Search research nuggets
Sizing Transformers
We used measurements of home loads to optimally size transformers.
Our first paper describes how to model home energy consumption using Markovian models for transformer sizing. We collected load measurements taken at six second granularities from 20 homes over four months to build these models, which can now be used by simulations and analyses for related problems [1].
Our second paper applies the Markovian model described above to determine the optimal transformer sizing for a neighbourhood. We demonstrated that the electric load distributed from a transformer can be modelled the same way as the traffic from data sources distributed from a router. Because of this equivalence, we were able to use teletraffic theory, usually used to size telecommunication access networks, to size power distribution networks. We showed that this method of sizing transformers gives results that match the sizes of transformers currently deployed in neighbourhoods. This method can be used to size transformers when building new neighbourhoods in the future [2].
[1] O. Ardakanian, S. Keshav, and C. Rosenberg. Markovian Models for Home Electricity Consumption, Proc. ACM SIGCOMM Green Networking Workshop, August 2011.
[2] O. Ardakanian, S. Keshav, and C. Rosenberg. On the Use of Teletraffic Theory in Power Distribution Systems, Proc. e-Energy, May 2012
On Using Storage and Genset for Mitigating Power Grid Failures
Using a two-way inverter to convert between AC and DC power, batteries can store power when the electricity from the grid is available and discharge to meet the load during a time of power outage. Advances in electric vehicle energy storage technology have led to a sustained decrease in the price of electrical storage. We studied the use of battery storage to allow a set of homes in a single residential neighbourhood to avoid power outages. In essence, the entire neighbourhood can be thought to be connected to a single large uninterruptible power supply.
Storage is still expensive, however, so our goal is to choose the smallest battery size such that, with high target probability, there is no loss of load despite a grid outage. Recognizing that the most common approach today for mitigating outages is to use diesel generators, we also study the related problem of minimizing the carbon footprint of diesel generator operation by minimizing generator operation.
Access the paper here
S. Singla. On Using Storage and Genset for Mitigating Power Grid Failures, MMath thesis, University of Waterloo, April 2013.
Quantifying the Benefits of Extending Electric Vehicle Charging Deadlines with Solar Generation
Imagine a parking lot that was powered by solar panels, where you could leave your electric vehicle to charge during the workday. If you specify that you will be driving home at say, 4 PM, the parking lot would ensure that your car is charged fully by that deadline.
If all cars in the parking lot have strict deadlines however, it might not be possible to charge all cars fully by the time their owners return. Instead, how much would it help if you added an hour’s leeway to your charging deadline? For example, “I will return home sometime between 4 PM and 5 PM” versus “I will return at 4 PM exactly”. If someone returns closer to 4 PM there is a bigger chance their car won’t be fully charged, whereas if they return closer to 5 PM there is a greater chance the car will be charged. If people are more flexible with their charging deadline, the parking lot is able to charge all the cars to a fuller capacity.
In this paper, we optimized when to charge each car, and how much to charge each car over a day to meet flexible and strict charging deadlines. The paper can be accessed here:
O. Ardakanian, C. Rosenberg and, S. Keshav, Quantifying the Benefits of Extending Electric Vehicle Charging Deadlines with Solar Generation, Proc. IEEE Smart Grid Communications, November 2014.
Optimal contracts for providing load-side frequency regulation service using fleets of electric vehicles
A fleet of vehicles can provide regulation services at 30-second intervals by controlling the rate of charging. We show how to choose the size of the regulation contract such that cars in a fleet will still meet their charging demands on time, while still making the most money by doing so. Access the paper here:
H. Zarkoob, S. Keshav, and C. Rosenberg. Optimal Contracts For Providing Load-Side Frequency Regulation Service Using Fleets of Electric Vehicles, J. Power Sources, vol. 241, pp. 94-111, November 2013.
How Similar is the Usage of Electric Cars and Electric Bicycles?
Problem
There is a lot of data-driven research being released on EVs. However, a lot of this research is based on assumptions of EV usage rather than actual data, because buying a fleet of EVs in order to collect its data is very expensive. Also, there isn’t an abundance of public data on EV usage available to researchers.
