The Indian Express (Delhi Edition)
Intelligence test
A computer programme that learns how to save fuel
FROM AVOIDING jaywalkers to emergency braking to eventually, perhaps, chauffeuring the vehicle itself, it is clear that artificial intelligence (AI) will be an important part of the cars of the future. But it is not only the driving of them that will benefit. AI will also permit such cars to use energy more sparingly.
Cars have long had computerised enginemanagement that responds on the fly to changes in driving conditions. The introduction of electric power has, however, complicated matters. Hybrids, which have both a petrol engine and an electric motor run by a battery that is recharged by capturing kinetic energy as the vehicle slows or brakes, need more management than does a petrol engine alone. Things get even harder with plugin hybrids, which can be recharged from the mains and have a longer electric-only range.
This is where AI could help, reckon Xuewei Qi, Matthew Barth and their colleagues at the University of California, Riverside. They are developing a system of energy management which uses a piece of AI that can learn from past experience.
Their algorithm works by breaking the trip down into small segments, each of which might be less than a minute long, as the journey progresses. In each segment the system checks to see if the vehicle has encountered the same driving situations before, using data ranging from traffic information to the vehicle’s speed, location, time of day, the gradient of the road, the battery’s present state of charge and the engine’s rate of fuel consumption. If the situation is similar, it employs the same energy-management strategy that it used previously for the next segment of the journey. For situations that it has not encountered before, the system estimates what the best power control might be and adds the results to its database for future reference. Ultimately, the idea is that the algorithm will also learn from the experiences of its brethren in other cars, by arranging for all such systems to share their data online.
Ideally, such a system would be fed its route and destination in advance, to make things easier to calculate. But Dr Qi and Dr Barth are realists, and know that is unlikely to happen. If a satnav were invoked, it would be able to pass relevant information on to the algorithm. But drivers use satnavs only to get them to unfamiliar destinations. Hence the researchers’ decision to design a system that does not rely on knowing where it is going.
It seems to work — at least, in simulations. Using live traffic information to mimic journeys in southern California, Dr Qi and Dr Barth compared their algorithm with a basic energy-management system for plug-in hybrids that simply switches to combustion power once the battery is depleted. As they report in a paper to be published in IEEE Transactions on Intelligent Transportation Systems, their system was 10.7% more efficient than the conventional one. If the system is aware in advance that a recharging station will be used as part of the trip (which might be arranged by booking one via the vehicle’s information screen) it can spread the use of electric power throughout the journey, to maximum advantage, knowing when the battery will be topped up. In such situations the average fuel saving was 31.5%.
Dr Qi and his colleagues now hope to work with a carmaker to test the algorithm on real roads. If all goes well, and their system proves able to cope with the nightmares of commuting in southern California, they will not be left stranded on the hard shoulder.