About the model
The model used by the simulation aims to be useful by being flexible and fast at the expense of being completely comprehensive and accurate. It is designed to realistically model journey sources and destinations in Canberra, to be conveniently runnable (in a web browser) for high usage scenario on common hardware, taking just a minute or two elapsed, and to present the results in a way to facilitate their interpretation, comparison and sanity-checking.
The two most significant limiting characteristics of the model are that time ticks over in "discrete" one minute intervals (rather than simulating a continuous stream of time), and that journey source and destinations are suburbs (albeit randomised locations within a suburb) rather than specific street addresses.
Before running each simulation, the model tries to estimate the number of arrivals and departures using the current parameters for each suburb for each minute as follows. It first generates an estimate for the number of trips which will depart in any one minute, based on the total number of journeys and their per-hour distribution as specified to the model. For each minute, this total number of departing journeys is randomly "dithered" to be within a -20% : +20% range, with a simple linear probability gradient up from 0 at the lower (-20%) bound, rising to the centre of the range and then back down to 0 at the upper (+20%) bound. This produces does not produce a normal distribution, but instead imagine 2 right-angled triangles back to back. This total number of departing journeys is then distributed randomly to each suburb based on the probability of each suburb being a journey source for the minute of the day under consideration [more...]. The destination suburb is then randomly chosen based on the probability of each suburb being a journey destination for that minute, with the added complication that nearby suburbs are slightly more probable. (Further, the increase in probability depends on where the suburb is; for example, suburbs in the Weston area of Canberra are over 5 times more likely to originate journeys to other suburbs in Weston than a suburb in Gungahlin is likely to originate journeys to other suburbs in Gungahlin. The so-called "self-containment" information used in this model was derived from data presented on page 12 of the Gungahlin Drive Extension Review prepared by Scott Wilson Nairn Pty Ltd for the National Capital Authority in 2002.)
The above estimation is run twice (representing 2 "days"), and the average values of arrivals and departures per minute for each suburb are passed to the "real" simulation for it to use as its best estimate of demand for and supply of cars.
The "real" simulation then runs, independently generating a dithered total journey demand rate for each minute and then independently assigning source and destination suburbs for each journey.
Actual demand for cars from each suburb and arrival of cars at each suburb will vary considerably from the estimate.
Major limitations of the model
Discrete rather than continuous time
All journey requests are for immediate travel
No specific addresses
Inter-suburb travel times are estimated from Google Maps
Simple sharing
Poor or no optimisation
- Once charging begins, a car is always charged to 80%. Often, a car charged to, say, 65% should be reassigned to service waiting passengers rather than continue charging to 80%.
- "Top-up" charging when there is a great surplus of idle cars only happens during a 1 hour window before the AM peak. It should be adaptive to loads.
- Forward load anticipation is naiive, being based solely on the anticipated difference between supply and demand of cars at each suburb in 15 minutes. Decisions to transfer cars between suburbs are mostly based on this simple measure.
- Car transfer is naiive. For example, imagine A is 10km north of B which is 10km north of C and that a 10km journey takes 15 minutes. On anticipated loads A has 10 spare cars, B has 1 spare car, C is short of 1 car. The system will currently transfer a car from A to C, which will take 15 minutes "too long". A better strategy is to simultaneously send a car from B to C and from A to B: same total distance, but C's need is satisfied 15 minutes sooner.
Distribution of chargers and concurrent use at any location
Charging time of day
It is possible that an operational system will have constraints on power consumption, for example, during the morning and evening household electricity usage peaks. Alternatively, there may be strong incentives to use power during solar-PV peak outputs.
Such constraints could influence choices such as car battery capacity, charging rates, number of chargers and even number of cars.
Single vehicle type
This model treats each vehicle as interchangeable and does not support vehicles of mixed specification (particularly range and seating capacity). It is not clear whether having a mix of vehicles with different specifications would be beneficial.
Weekend (non-weekday) outcomes are estimated, not simulated
Non-weekday travel patterns are very different, and are not simulated. Rather, the financial outcome for these days is estimated using a heuristic described below.
Financial incentives associated with car sharing
Perhaps unrelated passengers sharing a car due to a system scheduling decision should receive a discount. Perhaps guaranteed unshared peak period use of a car regardless of route should be available for an extra fee.
Major "known unknowns"
Commercial availability and cost of autonomous cars
Whilst there are demonstration systems, convincing arguments and expert opinion predicting their availability in a time-frame ranging from 2017 to 2025, the date of the commercial availability of a fully autonomous car is the biggest unknown and risk.
