AI – a recourse for telecom operators in the face of soaring energy prices

Energy costs account for 23% of CSP’s network OPEX, and up to 17% of their environmental footprint. In this article, we present a four-step approach to understanding consumption factors and achieving substantial energy savings in the near future.

How? By using AI and analytics, with no additional network hardware investments or sacrifices in service quality.

Energy efficiency has become a priority topic for economic and ecological reasons, but also for operational resilience in the event of power cuts. These considerations must be considered in future deployments, especially 5G. However, we must not forget the modernization of existing infrastructures, especially 4G, 3G and 2G radio networks, which account for 70% of operators’ energy bills. In this area, significant improvements are within reach without the need to radically change existing installations.

Radio equipment usually has features that allow it to automatically reduce or turn off power at predefined thresholds. It is thus possible to temporarily cut off certain elements of the network in a given area, during a given time period. This static approach is already in use at many CSPs, but its impact is limited and difficult to assess, as it can only be used sparingly to maintain a good level of service.

In contrast, the massive collection of operating data and their analysis by artificial intelligence (AI) enables the deployment of a dynamic strategy for switching down or putting to sleep specific elements or resources – one which is targeted and adjusted to the actual traffic. It is then possible to know exactly how much energy is being saved and to maximize savings without compromising quality of service. Automation makes it possible to achieve these benefits without placing additional demands on staff.

As the highest level of energy management, AI-based energy optimization of mobile networks can be implemented with the following steps:

1. Physical modeling to estimate network footprint

First, we build a physical model of the different network components. The aggregation of these results provides an estimate of the global footprint of energy consumption for the network and its distribution by use to identify the main areas for progress. This approach makes it possible to evaluate the impact of different scenarios, but being purely static, it can only suggest structural, configuration or hardware changes, and not optimizations associated with instantaneous capacity requirements.

2. Detailed measurement to centralize real detailed consumptions

Optimizing energy consumption requires a detailed picture of real-life activity. We need to collect data on the different network elements and consolidate it in a sustainability data hub for analysis. In general, this data exists and is accessible, but some work remains essential to harmonize it and make

it intelligible. We can then get a precise view of consumption by zone, time slot and traffic level. This information will make it possible to carry out realistic simulations and to validate the benefits of the planned actions by calculation. However, at this stage, we still do not have the leverage to intervene directly on the network.

3. Statistical modeling to understand consumption patterns

Provided that there is enough history and diversity of data, Machine Learning tools can be used to establish a statistical model of the network’s energy consumption. As compared to physical modelling, more parameters can be factored in: types of equipment, nature of services, number of users, types of terminals, weather conditions, etc. Much closer to reality, these models can be used to identify the factors that have the greatest influence on energy consumption, to determine the optimal settings according to the situation, and to detect surprising or aberrant behaviors, revealing anomalies, misconfigurations or errors. Verizon has disclosed more than $100M per year in electricity savings thanks to their new ”energy digital twin” which follows this approach.

4. Dynamic optimization to get significant savings

Once in possession of these statistical models, the operator can precisely anticipate the traffic at the level of each cell, to within a few minutes, and can immediately determine which equipment to switch off or on, which sleep mode to activate, which energy saving features to use, in order to minimize energy consumption while maintaining the expected quality of service. To perform these actions, it is necessary to have automated equipment control (in compliance with cybersecurity rules). The implementation of central, dynamic and real-time reporting and management of energy consumption makes it possible to reach an energy optimum. Capgemini has tested such an approach with Project Bose.

Such a dynamic optimization of mobile networks is not a pipe dream. Experiments have demonstrated its feasibility, reaching 15-18% in documented cases. Moreover, these approaches can be implemented in a relatively short period of time, the most important part of the work being the constitution of a reliable and accessible data history and the integration with the operator’s tools and network elements.

Physical and statistical modeling bring an understanding of empirical consumption factors, and allow a first level of optimizations and associated gains. For further gain, more sophisticated algorithms and deeper integration with the network are necessary, requiring deeper network technology and data analytics expertise. It is important to start on a perimeter (for example: 4G vs 5G, in a given region), make sure that optimization does not impact customer experience, and then extend, in an incremental implementation approach.

What’s preventing you from starting today? Let’s talk about it!

TelcoInsights is a series of posts about the latest trends and opportunities in the telecommunications industry – powered by a community of global industry experts and thought leaders.