We intervened, as part of our Drive support offer , in the digital transformation of one of our customers in the road transport sector. Following the implementation of Office 365 and the successful transformation of uses for Microsoft Consulting firm, we were able to identify a need for a new application integrating Machine Learning. Thanks to our team of “data scientist ” consultants, it is indeed possible for us to go further than a simple PowerApps application and create applications based on Machine Learning that improve your productivity.
The project
The Machine Learning project that we are setting up aims to optimize the journeys and fuel consumption of road transport companies .
In this specific case, the problem was the increasingly variable diesel prices . This creates difficulty in finding the best station to refuel. Our objective : To optimize the taking of fuel according to the prices in the stations close to the whole route .
Concretely, during a given trip from point A to point B, the algorithm will tell the driver which gas station to fill up at.
The driver is thus informed directly and makes the best choice. The key to winnings on each trip !
The data used
In order to be able to train the Machine Learning models and develop the solution, a lot of data had to be processed.
We had access to approximately 8 years of historical data from our client, including journeys made, various information on incidents/delays, fuel intake, position/price.
We then sorted and filtered them for use in the templates.
Models / algorithms applied
The main part of the work consists in indicating which station to take for a given path . For this we use a path finding algorithm .
This algorithm is derived from a* (1968) which consists of a global graph (here the path of a truck) between an initial node and an end node . The algorithm represents the stations along the path as s intermediate nodes s .
To estimate the best path , we first use a model based on heuristic reasoning ( a method that calculates the set of feasible solutions ) on each node ( here the stations ) .
We then visit the nodes in the order of this heuristic evaluation in order to find the one that will offer the best Price/Distance ratio ( Compared to the global path) .
0 Comments