Specifications and cost of suitable electric vehicles
The range of electric vehicles are improving and their costs are reducing. Whilst this model suggests that the viability of a transport system based on autonomous EVs is not particularly sensitive to range, it is extremely sensitive to capital costs and per-km running costs.
Battery degradation
Depending on battery capacity, fleet size and service levels, the simulation suggests that over a 3 year period, each battery will be cycled somewhere between 1500 times (for a large Tesla S battery) or 6000 times (for a small Smart ED battery). Although cycles will be limited to dropping from no more than 80% capacity to normally, no less than 20% capacity, these are larger numbers of cycles which may indicate severe battery capacity degradation with most current battery technology. More below..
Location of charging stations
Depending on the fleet size, car battery capacity and charging rate, between 200 and 2200 charging stations may be required, distributed over at least 8 locations to minimise travel times/distances required for recharging. It probably makes sense for some or all of these locations to also contain maintenance (including workshops and spares stored), cleaning and some admin facilities. Should they reuse infrastructure freed up such as government car parks and bus depots?
Access to high-voltage power supply is required.
Autonomous charging
Depending on the fleet size and car battery capacity, there may be many tens of thousands or even over a hundred thousand recharges required each day. Requiring human intervention to connect and disconnect a charging cable would be expensive and inefficient. Current generation "wireless" car recharging, and new systems proposed by BMW and Daimler deliver only 3.6 - 7.0 kw/Hr, which is too slow to be economic for a large autonomous fleet where time spent charging is not only monopolising an expensive charging station but is also idle time.
So, a "wired" charging system capable of delivering the same rates as the Tesla super-charging stations is required, but made simple, cheap, safe and reliable, perhaps along the lines of the "third rail" used in by some electric trains, or an overhead conductor used by trams and electric trains.
At an event to launch the Tesla S "D" on 9 October 2014, Tesla CEO, Elon Musk reportedly said that he'd like to add a system that automatically plugs the EV into a charger, as part the Tesla "Auto Pilot" technology that automatically garages your car when you get home and has the car ready at your front door when you want to leave home. [Update August 2015: a prototype.]
Actual travel patterns in Canberra, now and in 2020
The model uses vaguely plausible heuristics to generate a vaguely convincing set of journeys, but these are not backed by current actual journey data, let alone journeys using a cheap, door-to-door, 24x7 autonomous car system.
Administrative and operational costs and system requirements
The default model includes a fixed annual $1M for admin plus separate fixed annual provisions per car of $100 for general admin and $200 for communications.
It also includes a separate default provision for maintenance of 2.5 cents per-car-per-km, which equates to between $3000 and $4500 per car per year, depending on model parameters.
Are these amounts appropriate?
An operational system requires everything from an easy-to-use booking and billing system to a lost property office. It requires maintaining relationships with suppliers, customers and the government, skilled mechanics and other staff, management of spare parts, ...
Congestion
This model does not attempt to measure congestion. Travel-times are based on estimates from Google, conservatively increased by 20% for off-peak and 30% for peak periods as described here. However, it would be interesting to monitor route loadings to determine peak vehicles per lane kilometer and intersection crossings.
It is not obvious what levels of autonomous traffic can be tolerated. Simulations using the University of Texas at Austin's Autonomous Intersection Management project suggest "intelligent" intersections can coordinate vehicle approaches and traversals to handle three times the throughput of an intersection controlled by traffic signals. (See also the DriveWAVE intelligent intersection work from the MIT Senseable City Lab.) An analysis of coordination and "platooning" of autonomous vehicles suggests they could increase lane capacities by a factor of three.
This is not to say that congestion analysis is unwarranted, particularly when considering a mixture of autonomous and non-autonomous vehicle flows, but traditional measures of road and intersection capacity and congestion may need modification.
The simulation does provide some results suggesting that congestion in peak periods will be lower the higher the uptake, despite "dead running" needed to transfer cars in periods of "tidal" flows. For example, consider the results from the very high 1.1m journey simulation's "Occupancy data" section. It shows high average occupancy of cars arriving in AM peak at commuter "sinks" - work centres such as Civic, Parkes, Barton, Russell, and the ANU. It also shows the occupancy of cars leaving Gungahlin, Belconnen and Tuggeranong in the AM peak, regardless of the destination.
Consider typical output for Gungahlin:
- 15,625 people leaving Gungahlin in 9,891 cars, average occupancy of 1.58 people per car
- A further 1,116 cars transferring to other regions of Canberra (empty).
That is, a total of 11,007 cars travelled the roads leaving Gungahlin in the AM peak, carrying 15,625 people.
