Cities, infrastructure and operations

Traffic flow

Cars as individual drivers, as cells on a road that move by rules, or as a fluid flowing through the network.

Question it answers: Why do jams arise without an evident cause, and will this junction relieve one?

Leaves out: Individual driver quirks average out; one erratic driver's effect is lost.

Runs on: about 10¹³ FLOPS rough a city traffic authority's desktop or small clusterSince: 1955 Lighthill and Whitham's fluid theory of traffic flow

Run the interactive model

Transport and land use

Where people live, work and shop, and how they travel between them, over years to decades.

Question it answers: If we build this rail line, how will the city change around it?

Leaves out: Decades-long social change is hard to forecast from historical patterns alone.

Runs on: about 10¹⁴ FLOPS rough a metropolitan planning agency's server clusterSince: 1964 Ira Lowry's model of land use and transport in a metropolis

Urban growth

Cities spreading over land, cell by cell.

Question it answers: Where will this city grow over the next thirty years?

Leaves out: Cells grow by fixed rules; real politics and land ownership are harder to encode.

Runs on: about 10¹³ FLOPS rough a planning department's desktop, minutes per scenarioSince: 1979 Waldo Tobler's cellular models of geographic change

Evacuation and emergency response

People leaving buildings, stadiums or regions, and responders moving the other way.

Question it answers: How long does it take to clear the building, and where do people get stuck?

Leaves out: Panic and crowd-crush behaviour are simplified; real evacuations are more complex.

Runs on: about 10¹³ FLOPS rough a desktop or small cluster, minutes per scenarioSince: 1995 Dirk Helbing's social force model of pedestrian crowds

Public transit network simulation

Buses and trains running to a timetable, with passengers choosing routes and transferring between them.

Question it answers: If we change this timetable or route, how much time do passengers save?

Leaves out: Assumes passengers know and follow the best route, whereas many take the most familiar one.

Runs on: about 10¹³ FLOPS rough a transit agency's server, minutes per scenarioSince: 1970 by 1970, early computer models of bus and rail timetables

Ride-hailing and dispatch simulation

A fleet of vehicles matched to passenger requests as they arrive, with drivers repositioning between trips.

Question it answers: How many vehicles do we need, and how long will passengers wait?

Leaves out: Assumes drivers follow the dispatch system's suggestions, which real drivers do not always do.

Runs on: about 10¹³ FLOPS rough a company's cloud server, minutes per city scenarioSince: 2000 by 2000, early simulation studies of dynamic vehicle dispatch

Aircraft boarding simulation

Passengers walking and stowing bags in a narrow aisle, simulated under different boarding orders to find the quickest way to fill a plane.

Question it answers: Which boarding order fills an aircraft fastest?

Leaves out: Real passengers ignore orders and travel in groups, which breaks idealised schemes.

Runs on: about 10⁶ FLOPS rough thousands of boarding runs, each seconds.Since: 2002 Van Landeghem and Beuselinck's simulation-based study of reducing passenger boarding time in airplanes. source

Airport airside simulation

Aircraft taxiing, queuing at runways and parked at gates, with controller rules and weather, to measure how much traffic an airport can handle.

Question it answers: How many flights per hour can this airport take before delays explode?

Leaves out: Controller and pilot decisions are reduced to rules, and delays at other airports spread in ways the model leaves out.

Runs on: about 10⁸ FLOPS rough minutes to hours on a workstation.Since: 1981 A flexible simulation model of airport airside operations, Transportation Research Part A. source

Elevator traffic simulation

Passengers arriving at random floors, with morning, noon and evening peaks, and cars stopping and dispatched by a control rule, to size lifts for a building.

Question it answers: How many lifts and what dispatch rule keep waiting times short at the morning peak?

Leaves out: Passenger arrivals follow measured or assumed patterns, and a change in how tenants use the building breaks them.

Runs on: about 10⁶ FLOPS rough seconds to minutes on a laptop.Since: 1993 Siikonen's elevator traffic simulation, in the journal Simulation. source

Shared autonomous vehicle fleets

A fleet of driverless taxis serving requests in a city, with each trip, wait, empty drive and parked time followed, to measure the vehicles needed, cost and congestion.

Question it answers: How many shared self-driving cars could replace private cars in a city, and what would it do to traffic?

Leaves out: Whether people would accept sharing rides, and what they would do with the time saved, is assumed and not known.

Runs on: about 10¹¹ FLOPS rough hours to days on a workstation or cluster for a metropolitan region.Since: 2014 Fagnant and Kockelman's agent-based scenarios of the travel and environmental implications of shared autonomous vehicles. source

Airspace fast-time simulation

Flights moving through sectors and waypoints in a simulated sky, much faster than real time, to test routes, sector design and rules for controllers.

