Reducing EV range anxiety: How a simple AI model predicts port availability

Google Research just published something refreshingly unglamorous: a linear regression model for predicting EV charging port availability. No transformers, no deep neural nets, no billions of parameters. Just old-school statistics, tuned carefully and shipped fast.

The problem they're solving is real. Range anxiety isn't only about battery percentage — it's about arriving at a charging station and finding every port occupied. Google's team wanted a model that could answer a narrow question: given a station right now, what's the chance a port opens up in the next 30 or 60 minutes? They tested decision trees and small neural networks first, then found the simplest option won on both accuracy and, crucially, latency.

The model's core feature is almost embarrassingly basic: the hour of the day, treated as a set of independent buckets rather than a continuous number. Each hour gets its own learned weight, and that weight tells you whether ports are filling up (positive, like 7 a.m. commuters plugging in) or emptying out (negative, like 5 p.m. drivers heading home). Near-zero weights, say at 3 a.m., mean nothing much happens and the model just shrugs and repeats the current count.

That baseline it's up against — called Keep Current State — sounds trivial but is genuinely hard to beat. On the US East Coast, fewer than 10% of ports change status in any given 30-minute window, so guessing 'no change' is right most of the time by default. Yet Google's regression model still cut wrong predictions by roughly 20% during the 8 a.m. rush and about 40% around 8 p.m., specifically at larger stations where turnover swings are biggest.

One detail stands out: separate models for California and Germany outperformed a single pooled model, even though the shape of the daily occupancy curve looked similar in both regions. Driver habits vary by geography even when the underlying pattern — commute in, commute out — stays the same. Google says the next step is stretching the prediction window further out, aimed at long-distance trip planning rather than just the next hour.