Consider how these 15,625 people would have travelled without a fleet of autonomous cars. Given the current ACTION bus share of 8%, approximately 8%, or 1,250 would have travelled by bus. Assuming 40 people per bus, that's about 31 buses. Roughly estimating "car equivalents" to produce a car-equivalent congestion number, lets assume each bus produces the same road congestion as 3 cars, so these 1,250 bus travellers produce about 100 car equivalents of congestion on roads leaving Gungahlin over the AM peak.
The remaining 14,375 people travel by car. If the "effective" average occupancy is 1.2 travellers per car (which might be lower than the people per car, because some people in the car may be "chauffeurs", such as a parent driving a child to a school, or a carer taking a patient to the doctor, and hence not really people wanting to travel somewhere), these 14,375 people would require sharing of 11,979 cars.
Hence, without a shared fleet of autonomous cars, approximately 12,079 "car equivalents" (100 + 11,979) would be travelling the roads leaving Gungahlin, compared to 11,007 with a shared fleet. That is, a shared fleet reduces the number of car equivalents (cars and buses) on the road in the direction of peak flows by almost 9% for the same volume of travellers.
With "tidal" flows, a shared fleet of autonomous cars increases the volume of cars going "against" the flow, as empty cars need to reposition from the commuter "sink" to the commuter "source" to pick-up more passengers. But, by definition, these repositioning cars are travelling on otherwise under-utilised roads (few people want to be going in that direction, which is why the repositioning cars are empty).
Configurable model parameters
Cars
Number of cars on the road
Purchase cost
The default cost of $40,000
This cost is based on a calculation that starts with the current price in Australia of the Nissan Leaf of $40,000 "drive away", the availability of many models in the US and Europe at a lower retail cost, and significantly lower prices for cars such as the Smart ED (2 seater).
It then assumes that the price of EVs will fall as production increases, competition increases (particularly from Chinese manufacturers at least 5 of whom already have production models), and the cost of batteries (which form a significant component) continues to fall.
Purchasing a fleet of thousands or tens of thousands direct from a manufacturer rather than from a dealer's retail inventory could be expected to give considerable savings over a walk-up, one-off purchase from a dealer.
The Australian retail price of the Nissan Leaf dropped from $51,500 before on-road costs in 2012 to the current (2014) walk-up, one-off purchase on-road price of $40,000.
Assuming a 15% decline in EV prices over 6 years to 2020 due to mass production, increased competition, improvements in technology and cost reductions and a further 10% discount for a very large fleet purchase, we estimate a reasonable purchase and commissioning price of $40,000 * .85 * .9 = $30,600.
It is hard to predict the cost of an autonomous system because full Level-4 systems are not yet commercially available. The early versions of a key sensing component, the LIDAR (light detection and ranging) systems used by Google cost over $US70,000, with a total cost of $US150,000 in extra equipment.
The IHS report Emerging Technologies: Autonomous Cars—Not If, But When (Jan 2014) estimates:
The price premium for the SDC [Self-Driving Cars] electronics technology will add between $7,000 and $10,000 to a car’s sticker price in 2025, a figure that will drop to around $5,000 in 2030 and about $3,000 in 2035 when no driver controls are available.
The 2012 KPMG white paper Self-driving cars: The next revolution speculates (page 20):
.. perhaps bringing the total cost down to $1,000 to $1,500 per vehicle, as economies of scale are achieved.
Morgan Stanley's research report on Tesla Motors estimates that the cost to the consumer of adding "complete automation capability" to a vehicle will be between $5,000 and $7,000 and expect this to be commercially available between 2019 and 2024.
The upgrades to the Tesla S EV announced on 9 October 2014 include sensors to support their first steps to an autonomous system: long range radar, camera and image recognition (for detecting road signs), 360 degree ultrasonic radar, GPS integrated with real-time traffic updates. These sensors are included in the base price for the standard model, but much of their functionality (initially designed to support some level of autonomous driving on highways and automated parking and retrieval of the car) will only be activated as part of the purchase of a "tech package", a $4250 option.
The Economist's Technology Quarterly, Q3 2012 article on autonomous cars quotes Erik Coelingh, a senior engineer for "driver support" systems at Volvo, as estimating autonomous control systems will add $3,000 in costs.
A spokesperson for Ibeo, a German supplier of LIDARs claimed in 2012 that the cost per unit would be $250 by 2014.
We have assumed a conservative cost of the autonomous system of almost $10,000, giving our default simulation autonomous EV car price of $40,000 for a vehicle with 4 seats.
