Managing a network of relay antennas to which several operators are connected, an international entity of the Orange Group wanted to rationalize its energy bill and reduce its carbon footprint by adjusting the size of its antennas to the actual needs of users.

The hunt for oversized antennas

This challenge required meticulous comparison work on the different antennas in the network, as Erwan Josse, Data Science and Artificial Intelligence Expert at Business & Decision, a brand of the Orange Group, explains: "The challenge was to be able to compare the different antennas we operate. To do this, we had to identify oversized antennas that were consuming too much electricity in relation to their throughput and traffic."

An innovative data modeling tool

Industrializing an existing "Proof of Concept" (POC), Business & Decision teams provided antenna maintenance teams with a data modeling tool. This enabled sites to be referenced by location, type of geography (urban, dense, rural), type of network (3G, 4G, 5G) and number of antennas installed. A comparison of consumption (Kw/h) as a function of traffic generated (GB) and average throughput (Kb/s) is thus carried out between similar sites via a classification algorithm: for each site, the algorithm determines whether consumption belongs to the low, standard or high range. It is an unprecedented tool in the field of relay antennas.

This innovation requires the use of Python – the programming language of AI specialists – but also close collaboration between data specialists and telecom maintenance experts.

Erwan Josse confirms: "The involvement of the business lines remains essential to understand the link between the variables observed. If an antenna consumes a lot of power, is it due to its role as a relay point or because it belongs to an older generation of equipment? We didn't want to bias the algorithm with erroneous interpretations."

Iterative meetings were regularly organized to gain a clearer understanding of telecom operators' uses and to adjust the modeling tool.

Energy and component savings

Observing significant disparities in consumption levels, the telecom maintenance teams were able to draw up a plan to prioritize their interventions.

The approach has paid off, with energy consumption on some antennas reduced by a factor of three. Their scalable resizing also limits the quantity of components used, such as carriers. What's more, the reuse of this equipment at other sites means it can be given a second life in a circular economy. These financial savings and resource conservation measures have no impact on network quality or service continuity.

Our algorithm uses the right amount of energy. Following the logic of frugal AI, we've aligned it to the degree of complexity of the project, and it is now delivering very good results.


Erwan Josse, Data Science and Artificial Intelligence Expert, Business & Decision

In the future, this algorithm could even become predictive: by being able to anticipate traffic on a new antenna, it would then help determine how to size it as accurately as possible. The data modeling approach, which can be reproduced in other fields, could thus contribute, for example, to limiting the energy consumption of buildings or machines on production sites.

Energy consumption reduced by 66% on some antennas