Question it answers: How much extra traffic can a new sector layout or route structure carry safely?

Leaves out: Controller workload and judgement are represented by simple rules, so a busy sector's real limit may be lower.

Runs on: about 10⁹ FLOPS rough minutes to hours on a workstation per day of traffic.Since: 1964 Price's queueing theory in the fast-time simulation of air traffic control, in Navigation. source

Supply chains

Goods moving from suppliers through factories, warehouses and shops, with delays and shortages that echo back up the chain.

Question it answers: Why does a small change in demand produce a large swing in production at the factory?

Leaves out: Assumes rational reordering; panic buying and hoarding break the model.

Runs on: about 10¹³ FLOPS rough a company server, minutes per scenarioSince: 1958 Jay Forrester's industrial dynamics models of supply-chain swings

Manufacturing lines

Machines, workers, buffers and breakdowns on a production line.

Question it answers: Where is the bottleneck, and is the new machine worth buying?

Leaves out: Rare breakdowns and human factors are approximated from averages.

Runs on: about 10¹² FLOPS rough a factory planning desktop, minutes per runSince: 1961 Geoffrey Gordon's early discrete-event simulation language

Queues and service systems

Customers or passengers arriving, waiting and being served: airports, call centres, ports.

Question it answers: How many desks do we need to keep the wait under ten minutes?

Leaves out: Assumes people line up and behave predictably; real queue-jumping and abandonment vary.

Runs on: about 10¹² FLOPS rough a desktop, minutes per scenarioSince: 1961 by 1961, general discrete-event simulation tools begin modelling queues

Hospital patient flow

Patients moving through emergency departments, wards, theatres and discharge.

Question it answers: Where do patients wait, and what would free up beds?

Leaves out: Rare surges, like mass-casualty events, sit outside its normal patterns.

Runs on: about 10¹² FLOPS rough a hospital's desktop, minutes per scenarioSince: 1965 by 1965, early discrete-event simulations of hospital clinics

Warehouse and distribution centre simulation

Robots, conveyors, pickers and storage racks moving goods through a warehouse.

Question it answers: Can this warehouse pick and ship orders fast enough at peak demand?

Leaves out: Assumes equipment works as specified; real jams and breakdowns are harder to fully predict.

Runs on: about 10¹³ FLOPS rough a logistics company's server, minutes per scenarioSince: 1988 by 1988, early discrete-event simulations of automated warehouses

Last-mile delivery routing

Delivery vehicles and couriers routed through city streets to drop off many parcels in the least time.

Question it answers: What is the best order and route to deliver everything today?

Leaves out: Real traffic, weather and missed deliveries change the plan more than the model assumes.

Runs on: about 10¹³ FLOPS rough a delivery company's server, minutes per day's routesSince: 1964 early computer solutions to the vehicle routing problem

Project schedule risk simulation

Task durations drawn at random across a project network to give the spread of completion dates, instead of one critical-path estimate.

Question it answers: How likely is a project to finish by its deadline?

Leaves out: Task durations are assumed independent, though real delays are correlated.

Runs on: about 10⁶ FLOPS rough tens of thousands of schedule draws, cheap.Since: 1963 Van Slyke's Monte Carlo methods and the PERT problem. source

Warehouse robot fleet coordination

Hundreds of mobile robots carry shelves to pickers, and their routes, traffic jams and charging are simulated to size a fleet before it is built.

Question it answers: How many robots and what routing rules does a warehouse need to ship its peak orders?

Leaves out: Real robots break, humans work unevenly and orders arrive in bursts that are hard to predict.

Runs on: about 10¹⁰ FLOPS rough hours on a workstation per layout and day of orders.Since: 2008 D'Andrea's account of coordinating hundreds of autonomous vehicles in distribution facilities. source

Supply chain digital twins

A live model of a company's suppliers, factories, stock and shipping, kept in step with its data, used to test what a disruption would do.

Question it answers: If a port or supplier fails tomorrow, how long until shelves run empty and what is the cheapest fix?

Leaves out: Data from suppliers beyond the first tier is usually missing, so hidden dependencies stay hidden.

Runs on: about 10¹¹ FLOPS rough hours to days on a cloud cluster for many disruption scenarios.Since: 2020 Ivanov and Dolgui's digital supply chain twin for managing disruption risks and resilience. source

Container terminal simulation

Ships, cranes, trucks and yard stacks in a port, with random arrivals and service times, to find where containers queue and how many cranes are needed.

Question it answers: Where does this terminal jam, and how much would one more crane or gate help?

Leaves out: Ship arrival times slip with weather and upstream ports in ways that are hard to model, and yard stacking decisions are simplified.