Representative currently available electric vehicles
| Manufacturer, Model | Battery kWH | Battery range km | Retail Price (before incentives) |
|---|---|---|---|
| Nissan Leaf | 24 | 121 - 200 | $A40,000 |
| Mitsubishi i-MiEV | 16 | 100 - 160 | $A49,000 |
| BMW i3 | 18.8 | 130 - 160 | $A64,000 |
| Mercedes-Benz Smart ED (2 seater) | 17.6 | 90 - 121 | $US25,000 or $US20,650 plus $US80/month for "battery assurance plus" lease |
| Fiat 500e | 24 | 130 - 160 | $US35,200 or $US8,163 for a 3 yr lease |
| Renault Zoe | 22 | 100 - 150 | equivalent of $A26,000 plus battery lease |
| Ford Focus Electric | 23 | 122 | $US35,200 |
| Chevrolet Spark EV | 21.3 | 132 | $US35,000 |
| Bolloré Bluecar | 30 | 150 - 250 | €19,000 + €80/month for batteries (equivalent to a 3 year cost of $A31,600) |
| Kia Soul | 27 | 130 - 190 | $US33,700 |
| Tesla S | 60 - 85 | 335 - 500 | $A97,000 - $112,000 |
| Tesla 3 [announced only - 2017 availability] | 50? | 320? | $US35,000 |
| GM Bolt [announced only - 2017 availability] | ? | 320 | $US30,000 |
| BYD e6 | 61 | claim 300, 196 using EPA | $US52,000 |
Looking at the Smart ED in more detail
The Smart ED (2 seater) is a particularly interesting fleet option. The "battery assurance plus" is a 10 year lease which guarantees a battery replacement when battery capacity drops to 80% of specification, regardless of km travelled. It is also transferable on the sale of the vehicle. (Nissan has a similar program which costs $100/month and is also unlimited km).
Assuming a fleet purchase discount of 10% and a 5-year technology and competition price reduction of 15% gives an $A price of $18,200 per car. An up-front 3 years for the battery lease up-front (also with 10% and 15% reductions anticipated) adds $A2,600, giving an upfront EV car price of almost $21,000. Assuming an autonomous driving option on a smaller car costs adds almost 25%, or $5000 (rather than the pessimistic estimate of $10,000 for the 4-seater), the total car cost is $26,000. The residual value of the car is now significant, as after 3 years it comes with a guaranteed 7 years of 80% battery range, regardless of km travelled. Assume the residual value is $3000.
Mercedes-Benz claim a city range of 121km for the Smart ED, so using 80% of this figure as our maximum charge-range, we set 100km as our maximum range.
If you run a simulation with these figures: $26,000 cost, residual $3,000, range 100km, but boost fleet size by, say, 10% - 33% to compensate for it having only 2 seats and shorter range hence requiring more non-productive charging transfers, the simulation demonstrates both much greater profitability (allowing for much greater fare subsidies for promoting community mobility regardless of income) and greatly reduced waiting times.
For example, the "very low usage (current ACTION load)" scenario with a fleet of 2,500 vehicles carrying a maximum of 2 passengers costing $26,000, residual $3,000, financed at 10% with a theoretical max range of 100km and 250 chargers typically runs at an annual loss of $4M, with typically over 94% of journeys having a wait time of <= 1 minute, and very few waits of over 5 minutes. (By way of comparison, the default settings based on a $40,000 4-seater car with a theoretical range of 240km and fewer chargers typically runs at a $12m annual loss: that $8m gap could be used to pay for over 7,500 off-peak journeys every day).
Another example, "very high usage (750K trips/day)" scenario with a fleet of 32,000 vehicles carrying a maximum of 2 passengers costing $26,000, residual $3,000, financed at 10% with a theoretical max range of 100km and 2200 chargers typically runs at an annual surplus of around $100M, with typically almost 99% of journeys having a wait time of <= 1 minute, and hardly any waits of over 3 minutes. Such a surplus could be used to pay for over 90,000 off-peak journeys every day.
A fleet comprised entirely of 2 seaters may not be optimal (for example, for transporting a family of 3 or 4), but modelling a mixed fleet is beyond the capabilities of the current model.
Residual Value
Depending on simulation parameters, cars can be expected to have travelled between 340,000 and 480,000 km over 3 years, and undergone between 1500 and 6000 full battery cycles (albeit confined to the 20%-80% charging range). It is unlikely that the car will have much value except as scrap, and the scrap value is conservatively estimated to be worth no more than the disposal costs.
There have been suggestions that old EV batteries could have a valuable second life as static batteries in home PV systems. Maybe...
Useful Life
The default is 36 months.
A longer life reduces capital costs but should be accompanied by an increase in per-km maintenance costs and perhaps size of the spares.
According to the New York City Taxicab Fact Book, the average NY cab is 3.3 years old (so its life would be considerably longer, but must not exceed 6 years by law) and drives 70,000 miles (112,700 km) each year.
