How AI can revolutionise battery energy management systems
The Plug-in Hybrid Toyota Prius toiling in the downtown Los Angeles traffic looks like any other. Essentially it is no different to the other cars and trucks caught up in the latest gargantuan jam on Interstate 5, one of Los Angeles’s busiest sections of road. But curiously the driver of the Prius is not exhibiting the same blank and defeated expression of many of the other motorists who sit uncomfortably bumper to bumper in this five-mile tail back. On the contrary, he seems happy that his car is mired in the city gridlock. That’s because this is no ordinary vehicle and this is no ordinary journey.
While it is true that the man at the wheel of the car is a commuter, Dr Xuewei Qi is also a scientist collecting data for an Artificial Intelligence project that could one day help to eliminate ‘range anxiety’ in electric vehicles, and in the shorter term, reduce energy consumption and toxic exhaust emissions in hybrid electric vehicles.
Dr Qi, who is a researcher faculty member at the University of California, Riverside, says the new innovation, which has taken four and half years to create, develop and test, “is game-changing technology, as it will enable equipped electric and hybrid vehicles to use energy, generated from the battery, much more moderately and efficiently than current systems allow”.
Qi, who belongs to the University’s Intelligent and Sustainable Transportation Research Group at the Center for Environmental Research and Technology (CE-CERT), led by Professor Matthew J. Barth, which also includes two research engineers, Guoyuan Wu and Kanok Boriboonsomsin, explains, “When it comes to alternative fuel and hybrid vehicles, there are a myriad of options for the consumer. But this has only made the energy management system (EMS) landscape more complex.”
Qi says, “Today, anyone who purchases a Plug-in Hybrid Vehicle (PHEV) or standard hybrid car has no choice but to use the EMS of that vehicle providing the driver with the most efficient and cost-effective drive. Our research quickly revealed that none of that the computerised technology currently found in these vehicles took into account speed, longitude, latitude and elevation when making calculations. Nor were they able to evaluate the road gradient, and real-time battery charge and fuel consumption. Most, PHEVs simply revert to conventional fuel once the battery is drained.”
Dr Qi, who has spent the over six years researching evolutionary computation, including, how swarm intelligence can transform connected and automated vehicle technology, continues, “Our findings led us to Artificial Intelligence. So we decided to create a state-of-the-art algorithm, which leverages deep learning capability – one which could not only react to its surroundings, but learn from them too.”
A Ground-breaking solution explained
So how does the technology work?
“The algorithm, which we have created divides each car journey into small portions, which range from ten seconds to a minute. The algorithm, which works in real-time – meaning a Control decision is made every 0.1 seconds – evaluates a series of data (including the speed of the vehicle, road elevation, gradient) and compares this information to current battery life and how much fuel the vehicle has consumed. If, in carrying out its calculations, the algorithm verifies the next segment of the journey is the similar to the last, then it sends a message, via the Electronic Control Unit, to the Energy Management System, telling the EMS to behave in exactly the same was as in previous section of the trip. If the next segment of the journey is never encountered before, then the system is also able to estimate the optimal control strategy based on what it has learned from the past journey.”
So far, the results of the study, which is partially being funded by the US National Center for Sustainable Transportation to the tune of USD$80,000, are astonishing as they are impressive. When Dr Qi, Professor Barth and the wider CE-CERT team compared their algorithm to that of a standard Energy Management System for a PHEV, they discovered that their invention is 10.7 per cent more efficient than a text-book standard EMS. Even more staggeringly, a series of tests revealed that when the opportunistic charging information is considered, the energy savings can amount to a whopping 31.5 per cent.
But, there is an important caveat to stress - the technology has not yet been tested in real-world OEM proving grounds.
Instead, using MatLab Simulink for algorithm design and testing, and Autonomy for vehicle model simulation, real-life car journeys are fed into a leading edge simulator, and comparisons and calculations made. But how can Qi be sure that the data collected truly mimics that of a real-life traffic scenario?
Qi says, “We have replicated live traffic situations by collecting information using an On Board Data Logger (OBD) mainly deployed on Interstate 210 - one of the main highways, which cross-crosses Greater Los Angeles from west to east - over a four-month period, making a series of 40 minute trips on a range of different days, including weekdays and weekends. Each day, our vehicle spent four hours on the road, which included both the morning and evening rush hour.”
Qi, who made many of the road trips himself, explains how the OBD logger gathers relevant data.
“The Data Logger, which is plugged into the OBD port under the steering column of the vehicle is directly linked to the car’s Electronic Control Unit (ECU) which takes into account vehicle speed, and also the geography, elevation and route, so that it can assess how much power is needed to compete the journey in the most efficient way possible. We then relay this data to our simulation software,” he says.
But, if, and when, a major automotive company agrees to test the system in its vehicles, just how easy will it be to convert the technology for real-world use?