Runs on: about 10⁸ FLOPS rough minutes to hours on a workstation.Since: 1975 Borovits and Ein-Dor's computer simulation of a seaport container terminal. source

Construction process simulation

Crews and machines such as trucks, excavators and cranes working through repeating cycles on a site, with random durations, to find bottlenecks and cost.

Question it answers: How many trucks per excavator will finish this earthwork fastest?

Leaves out: Weather, breakdowns and human variation are hard to put in, so the schedule is usually too optimistic.

Runs on: about 10⁶ FLOPS rough seconds to minutes on a laptop.Since: 1977 Halpin's CYCLONE method for modelling job-site processes. source

Power grids

Generation, lines and demand balanced second by second, including how failures cascade.

Question it answers: Can the grid take more wind and solar, and could one failure black out a region?

Leaves out: Rare cascading failures across many linked grids are hard to fully anticipate.

Runs on: about 10¹⁴ FLOPS rough a grid operator's control-centre server, seconds per scenarioSince: 1956 by 1956, early digital load-flow programs replace physical network analysers

Water and drainage networks

Water flowing through pipes, pumps, sewers and drains.

Question it answers: Will the drains cope with a heavier storm?

Leaves out: Ageing pipes and leaks are hard to know exactly, so results depend on assumed condition.

Runs on: about 10¹² FLOPS rough a water utility's desktop, minutes per scenarioSince: 1968 by 1968, early digital hydraulic models of water distribution networks

Building energy and daylight

Heat, air and light through a building across a whole year of weather.

Question it answers: How much energy will this design use, and is it comfortable inside?

Leaves out: Real occupant behaviour rarely matches the assumed schedules.

Runs on: about 10¹³ FLOPS rough a design office desktop, minutes to hours per buildingSince: 1977 early US building energy programs following the 1973 oil crisis

Infrastructure interdependence

Power, water, communications and transport linked, so that a failure in one spreads to the others.

Question it answers: What fails next when the power goes out?

Leaves out: Needs detailed data on every linked system, which is rarely all available at once.

Runs on: about 10¹⁴ FLOPS rough a research cluster, minutes per scenarioSince: 2001 Rinaldi, Peerenboom and Kelly's framework for interdependent infrastructures

Waste collection and treatment simulation

Waste and recyclables collected by trucks on routes, then sorted and processed at treatment plants.

Question it answers: Can we collect all the waste on time, and where should the next treatment plant go?

Leaves out: Assumes households sort waste correctly, which they do not always do.

Runs on: about 10¹² FLOPS rough a city council's desktop, minutes per scenarioSince: 1974 by 1974, early computer models of municipal waste collection routes

District heating and cooling networks

Hot or chilled water piped from a central plant to heat and cool many buildings at once.

Question it answers: Can the network deliver enough heat on the coldest day of the year?

Leaves out: Assumes pipes and pumps perform as designed; real networks leak and silt up over decades.

Runs on: about 10¹² FLOPS rough a utility engineer's desktop, minutes per scenarioSince: 1979 by 1979, early hydraulic and thermal models of district heating networks

Power-system electromagnetic transients

Fast voltage and current surges in a grid, from lightning, switching and faults, followed microsecond by microsecond to protect equipment.

Question it answers: What happens to a grid in the first milliseconds of a fault or lightning strike?

Leaves out: Lines and equipment are modelled with frequency-dependent approximations, and component data are often incomplete.

Runs on: about 10¹⁰ FLOPS rough time-stepped network solves on sub-microsecond steps.Since: 1969 Dommel's digital computer solution of electromagnetic transients in single- and multiphase networks. source

Energy-system capacity-planning models

The choice of power plants, fuels and networks over decades found by minimising cost under demand and emission limits.

Question it answers: What mix of energy technologies meets demand cheapest under a carbon limit?

Leaves out: Perfect foresight and fixed demand assumptions give tidy but unrealistic transitions.

Runs on: about 10¹⁰ FLOPS rough large linear programs over decades and regions.Since: 1981 Fishbone and Abilock's linear-programming model for energy systems analysis. source

Agent-based electricity markets

Power generators as adaptive agents bidding into an auction, to find how pricing rules, strategies and market power shape prices.

Question it answers: Do generators learn to raise prices in an electricity auction?

Leaves out: Agent learning rules are assumed, and human traders are not so simple.

Runs on: about 10⁹ FLOPS rough many agent-learning runs.Since: 2001 Bower and Bunn's experimental analysis of uniform-price versus discriminatory auctions in the England and Wales electricity market. source

Cascading-blackout models

Power lines tripping one after another as load shifts onto the rest, simulated to find how small faults grow into regional outages.

Question it answers: How does a small failure in a grid grow into a blackout?

Leaves out: Operator response and protection settings are simplified, and rare cascades need huge sampling.