Financing interest rate
The default is 10%.
Governments can normally raise finance considerably cheaper. A 10% return may be attractive for many superannuation fund investments, particularly as most of the funds are secured against assets (cars) which can be resold and this model assumes the outstanding amount is paid back evenly over the life of the loan (ie, the amount "at risk" decreases linearly).
Maintenance costs
The default cost is 2.5 cents per km, and was estimated as follows:
- General mechanical maintenance: 0.5 cents/km. The Electric Power Research Institute's report Total Cost of Ownership for Current Plug-in Electric Vehicles (June 2013) estimated cumulative maintenance costs of a 2013 Nissan Leaf as $US1,183 after 150,000 miles. This figure was based on scheduled maintenance from the owner's manual and excluded tyres. Converting to cents per km gives an approximate cost of 0.5 cents/km.
- Tyres: 0.8 cents/km. Assuming $400 to replace a set of tyres with a lifetime of 50,000 km.
- Cleaning: 0.7 cents/km. Assuming employing cleaners including overheads costs $30 per productive hour, and a daily clean of a car takes one person 5 minutes, then each car clean costs $2.50 in labour. Assuming consumables of 30 cents brings cost to $2.80. Assuming a car averages 400km/day, cleaning cost per km is 0.7 cents.
- Unforeseen and not covered by warranty and/or spare parts from fleet "spares": 0.5 cents/km.
Notably absent from these maintenance costs are explicit allowances for:
- Accident repairs (such as panel-beating). Tongue-in-cheek: the liability for these will rest with the human driver crashing in to the autonomous vehicle, or the manufacturer of the autonomous system that failed.
- Wear and tear (such as door-handles and electric windows breaking, cars catching fire). The vehicles will be less than 3 years old, and a manufacturer's warranty is assumed to be negotiable as part of the large fleet purchase, plus the extra (by default) 5% of the fleet size purchased as "spares". At a 5% ratio of spares, 1 in 60 cars can be assumed to be "written off" each year.
- Battery replacement. Hitherto, the biggest uncertainty regarding maintenance costs has been battery replacement due
to degradation resulting from both calendar age and cycles.
Mercedes-Benz and Nissan have both introduced battery lease options which guarantee
battery performance to 80% (Mercedes-Benz) and 70% (Nissan). In addition, battery technology is undergoing rapid improvements aimed
at increasing cycle life and power density and reducing costs. Further, the charging regime modelled involves never charging
batteries to over 80% of their theoretical capacity, and never undertaking a new journey when they fall to below 25% of their
theoretical capacity, practices which are thought/hoped/rumoured to improve battery performance. Hence, we feel justified in
assuming that battery performance to at least 80% does not require additional maintenance costs.
In March 2015, the largest electric bus manufacturer, BYD, announced a 12-year unconditional battery warranty on their lithium-iron-phosphate battery-packs, claiming an operational life of over 7,000 charging cycles.
Annual insurance/rego/admin
The default is $2000, estimated as follows:
Insurance: $1200 for Compulsory Third Party Insurance and Third-party property damage.
It is hard to estimate insurance costs, as one of the primary advantages posited for autonomous cars is the reliability of the "driver" who is never tired, distracted by passengers or their phone, fiddling with their radio, drunk, incapacitated nor overcome by road-rage.
What about insurance for passengers? Who is liable if autonomous car drives into a tree? (The autonomous system manufacturer?) Who is liable if the boot lid falls on a passenger's head as they are removing their shopping? (The car manufacturer for faulty design/manufacturer or the fleet operator for negligence in maintenance, or for not removing legal liability for such accidents?)
The "spares purchased" part of the fleet provides some self-insurance for repairs and write-offs.
A government-run fleet would probably "self insure" - what are the equivalent costs for ACTION buses?.
Registration: $500
Does this depend on ownership? Do ACTION buses pay registration?
Communications: $200
Cars may send a 1K of TCP data to report their status every 10 sec, and another 1K for events such as pick-ups and set-downs. Bulk data (video? software updates?) can be transferred using wifi during charging. So, mobile data requirements may be 24*60*6 1K status packets plus ~50 1K event packets = max 10MB/day. Assuming a commercial plan charging 5c/MB (eg, ALDI mobile on the Telstra network), that's max $200/year.
Administration:$100
Cover irregular, unspecified but inevitable events in the life of the car (not related to acquisition or disposal, which are included in other categories).
Theoretical Maximum Range
The default is 240km.
This default is very high for 2014 model cars, currently exceeded only by Tesla models. However, as EV battery power densities are increasing and costs are decreasing, it is widely predicted that batteries with maximum range of well over 240km will be cheap and common by 2020.