Solution
Our idea was to model EVs using eBike data. This would be a much cheaper solution than purchasing an entire fleet of vehicles. In this study, we analyzed 70GB of usage data from eBikes, as well as data logged from pure EVs and hybrids to determine whether eBike usage could model EV usage.
Evaluation
We found that EVs actually cannot be modelled using eBike data, because they are used differently. This includes the time of charging, the state of charge when charging begins, duration of charging, destinations, and trip durations. Moreover, the distribution of months of operation and trip lengths differ at a fundamental level.
Access the paper here
C5. Simon Dominik Fink and Lukasz Golab and Srinivasan Keshav and Hermann de Meer. (2017). How Similar is the Usage of Electric Cars and Electric Bicycles?. ACM eEnergy EVSys Workshop, (334–340)
The return on investment for taxi companies transitioning to electric vehicles: San Francisco
Problem
What would be the return on investment of transitioning a taxi company’s fleet to electric vehicles?
Solution
We built a Bayesian model to calculate the return on investment when switching to electric vehicles. The model can be configured with several input parameters, including the type of vehicle to be tested, electricity and gasoline prices, and roadside charging/battery switching infrastructure assumptions.
Evaluation
We evaluated our model using location data collected from over 500 Yellow Cab San Francisco (YCSF) taxis. Using prices at the time of the study, 2014, we found that transitioning YCSF’s fleet to battery electric vehicles and hybrid vehicles was indeed profitable. Moreover, given that gasoline prices in San Francisco were only 5.4% higher than the rest of the US, but electricity prices were 75% higher, taxi companies with similar mobility patterns in other cities were likely to profit more than YCSF. Generally, when gas prices rise above a certain level, EVs will give a better return on investment.
We’d like to note that at the time of this study, we made the assumption that there would be battery swapping stations set up around San Francisco to allow EVs to recharge immediately (see: Better Place). As part of our study, we determined that the optimal place to put these swapping stations would be at the airport, where longer taxi trips would likely drain the battery of an EV. Unfortunately, the infrastructure for these stations was never realized.
Access the paper here
J20. T Carpenter*, AR Curtis*, S Keshav. (2014). The return on investment for taxi companies transitioning to electric vehicles. Transportation. 41(4)
Sizing vehicle and bikeshare pools
Sizing Finite-Population Vehicle Pools
Range anxiety, or the fear that a battery will run out before reaching a destination, is a factor that prevents customers from buying electric cars. Solutions like battery switching, adding more charging outlets, and larger batteries have been proposed, but in 2014 BMW offered a simpler solution: what if when a dealership sells an electric car, they automatically “subscribe” the buyer to a program that allows them to borrow gasoline cars for longer trips? That is, whenever an EV car owner needs to make a longer trip that will outlast the car’s battery range, they can come to the dealership, trade in a subscription coupon for a gasoline car, and rent out that gasoline car free of charge that day.
The question then becomes: how many gasoline cars does a dealership need to reserve for this program? In this study, we analyzed four different methods of sizing this vehicle pool, one of which doesn’t need any prior data or training. The following paper describes these methods in detail.
J16. Tommy Carpenter and Srinivasan Keshav and Johnny Wong. (2014). Sizing Finite-Population Vehicle Pools. IEEE Trans. Intelligent Transportation Systems. PP(99)
Bikeshare Pool Sizing for Bike-And-Ride Multimodal Transit
In shared bike-and-ride systems, commuters can ride a shared bicycle from home to a public transportation station, drop the bicycle off in a pool, take public transportation, pick up another bicycle from a pool at their destination stop, then ride to their final destination. So, how many bicycles would be needed at each transportation stop? A naive solution would be to have two bicycles for each commuter, one at the stop they board public transportation, and one at the stop they get off. The caveat is that this would be prohibitively expensive. So, what would be the smallest number of bikes that should be available at each public transportation station for this to work?
It turns out that the mathematical problem for sizing these bikeshares has the same structure as the mathematical problem for sizing vehicle pools, which was described above. By using this method, we were able to reduce the number of bicycles in each pool by between 39% to 75%, compared to having two bikes per commuter.