Explains Qi, “When applying the algorithms to a real vehicle, we only have to tailor the algorithm accordingly and update the software in the car’s Electronic Control Unit (ECU). In terms of infrastructure, the technology need to be integrated into the vehicle’s ECU and needs to connect to an in-car visual display unit located on the dashboard, which would relay real-time energy management data to the motorist. So the system engineering is fairly straightforward.”
Operating outside of V2i Connectivity architecture
Instead, the one major obstacle that the CE-CERT team had to overcome was how to ensure that the vehicle knew every inch of the natural environment surrounding it. “The simulation that we created is based on a known journey,” begins Qi.
“But we anticipated from the early stages of the project that, in real-world conditions, there would be many scenarios that the technology simply would not have encountered.
Dr Qi believes that the advent of connected V2V and V2i landscapes, such as Mcity, a Connected Vehicle currently project being developed by the University of Michigan’s Mobility Transformation Center, and, also the initiatives being trialled by the US Department of Transport in the states of Wyoming, Florida (Tampa) and in the city of New York “will one day provide a ready-made solution”.
He explains, “When these projects are rolled-out across the world, and when cars can communicate with each other, if a vehicle is embarking on a journey for the first time, the algorithm in the vehicle will simply be able to receive the trip data from the another car in real-time or via the local traffic management centre server, which will also be able to receive and distribute journey information in real-time.”
However, while there are number of these next-generation initiatives exist globally, most are at pilot stage or still in concept phase. Therefore, in the interregnum period before vehicle-to-vehicle and vehicle to infrastructure systems becomes the norm, how has the CE-CERT team managed to solve this conundrum?
Says Qi, “While the cellular network, crowdsourcing platforms and innovative V2V and V2i schemes are key enablers, they are still unable to provide the intense information sources, such as real-time traffic conditions around the vehicle, so the car’s speed can be predicted. Therefore, we created a ‘deep-learning system’ which would work on today’s road networks where connected environments are still few and far between.
Therefore, any car fitted with the technology would have the capability, through AI, to learn how to save fuel from past driving records. However, if historical data is not available, the deep learning system is capable of carrying out an informed estimation, based on real-time data, in regard to what the most efficient power control is likely to be. It them transmits its findings to an on-board database, which stores the information for future journeys.”
AI: A cure for range anxiety?
But what do leading automotive experts think of this pioneering and potentially break-through technology system?
Iain Mowat, a senior analyst for Wood Mackenzie, who specialises in forecasting the impact of electric vehicles in the European car fleet, says, “While I am unable to comment too much on the exact specifics of this particular project, what I can say is that improved battery energy management will support improved EV fuel economy, and so enable improved range without increasing the size of the battery pack. Thus, the technology could be a contributing factor in improving overall range per charge.”
Mowat, who has worked for Wood Mackenzie for five years, believes any technology which could one day help cars to use energy more conservatively and sparingly, could help ease ‘range’ challenges at the lower end of the market.
Says Mowat, “Range anxiety is being resolved through increasing the BEV battery pack to around 60kWh, which provides sufficient range for a mainstream family car.
This is a more feasible strategy for the luxury end of the EV market, as the cost of a 60 kWh battery is currently prohibitive for less expensive cars. Thus, high battery costs are a key constraint on the future deployment of BEVs, particularly in the small and lower medium segments, where ICE cars are less expensive and have better fuel economy compared to the executive and upper medium segments."
And with range anxiety being regarded “as one of the three key obstacles holding back EV adoption”, Anil Valsan, EY’s Global Automotive and Transportation Lead Analyst, says, “There is a wider industry debate about how we solve this most pressing of issues: finding an answer to combat range issues is not just down to technology. Consumer education can play a part too. A significant proportion of owners overcharge their vehicles, which limits range in the long-term. In the same way as a person charges a lap top computer, battery life, and in a cars’ case – range – can be extended by charging the battery at the right time and for the right amount of time.”
Valsan, who has worked for EY for the last seven years and graduated at LSE, also believes that “governments” and “cross-industry collaboration” can also go a long way to quelling driver’s fears.
“OEMs and legislators need to demonstrate to consumers that the current range of EVs will often meet the majority of their commuting and mobility needs. But for longer journeys – perhaps through innovative next-generation car-sharing schemes - range anxiety can be solved by making conventional and/or hybrid vehicles widely available for consumers who wish to undertake longer journeys. This would mean that potential EV converts don't discount zero-emissions vehicles because of a few long journeys,” he adds.
Meanwhile, back in California, the CE-CERT team, according to Dr Xuewei Qi, “is currently negotiating with a number of leading automakers with the purpose of commercialising our algorithms in their commercial vehicle models”.
If talks are successful, and this ground-breaking energy management system is adopted universally by OEMs, then technology alone could go a long way to banishing the collective fear we all have of running out of gas on a busy motorway.