South Korean battery manufacturer, LG Chem, announced in July 2014 that they will be producing electric vehicle battery packs with a range of 200 miles (320km) in 2016.
At the Detroit Car Show in January 2015, GM displayed their all-electric BOLT concept vehicle with a range of 320km and price of $US30,000 and availability targetted for 2017 (according to "A source with knowledge of the company's plans").
When specifying a maximum range, consider using the most conservative estimate of typical city driving with at least some heating and cooling, and then using no more than 90% of that figure to represent maximum range after battery capacity degradation (see the discussion above on battery replacement).
Passenger capacity
The default is 4.
The Nissan Leaf transports 5 adults, but it is assumed paying passengers would rather wait a minute or two for an empty car than be stuck in the middle of the "back" seat. The Smart ED transports 2 adults.
The current model tries to share cars between passengers starting from the same suburb and travelling to the same suburb or a destination that requires only a small extra travelling time for any passengers. For the "low" and "medium" usage models, car sharing is rarely possible because the journey volumes are low and the likelihood of passengers being able to share cars is low. For the "high" usage models, car sharing is only likely to be used in peak hours. Most of the time, a single passenger is effectively booking a "car", not a seat.
In off-peak periods, cars, not seats are booked. That is, a family of 4 (or 5) going to the movies, or 2 friends going home together after dinner, or a single person going to start their late shift all pay the same rates (a total of $2.87 for a average 13.4km trip using default flag fall and per km rate). During peak periods, some trips will be shared, but perhaps a premium could be paid by passengers who did not want to share, or wanted to book a car not a seat (for example, when travelling with their children). However, many peak-period trips are not candidates for sharing. For example, the fleet management system is keen to return an excess of cars accumulating at town centres back to suburbs during the morning peak period: at such times, trips from town centres to suburbs can be booked without sharing. Similarly, most suburb-to-suburb trips will not be shared.
What about large families, with say, more than 3 children, or a large group of people travelling together? As is required now with current sedans and taxis, more than 1 car will be needed. At least most of the time (and all of the time "off peak"), only one fare needs to be paid per car per booking, so a family of, say 7 people booking 2 cars would almost always only pay 2 fares. For some families, being separated into 2 cars will be unacceptable and they will keep their Tarago. The NSW Bureau of Transport Statistics 2011/12 Household Travel Survey measured just 1% of weekday trips and 3% of weekend trips where the vehicle occupancy was 5 or more (Table 4.8.4). Unfortunately, the vital "6 or more" figure is not available from this report.
Spares purchased
The default is 5%.
It is assumed that the on-road fleet is sized to just meet peak capacity. However, cars will be off the road for maintenance, or will be written-off in some circumstances. Hence a pool of "spares" is required as replacements, but note, not as a contingency for a load spike, as this would require them to be also insured and registered, which is not costed in this model.
These spares could also be considered as a source of spare-parts: panels, windows, motors, etc, and hence prepay the costs of these and guarantee their availability. A 5% spare capacity represents a capacity to provide 1 complete spare car per year for every 60 cars on the road.
Charging
Number of chargers
Charge power
The default is 75kW.
Purchase cost
The default is $15000.
This default is estimated from the 2012 price of Nissan's 50kW chargers ("below €10,000"), and ABB's similar price for small-volume orders for its Terra Smart Connect fast charger. Tesla are deploying hundreds of high-power chargers around USA, Europe and now Asia. Nissan and Renault are also active in rolling out public charging stations. The basic technology is understood, but this model assumes the availability of autonomous high-power charging (see above), which currently does not exist.
Useful Life
The default is 120 (10 years).
Annual Costs
The default is $3000.
Cost of electricity per kWHr
The default is $0.20 per kWHr.
Trends in ACT Electricity Consumption by, Sinclair Knight Merz Pty Ltd (Jacobs), May 2014, predicts that the retail price of a high voltage supply will be $0.15 per kWHr through to at least 2018/18.
Operating surplus is quite sensitive to the price of energy.
Journeys
Journeys per day
Per hour distribution
The default is taken from Estimating urban traffic and congestion cost trends for Australian cities, Bureau of Transport and Regional Economics Working Paper No 71 Figure 2.23 page 84, Network (arterial) average.
Interesting alternatives include: VicRoads Traffic Monitor 2012-13, September 2014 page 15, "Monitored Network Traffic Profile" (although the right-hand scale seems a bit strange with uneven increments), and NSW Bureau of Transport Statistics 2011/12 Household Travel Survey Fig 3.9.1 (note: travellers, not vehicles).
