Better decision making in public electric vehicle charging based on artificial intelligence
Written by Rebecca Fortmann and Christopher Burgahn
In our last blog post, we’ve described how crucial charging decisions are to make driving an electric vehicle (EV) truly efficient, sustainable and convenient. Where, when and how to charge decides if the driver finds a vacant charge point, if the charging electricity is based on renewable energy production and if the overall costs are low. But humans sometimes struggle with making the right charging decision, which creates a frustrating charging experience and thus blocks our path to CO2-neutral transportation.
We believe that artificial intelligence (AI) can play an important role in supporting humans and that such charging solutions embracing this technology will win the most EV drivers. With this blog post, we want to look deeper into the potential of AI by considering how the end-user experience in public charging could be improved. We hope this will inspire you to take the next step in developing the charging solutions of tomorrow. Let’s dive right in.
Samantha’s problem with public charging
To understand how EV drivers make decisions, let’s take a look at a real-life scenario of a typical EV driver. Imagine Samantha, 32 years old, living in a medium-sized European city and working as a software engineer at a local bank. She just recently decided to drive electric (with 200km of range) and she definitely loves it.
On a sunny Thursday afternoon, she gets a call from a friend, living in a city about 120km away. Her friend just got fired without any prior notice and therefore would enjoy some comfort from Samantha. They both decide that they will have dinner together at her friend’s place and that Samantha will stay over until tomorrow. Before Samantha starts her drive, she immediately thinks about charging as she knows that a little bit of planning is required. After checking on her favorite charging app, she finds a public 22kW charger close to her friend’s house — busy at the moment — and an HPC charger right before the exit on the highway. She decides that she will charge at the HPC charger, as she doesn’t want to risk being unable to charge her car that evening. She will not have the time the next morning right before work. No sooner said than done, she gets into her car, drives down the highway, charges at the HPC charger and arrives at her friend’s house. On arrival, she sees that the AC charger is now available and she could have charged there. She is frustrated as she has paid a lot of money for charging at the HPC charger already.
Why did she end up with this frustrating charging experience? Her decision, that turned out to be not optimal for her, is mainly based on three heuristics that humans regularly rely on in decision making: Availability, representativeness and adjustment (learn more about these heuristics here). All these also have influenced Samantha in her decision making:
- Availability heuristic: She just recently tried to charge at an AC charger but found it blocked. This is the most present memory regarding AC charging in her mind. She doesn’t want to experience this again.
- Representativeness heuristic: The AC charger next to her own home is used a lot and her neighborhood is quite similar to the one of her friend. At least she thinks that. So she projects the characteristics of the AC charger next to her place to the AC charger next to her friend’s place, without knowing if that is actually true.
- Adjustment heuristic: In her mind AC chargers are typically blocked. Even if she would invest more time in understanding the local situation at her friend’s place better, she most likely will not adjust her estimation about the availability enough to come to the conclusion that it will be available on arrival.
The overall lack of information and therefore relying on heuristics, combined with the typical high risk aversion of human beings, leads to Samantha going for the HPC charger — even if that means that she needs to pay more for charging. But can a smart charging solution do anything about this and help her out?
Why displaying statistical information is not enough
Often discussed is the use of statistical data for helping Samantha out. A charging application could display a histogram of usage data based on the time of the day and day of the week. This way Samantha could get an estimate of how likely it would be that the station is available on arrival. We all know this type of diagram from Google Maps.
Calculating the mean of the occupancy on Thursdays, 7 p.m. on historic data would approximate the a-priori likelihood P of the charging station being available:
Samantha’s charging app additionally knows the current status of the charger — it is blocked. Combining this information could be expressed as the a-posteriori likelihood of the AC charger being available in an hour given the AC charger is blocked right now:
In addition, the duration of the charging session could be taken into account:
But there is much more information available and some has already shown to be very important to change the estimate of availability, e.g. weather, events in the area and traffic. Do you see where this is going? Nobody wants Samantha or her charging app to gather all this information and do the math. Going for the statistical approach would be tedious and in the end maybe maybe not as accurate as needed. We think that Artificial Intelligence can take over, integrating information and coming up with a valuable recommendation about future availability.
How AI can make a difference
Most people know from biology class that the human brain is constructed of neurons and axons sending action potentials. Even though it’s a lot more complex and scienctists are still puzzled, this basic knowledge is enough to inspire computer science: In Neural Networks (only one part of AI), the basic structure of our brains is recreated. This is done to artificially create one of the most important functions for decision making: to learn from experience.
Humans are able to generalize or imitate from only a few samples. Depending on the task, artificial neural networks need thousands up to millions of samples to learn from them. As they focus on one single task, they may discover correlations between different parts of the data that a human would oversee. The “input” differs here: Humans have their personal experience, knowledge and situational perception, while Neural Networks can only read numbers. Both filter for useful information to solve the given task. While humans tend to (once decided) not change their filters for a given task (learn more about that in this blog post), Neural Networks stay completely rational and consider all given information equally. This can be a huge advantage, depending on the given task.
Looking at Samantha, she already learned from experience that the charger near her own home is often occupied in the evening. She thought that the charger near her friend’s home is used in a similar fashion. Maybe they are both placed in the city center, but some surrounding characteristics are different such as the next charging station, restaurants, etc. In contrast to the human decision making process, an AI can integrate multiple types of data that influence charging availability, but aren’t feasible to be considered manually by an EV driver. The historical usage of charging stations is just one part next to the unlimited possibilities of data of the surrounding area and similar locations. Such a system will be able to predict the availability of charging stations in the near future with high accuracy, thus giving Samantha a rational risk estimation for her charging decision. If her favorite charging application would have made use of such a system, paying too much at the HPC charger could have been prevented.
We need to make AI accessible for all charging applications
We believe that there are a lot of amazing charging products out there already, but at the moment of writing, they are just not good enough in helping EV drivers in their charging choices. Especially in times and areas where the number of public charge points grows slower than the number of EVs on the streets, we need to be better in helping drivers to find vacant charge points.
The use-case described in this blog post is only one of many where integrating advanced AI technology into a charging application can make the difference between a “frustrating” and a “best-in-class” charging experience. We know that AI can be troublesome and requires lots of up-front investment. That’s why we made it our mission to make the best AI technology as easy as possible to integrate into our customers’ charging applications. Stay tuned for more.