Runs on: about 10⁹ FLOPS rough many Monte Carlo cascades on a grid model.Since: 2002 Carreras and colleagues' critical points and transitions in a power transmission model for cascading failure blackouts. source

Smart-grid co-simulation

The power network, houses, batteries, solar panels and their controllers are each simulated separately and run together, to test how the whole system behaves under different strategies.

Question it answers: How does a control scheme for thousands of home batteries affect the local grid?

Leaves out: Household behaviour is represented by synthetic load profiles, which miss rare patterns of use that stress the grid.

Runs on: about 10¹⁰ FLOPS rough hours on a workstation for a day of a district.Since: 2011 Schütte and colleagues' mosaik framework for modular simulation of active components in smart grids. source

Gas pipeline network transients

Pressure and flow waves in a network of pipes, compressors and valves as demand changes, so operators can schedule supply and keep gas moving without leaks or shortfalls.

Question it answers: How will the pressure in a gas network respond when a large user switches off or the gas mix changes?

Leaves out: Demand is uncertain and the pipes are treated as one-dimensional, so local effects at joints are simplified.

Runs on: about 10⁹ FLOPS rough minutes to hours on a workstation.Since: 1970 Distefano's Pipetran program for simulating gas pipeline network dynamics. source

Mobile network planning

Radio signals from phone masts spreading across a city or region, mapped against how many calls and data connections each mast must carry.

Question it answers: Where should the next mast go, and will callers get a signal there?

Leaves out: Real buildings and weather scatter radio signals more unpredictably than the terrain model assumes.

Runs on: about 10¹³ FLOPS rough a carrier's planning server, minutes per areaSince: 1978 by 1978, early cellular network coverage and capacity planning tools

Internet backbone and cable resilience

Data routes across the internet's backbone links and undersea cables, tested against breaks and outages.

Question it answers: If this cable or exchange point fails, whose connection goes down?

Leaves out: Real routing decisions during an outage depend on business deals the model does not see.

Runs on: about 10¹² FLOPS rough a research or carrier server, minutes per scenarioSince: 1995 by 1995, early studies of internet routing failures and resilience

Data centre facility simulation

Servers, cooling systems and power feeds inside a data centre, balanced against the computing jobs they run.

Question it answers: Will this data centre stay cool and powered as we add more servers?

Leaves out: Real equipment ages and fails in ways average models smooth over.

Runs on: about 10¹³ FLOPS rough an operator's own server, minutes per scenarioSince: 1999 by 1999, early computational fluid dynamics models of data centre cooling

Satellite broadband constellation simulation

Hundreds or thousands of satellites moving in orbit, handing signals between each other and down to ground users.

Question it answers: Can this constellation give everyone on the ground a fast, steady connection?

Leaves out: Real weather, ground terrain and interference cut into the coverage the geometry alone predicts.

Runs on: about 10¹³ FLOPS rough an operator's planning cluster, hours per scenarioSince: 1997 by 1997, early orbital and coverage simulations for satellite phone networks

Telecom traffic engineering

Calls and data connections arriving at random, sized against the lines and equipment available to carry them.

Question it answers: How many circuits or channels do we need before callers start getting busy signals?

Leaves out: Assumes arrivals are random and independent, which breaks down during a big shared event.

Runs on: about 10¹¹ FLOPS rough a carrier's desktop, seconds per scenarioSince: 1917 Agner Krarup Erlang's traffic theory for telephone exchanges

Smart-city sensor network simulation

Thousands of small sensors across a city reporting traffic, air quality, water or lighting data over a wireless network.

Question it answers: Will enough sensor readings get through, and how long will the batteries last?

Leaves out: Real sensors drift, fail without notice and drop off the network more often than the model assumes.

Runs on: about 10¹² FLOPS rough a city IT department's server, minutes per scenarioSince: 2010 by 2010, early simulation studies of city-scale wireless sensor networks

Internet worm propagation models

Simulated hosts on a network are infected by self-spreading code, using measured scanning strategies and bandwidth limits.

Question it answers: How fast could a worm infect the vulnerable internet?

Leaves out: Defences and patching speed are assumed.

Runs on: about 10¹⁰ FLOPS rough Cheap epidemic-style runsSince: 2002 How to 0wn the Internet in your spare time (2002) source

Vehicular network simulation

Cars exchanging messages while a road-traffic simulator moves them, to test collision warning and platooning over radio.

Question it answers: Will cars exchange warnings reliably while driving?

Leaves out: Radio shadowing by other vehicles and buildings is only approximated.

Runs on: about 10¹⁰ FLOPS rough coupled traffic and packet simulators.Since: 2011 Sommer, German and Dressler's bidirectionally coupled network and road-traffic simulation. source