The model distributes traffic volumes with a "tidal" bias over the day:
- 04:00 - 9:59: from suburbs to centres
- 10:00 - 14:59: no bias
- 15:00 - 18:59: from centres to suburbs
- 19:00 - 20:59: from suburbs to centres
- 21:00 - 03:59: from centres to suburbs
The propensity for a suburb to be a journey source and destination depends on the suburb definition in the TownAndSuburbs.js file (parameters which for each suburb define population and factors which define the factors: peakSource, peakDest, dayOffPeakSource, dayOffPeakDest, nightOffPeakSource, nightOffPeakDest.
Fares
Peak flag fall
The default is $0.45
The logic behind having a flag fall is that the system incurs a real and fixed cost to accept a booking, schedule a car and send the car on a probably empty, and in any case, unbillable journey to the pick-up location, and wait whilst the passenger boards.
The fleet must be sized to meet peaks. Outside of peaks, much of the capacity of the fleet is wasted, so peak period travel "costs" the system more, which is passed on to peak travellers.
Peak per km rate
The default is $0.25
Off-peak flag fall
The default is $0.20
Off-peak per km rate
The default is $0.20
Other
Annual Fixed System cost
The default is $1,000,000
This cost is distinct from per-car and per-charger costs which are intended to fully cover their commissioning, operation and maintenance.
Rather, it is for costs which are largely fixed, regardless of fleet size, such as:
- liaison with customers, suppliers and government
- creation and maintenance of systems to facilitate car booking, billing and optimal operation of the fleet
Hard-coded model parameters
Peak periods
Used for peak/off-peak tariff calculations.
Car-sharing periods
During car-sharing periods, try to allocate passengers to an already "in-use" car, that is, a car already occupied by one or more passengers with at least one empty seat, either already travelling to the same destination as the new passenger, or "nearby", as long as the deviation to a new destination does not increase travel time by more than 20% or by more than 5 minutes for either the existing or prospective passengers.
Outside of car-sharing periods, an already "in-use" car will only be considered if there is no idle car nearby (that is, "in the same suburb") as the requesting passenger: if an idle car is available (as it almost always is outside of the car-sharing period), it will be used.
As a result, for most purposes, outside of car-sharing periods, a passenger almost always effectively books exclusive use of the entire car, whereas frequently during car-sharing periods, and particularly when travelling with the commuter "tide", a single seat, rather than the entire car, will be booked.Km travelled per kWHr used
The battery capacity is not part of this model. Rather, each km of range added assumes that 1/6th of a kwHr is added to the battery (and furthermore, more grid power is required to achieve this charging - see charging efficiency below.
6km per kwHr seems to be commonly achieved in current EVs in city driving conditions with some use of heating or cooling as would be expected in Canberra. It is likely that an autonomous system, particularly when linked with a wide-area traffic management system, would be able to more effectively optimise energy usage than a human driver, so the setting for this parameter seems suitably conservative.
Maximum range used
It is assumed that the battery is only charged to a level which will give it a range of 80% of the theoretical maximum range.
This figure has been chosen to both honor the collective wisdom which asserts damage to current lithium ion batteries is minimised by not exceeding 80% charge, and to represent in the model battery degradation with age, particularly due to cycle frequency.
Range threshold to trigger recharging
When the remaining range falls to 25% of the the theoretical maximum range (not the maximum range used) or 30km, the car will be routed to a recharger. It is possible that for small range simulations, some cars will embark on a trip longer than the remaining range. Yes, that is embarrassing, but it is something a real system would have to avoid and occasionally cope with by dispatching cars to pick-up passengers.
This figure has been chosen to both honor the collective wisdom that damage to current lithium ion batteries is minimised by not taking them under 20% charge, and the collective wisdom that cars shouldn't roll to a halt without power a long way from home.
Range threshold to trigger opportunistic recharging
If a car has been idle for at least 1 minute and has less than this range remaining, it will be routed to a recharger.
A high value of this setting may increase overheads by causing dead-running of the car to a charger. However, frequent "top ups" may be good for long battery life, and by charging in apparently idle times, may reduce wait times in peak periods.
Range threshold to trigger early morning recharging
If a car has been idle for at least 1 minute and has less than this range remaining, and the time is between 5am and 6am (ie, before the morning peak), it will be routed to a recharger.
The idea behind this setting is to attempt to top-up as many cars as possible before the morning peak, which you should see a charging burst starting at 5am. Depending on grid capacity and sources of power for EVs, this may or may not be appropriate.
Charging energy efficiency
The charging system is not 100% efficient. That is, when charging a battery, you will consume more energy from the grid than you can subsequently retrieve from the battery.
