequence of activities and connecting trips for every person in the synthetic population on a specific day, and use computer simulation to model their interactions with transportation supply and each other as they attempt to realise their activity schedules.

This helps us confidently replicate trip patterns across a city or region. We compare our models to real data to ensure accuracy and reliability.

What isSynthetic Population Data ?

In essence, GeoTwin creates a synthetic population by using government-supplied census data, along with other sources for socio-demographic and location information, on residents living in a given area, including regional housing and employment availability. We apply modelling and optimization algorithms to generate this synthetic population, one that statistically mimics (“twins”) the census population via a person’s behaviors, individual household composition, and in aggregate. These synthetic 'people' and 'households' are then assigned housing units, workplace locations, and schools. GeoTwin then creates personas, which are profiles that extract behavioral patterns from de-identified mobile location data.

The core of our Digital Twins involves a synthetic assimilation, a digital replication of real-life observations that map, match, and emulate human behavioral patterns in space and time. Persona creation springs from a region’s spatial and socio-demographic data, plus other records of regional human activity. It is composed of three main underlying behavioral choice models: activity scheduling, destination location, and travel mode. Each synthetic person, with its assigned persona and travel behavior models, is “motivated to travel.” Finally, we simulate a given day of an average week, to represent travel patterns for the entire city area. We generate a travel itinerary for each synthetic person on a given day. This serves as the input for GeoTwin City Insight and Scenario Simulator. GeoTwin uses multiple sanity checks to preclude any inconsistency problems from the start and validate the travel metrics with some ground-truth data.

Why Activity and Agent-based Modelling ?

Agent-based modelling captures the behavior and interaction of different agents within an environment, recreating complex dynamics. This means it can solve problems where traditional methods are insufficient, such as what-if analysis in transportation, market research and retail analytics.

Our system uses multi-agent simulation and AI to better understand and predict a city’s future mobility patterns. It does this by simulating people’s behavior based on their individual preferences.

The optimization suite is a set of mathematical algorithms and data structure to help different processes such as routing, or dispatching. It also helps during the synthetic population pipelines to increase the accuracy of the activity plans and population characteristics. These algorithms are formally provable by nature.

The optimization suite can also be used on top of the simulation, for instance if you want to optimize a fleet size or station location. With the help of a mathematical model as a master (formal but simplified), it can compute fewer simulations to search for the optimum. The point for the simpler optimization model is to have feedback about the simulated reality and take into consideration the differences between expected and realized. This is sometimes called black-box optimization. This avoids you to do the job yourself by an exhaustive set of simulations, or make a big heuristic iterative analysis.

Geotwin's agent-based modelling characteristicsGeotwin's agent-based modelling characteristics

The optimization suite is a set of mathematical algorithms and data structure to help different processes such as routing, or dispatching. It also helps during the synthetic population pipelines to increase the accuracy of the activity plans and population characteristics. These algorithms are formally provable by nature.

The optimization suite can also be used on top of the simulation, for instance if you want to optimize a fleet size or station location. With the help of a mathematical model as a master (formal but simplified), it can compute fewer simulations to search for the optimum. The point for the simpler optimization model is to have feedback about the simulated reality and take into consideration the differences between expected and realized. This is sometimes called black-box optimization. This avoids you to do the job yourself by an exhaustive set of simulations, or make a big heuristic iterative analysis.