J6. G. Tang and S. Keshav and L. Golab and K. Wui. (2018). Bikeshare Pool Sizing for Bike-And-Ride Multimodal Transit. IEEE Transactions on Intelligent Transportation Systems.
The relative importance of price and driving range on electric vehicle adoption: San Francisco and Los Angeles
Problem
The adoption of electric vehicles has increased over the past few years, partly due to different government policies such purchase rebates, charging station rebates, high-occupancy lane access for EVs, and more. However, it isn’t obvious which of these policies are the most successful in increasing EV adoption. Another thing to consider is that although rapid EV adoption is desirable, it has potential drawbacks including increased grid load and the need to build additional expensive charging stations.
Solution
We developed a Python-based tool that predicts EV adoption and usage, by simulating people’s behaviour when purchasing, driving, and charging EVs in a city. This tool can be used for purposes such as the following:
- Policy makers can estimate the impact of different policies on EV adoption
- Electrical utilities can estimate the spatial and temporal changes in energy demand on the electric grid, based on different levels of EV adoption and different EV technologies
- Charging station planners can estimate how different levels of EV adoption affect public charging station activity
- Battery manufacturers can determine how battery sizes would affect EV adoption and electrical load
If you are interested in using this tool, please contact Prof. Keshav at keshav@uwaterloo.ca.
San Francisco
We first used the tool to model EV adoption in San Francisco, tuning the parameters using data from a prior survey of people’s transportation needs. We did a ‘what-if’ analysis using the tool – for example, what are the effects of different policies and technology changes on EV adoption and usage in San Francisco? We found that:
- Increasing rebates by $2000 had the highest impact.
- Range anxiety is not an important deciding factor in EV adoption since most driving distances are short as in San Francisco.
- A consumer education program focused on the lower lifetime costs of EVs may not be worthwhile.
- Increasing the battery sizes of EVs results in increased charging load due to less gasoline use and less public charging station activity.
Adepetu, V. Arya, and S. Keshav. (2016). An Agent-Based Electric Vehicle Ecosystem Model: San Francisco Case Study. Transport Policy. 46: 109-122.
Los Angeles
We also used the tool to model EV adoption in Los Angeles, California to determine the impact of high-capacity batteries and EV rebates on EV adoption. Simulating the different cases of battery costs and prices, we find that even in Los Angeles, a geographically spread out city, quintupling the battery size of cars at no additional cost improves EV adoption by only 5%. Since the price of EVs is a more significant barrier to adoption than range, to promote EV adoption, policy makers should focus more on affordability than giving manufacturers credits for improving EV range.
Adepetu and S. Keshav. (2015). The Relative Importance of Price and Driving Range on Electric Vehicle Adoption: Los Angeles Case Study. Transportation Journal. : 1-21.
Recent Trends
The above two studies, written in 2015 and 2016, were conducted at a time when EV costs were relatively high. This meant that a large percentage of EV adopters were more affluent, and could afford to live close to their workplaces. This in turn meant that EV owners had very short commutes to work, causing battery size and range to be a minor consideration when purchasing a vehicle.
At this moment EVs are still not the cheapest alternative, so we will continue seeing a trend where car prices are a higher barrier to EV adoption than car range. We’re predicting that range will become a major factor in EV adoption in North America only when EV prices fall on par with gasoline car prices.
Utility-scale solar
Detecting anomalies in solar power generation
Solar panels lose efficiency when they’re covered with dust, dirt or snow, or are shaded. Although cleaning them is not a huge issue, keeping an eye on them to determine when they need cleaning is physically demanding, because solar panels are usually installed in inaccessible locations like rooftops. So as part of our study, we explored some automated methods to detect soiling.
In our first study, we used solar power traces to detect soiling using an algorithm we developed. We were able to detect ten types of anomalies, including temporary shading, permanent shading, fallen leaves, accumulating snow, and melting snow [1].
While working with solar panels, we noticed that shading affects power output in a different way than soiling. In our second study, we used solar power traces along with solar intensity information to detect obstructions of both types. Our detection algorithm achieved an 85% accuracy when tested on two real PV installations in Ontario, Canada [2].