Figures of 85%-90% are commonly quoted, but the whole charging and system infrastructure will have power requirements of its own, so a figure of 80% should conservatively include these.
Max diversion overhead factor
When a car is already allocated and we're thinking of sharing it to a new destination, what is an acceptable increase in travel time for the allocated and possible new passenger caused by this "diversion"? This parameter specifies that an increase of up to 20% is ok.
Max diversion overhead minutes
When a car is already allocated and we're thinking of sharing it to a new destination, what is an acceptable increase in travel time for the allocated and possible new passenger caused by this "diversion"? This parameter specifies that an increase of up to 5 minutes is ok.
Off-peak travel time increase factor
Inter-suburb travel times are obtained from Google Maps. These times are increased by 20% for off-peak travel and rounded up to the next minute.
Peak travel time increase factor
Inter-suburb travel times are obtained from Google Maps. These times are increased by 30% for peak travel and rounded up to the next minute.
Pick-up and set-down times per passenger
When a car arrives at a pick-up or set-down point, it is assumed that each passenger will take 1 minute to embark or disembark. As well, if multiple passengers are being picked up from the same suburb, it is assumed that travelling from one pick-up point to the next pick up point will take an average of 1 minute. Ditto for set-downs.
It may be useful to model variable times. For example, a disabled passenger may take more than 1 minute to embark; 4 school kids arriving at their high school should take less than 7 minutes to disembark (4 x 1 minute set-down + 3 x 1 minute move-to-next-set-down-location).
Estimating non-weekday variable costs and revenues from weekday data
The number of non-weekday trips is estimated to be 70% of weekday trips. This is considerably below estimates of 85%-90% (see for example, the NSW Bureau of Transport Statistics 2011/12 Household Travel Survey).
However, for the purposes of this simulation, we are interested in the number of cars required and the fares that will be paid. The above-mentioned survey also indicates that work trips have a considerable lower average occupancy per car (1.10 people) compared to non-work trips (1.66), and non-work trips will be more likely on non-weekdays. Travel on non-weekdays will very rarely be shared by separate fare-paying passengers: almost all travel will be by fare payers paying for the entire car, rather than a seat, and so a family of four will travel for the same cost as one person. Hence, for revenue purposes, although the number of people travelling and the number of trips they take may be 85%-90% on non-workdays of the workday number, we conservatively estimate that the number of fares collected will be just 70% (and all will be at off-peak rates).
Unfortunately, this also implies that more than 70% of week-day car numbers will be required: basically, no car sharing is assumed, so "per km" non-weekday fleet costs (energy and maintenance) are inflated from the 70% level inversely proportional to the level of single occupancy trips made during a weekday.
For example, if 90% of weekday trips were single occupancy, then non-weekday "per km" costs are estimated as weekday "per km" costs times 0.7 and then divided by 0.9.
These calculations are probably slightly pessimistic (conservative) because the fleet is greatly underutilised during the non-weekdays and travel is less "peaky", so forced transfers between suburbs to meet demand and transfers to charging stations will be more than proportionally lower. On the other hand, there is some evidence that non-weekday trips may be shorter on average than weekday trips, as trips to work tend to be the longest journeys (at least as reported in the NSW Bureau of Transport Statistics 2011/12 Household Travel Survey).
Note, that you can mostly simulate a non-weekday outcome yourself, by specifying a per-hour traffic distribution and by making the peak and off-peak fares the same, setting vehicle capacity to 1 person (to eliminate any possibility of sharing and hence gathering multiple fares for a trip), and supplying the number of journeys (not passengers) you think are needed. You cannot, however, change the spatial distribution of journey (the source and destinations, which may be characteristically different enough on weekends to matter) without amending the contents of the TownsAndSuburbs.js file.
Sensitivity of operating surplus and waiting times to various parameter settings
The following graphs show how the operating surplus is affected by changing various financial assumptions, and how the operating surplus and percentaged of journeys with a waiting time of less than 1 minute are affected by changing the battery range and number of journeys assumptions.
The "base" scenario is the "Very High" uptake, with 23,000 cars, 1,500 charging stations and 750,000 journeys per day, and only one parameter at a time is varied.
| Charger costs are small relative to other system costs, and defrayed across many cars. | |
| For a given fleet, wait times (most particularly at peak periods) are very sensitive to the number of journeys. | |
As range decreases, recharging requirements are higher. Hence unless the number of chargers and/or cars is increased, waiting times increase dramatically as range is reduced. For an example of modelling an EV with a small range, see this simulation of a fleet based on the 2 seater Smart EV. |
Canberra Autonomous Car Simulation by Kent Fitch is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.