Author: Prof. Panos Georgakis / University of Wolverhampton 

GREEN-LOG is a European project building smarter, greener city logistics; think delivery bikes instead of vans. Behind the scenes, a suite of AI and machine-learning tools does the heavy lifting. Here’s what they do. 

A nowcasting and forecasting engine tries to answer two questions minutes before they matter: how busy will the roads be, and how many parcels will need to be picked up in a given neighbourhood? It analyses recent sensor data, including average car speeds, vehicle counts, and past delivery patterns and projects them six hours ahead. To do this, it relies on a mix of statistical and machine-learning methods capable of spotting repeating patterns, for example, rush hour every weekday, slower weekends, and extending them into the near future. Others work like a panel of experts voting: many simple rules are combined, each correcting the last one’s mistakes, until the group’s guess is quite accurate. A related tool, the spatio-temporal demand prediction service, tackles a similar problem over longer horizons, aiming to predict parcel demand roughly a week out. It uses a pair of models: one good at recognising spatial patterns, such as which neighbourhoods behave similarly, and one good at remembering trends over time, such as demand rising before a holiday. 

A dynamic routing service tries to solve the puzzle of “who should deliver what, in what order, given how much each bike can carry and when each customer needs their parcel.” Rather than trying every possible combination (which would take forever), it uses a smart trial-and-error approach. It starts by building a decent starting plan, then repeatedly rearranges it by swapping stops and relocating deliveries until no more improvements can be found. It also reacts on the fly: if a new pickup request comes in or a bike breaks down mid-route, it recalculates instantly rather than waiting for the next planning cycle. 

Together, these tools can give city planners and logistics companies a sharper, faster set of educated guesses to work with. For city authorities, that means spotting congestion before it forms and proactively adjusting traffic flow rather than reacting after the fact, improving both safety and air quality on the streets most affected by delivery traffic. For logistics companies and the local shops that rely on them, it means knowing roughly how many parcels to expect and where, so vehicles and staff aren’t sitting idle in some areas while overwhelmed in others. Routes are recalculated the moment something changes, such as a late order or a broken-down bike, so delays don’t cascade through the rest of the day’s deliveries. The net effect is fewer wasted trips, fewer empty miles, and a logistics network that adapts in near real time rather than running on yesterday’s guesswork.