[1] B. Hu. Solar Panel Anomaly detection and classification, MMath thesis, University of Waterloo, May 2012.
[2] X. Gao, L. Golab, S. Keshav, What’s wrong with my solar panels: a data-driven approach, Proc. Workshop on Energy Data Management, March 2015, pp. 86-93.
Allowing solar farms to enter into firm contracts with electric grid utilities (TSOs)
A large barrier to integrating solar power into the electric grid is the variability of sunlight. Because solar power generation depends on the intensity of sunlight in the moment, it’s tough to control solar energy generators to match fluctuations in demand from the electric grid. This means solar farms cannot confidently participate in the day-ahead electricity market, unless they add generous safety margins when committing to a certain level of power generation for the next day.
To allow solar farms to participate more confidently in the day-ahead electricity market, we developed a technique that considers the variability of solar intensity to predict how much electricity can be generated in the upcoming day. We also show how the same technique can be used to determine the best storage size for a solar farm to help meet fluctuating demands in energy. We tested our technique using a 10-year dataset, and found that it attains 93% of the maximum revenue that would have been achieved in the daily market if the entire schedule of the sun’s intensity was known ahead of time.
For energy storage specifically, our model determines the optimal size by taking into account the following three factors that affect solar power generation:
- The position of the sun in the sky
- Long-term cloudiness at time scales ranging from 10-minutes to a few hours
- Short-term cloudiness that last less than 10-minutes
Given a target output power and an allowable loss of power threshold, our technique computes a near-optimal storage size.
[1] Y. Ghiassi-Farrokhfal, S. Keshav, and C. Rosenberg. Firming Solar Power, Extended Abstract/Poster, Proc. ACM SIGMETRICS, June 2013.
[2] Y. Ghiassi-Farrokhfal, S. Keshav, C. Rosenberg, and F. Ciucu. Solar Power Shaping: An Analytical Approach, IEEE Transactions on Sustainable Energy, Vol 6, No. 1, Jan. 2015.
Optimal budget allocation
Suppose you had a million dollars to spend, some on solar panels and some on a storage system. While adding more solar panels increases the amount of power your solar farm can generate, adding more storage means you can source excess power on days with less sun so that the overall power supply is less varied. With this in mind, what would be the best way to split your budget if you’d like to maximize revenue by selling electricity in the day-ahead market? We provide an algorithm to determine this, taking into account factors such as:
- the power level committed by a solar farm owner
- the capacity of the transmission link between the solar farm and the power grid
- the need to curtail excess solar generation
- the degree in fluctuation in the purchase price of energy
- the inherent variation in solar generation due to fluctuations in sunlight
Ghiassi-Farrokhfal, F. Kazhamiaka, C. Rosenberg, and S. Keshav, Optimal Design of Solar PV Farms with Storage, IEEE Transactions on Sustainable Energy, October 2015.
SPOT: A Smart Personalized Office Thermal Control System
Problem
Heating, Ventilation, and Air Conditioning (HVAC) accounts for about 40% of the energy consumption in buildings. By changing the indoor air temperature of a building to be closer to the outdoor air temperature—for example, maintaining the building at a warmer temperature during summer months—HVAC energy consumption can be reduced by 10-40%. However, this comes at the cost of a reduction in individual comfort.
Solution
We have designed and implemented SPOT: a Smart Personalized Office Thermal control system. A SPOT device is placed in individual office spaces to heat or cool the immediate area to a comfortable temperature when an occupant is present. This allows the temperature of a building to be set to a value lower than normal in winter and to a value higher than normal in summer.
The first version of SPOT used 6 parameters to predict personal comfort: air temperature, radiant temperature, humidity, air speed, clothing level, and activity level. We made three iterations on SPOT’s design to improve its balance between energy conservation and personal thermal comfort, as described next.
SPOT
The first iteration of SPOT uses a Microsoft Kinect and a variety of sensors to measure the six parameters mentioned above. SPOT is able to calculate the amount of clothing a person is wearing using data from the infrared sensor and the Kinect. Once a user enters a room, SPOT measures these parameters, then controls a radiant heater to heat the workspace to a comfortable temperature. More details can be found here:
P. X. Gao. SPOT: A Smart Personalized Office Thermal Control System, MMath thesis, University of Waterloo, May 2013.
P. X. Gao and S. Keshav. SPOT: A Smart Personalized Office Thermal Control System, Proc. ACM e-Energy, May 2013.
SPOT+ (SPOT Plus)
SPOT+ improves upon SPOT by performing predictive control rather than reactive control. That is, SPOT+ will begin heating a workspace 10 minutes before a user walks in, so when they arrive the workspace is already at a comfortable temperature. It will also predict when a user will leave, so that it can begin cooling earlier to save energy.
After deploying SPOT+, we found that it reduced energy usage by 60% compared to a fixed temperature setting, and it reduces personal thermal discomfort from 0.36 to 0.02 (in the ASHRAE comfort scale) compared to SPOT.
P.X. Gao and S. Keshav. Optimal Personal Comfort Management Using SPOT+, Proc. BuildSys Workshop, November 2013. (Winner of the Best Student Paper Award.)
SPOT* (SPOT Star)
Our third version improves on SPOT and SPOT+ in 5 distinct ways:
- It provides both heating and cooling, unlike the previous iterations which only provided heating.
- It uses a speed-controlled desktop fan instead of a radiant-heater, making it possible to rapidly change the room temperature in response to discomfort.
- It is far less intrusive than the prior systems, because it does not use a camera.
- SPOT* is about an order of magnitude less expensive – while SPOT/SPOT+ were $1000 a unit, this prototype model costs only $185 a unit, with costs dropping to below $100 in mass production.
- Flexible placement of its software components allow for a balance between cost, privacy, and data durability.
In our deployment, we found that SPOT* improved user comfort by 78% compared to traditional HVAC systems.
Rabbani and S. Keshav, The SPOT* System for Flexible Personal Heating and Cooling, Poster, Proceedings of the 2015 ACM Sixth International Conference on Future Energy Systems (e-Energy), July 2015, 209-210. Best Poster Award
Rabbani and S. Keshav, “The SPOT* Personal Thermal Comfort System,” Proc. ACM BuildSys’16, November 2016.
Controlling an HVAC system with SPOTs deployed
To integrate SPOT systems into everyday use, we explore how to make HVAC systems SPOT-aware. We propose a control strategy to update the temperature set points for an HVAC system using the following factors: occupancy status in each room, preferred comfort requirements of occupied rooms, zones which have SPOT systems in it, outside temperature, and thermal properties of each room.
In a simulation setting, when users have homogeneous comfort requirements, we find that our system provides 45% savings in energy during the summer, and 15% during the winter compared to current predictive HVAC systems. When users have heterogeneous comfort requirements, our system provides 50% improvement in comfort in the summer and about 30% in winter, on top of significant energy savings.
Kalaimani, M. Jain, S. Keshav, and C. Rosenberg, ”On the Interaction between Personal Comfort Systems and Centralized HVAC Systems in Office Buildings, ” J. Advances in Building Energy Research, August 2018, V7:p.1-29.
Mitigating the impact of occupancy prediction errors in HVAC performance
Many commercial buildings use model predictive control (MPC) to control their HVAC systems – the model predicts outside air temperature and the number of people that will be in each zone of a building on a given day and adjusts the HVAC system accordingly. A prediction model cannot be perfect however – when the prediction errors of a model increase from 5% to 20%, the performance of the HVAC controller, as measured by occupant comfort and building energy use, becomes worse than that of a simple static scheduler that changes the temperature setpoint at the beginning and the end of the day.
We found that by employing the SPOT Aware strategy for HVAC systems, we stay in the acceptable region of occupancy comfort 95% of the time as opposed to only 83% when there are prediction errors in the MPC system. Thus, installing SPOT systems can not only save energy, but also make building occupants more comfortable, even in the presence of forecasting errors.
M. Jain, R. Kalaimani, S. Keshav, and C. Rosenberg, ”Using Personal Environmental Comfort Systems to Mitigate the Impact of Occupancy Prediction Errors on HVAC Performance,” Energy Informatics, 2018 1:60, https://doi.org/10.1186/s42162-018-0064-9, December 2018.