Reducing network energy consumption has become central to communication service providers’ (CSPs) business operations and can help them drastically reduce operating costs. In fact, according to Nokia, energy consumption accounts for nearly half of all telco network operations cost, and as a result, an astonishing 78% of telcos are turning to artificial intelligence (AI) to help them cut down on this energy use.
A recent Nokia white paper further claimed that, within the Radio Access Network (RAN), the base stations and the cooling systems are the two biggest power users, consuming 36% and 55% of the network’s total, respectively. Therefore, effective energy management becomes about using artificial intelligence to shut down certain parts of the RAN dynamically and automatically when they are not needed, reducing energy waste and operating costs. AI-based solutions, said Nokia, can achieve two to five times more savings than non-AI systems that perform temporary shutdowns based on fixed schedules.
“A typical network consumes 300 to 350 GWh electricity per 10K base station sites,” revealed Mohammad Nur-A-Alam, Nokia’s head of sustainable products. He added, though, that an intelligent, comprehensive and end-to-end energy management solution can reduce this consumption by up to 35%.
However, energy savings tactics that require operators to temporarily shut down network elements come with potential risks. In the case of an unexpected traffic event, for instance, the network may underperform, resulting in a degraded customer experience. Today, equipment shutdowns are performed according to fixed schedules that do not consider live, real-world conditions, and therefore, these shutdowns risk leaving the network short-handed.
Dynamic, then, must also mean acknowledging the distinction between the traffic profile and the required network performance. AI can help a CSP prioritize high-profile customers or customers with critical processes, for example, so if a failure occurs, the energy management system can “wake up” a sleeping base station to ensure that customer KPIs are maintained. Put another way, the AI system dynamically calculates the threshold based on the required performance every day for every site and cell.
“It is most important is to have a mature and field-proven AI model that provides greater than 90% prediction accuracy by keeping network performance KPIs and counters intact,” said Nur-A-Alam. “Mobile networks are critical network infrastructure and should always maintain high reliability and should have the capability to react in real time. A good AI model must consider this.”
By predicting network traffic and adjusting shutdown times dynamically, AI and machine learning make it possible to both extend energy — and therefore, cost — savings compared to static schedules while avoiding any degradation of network performance. It does so by predicting the network needs based on the latest load and network performance feedback, which can differ in a variety of ways including metrics like location — rural vs urban — or type of traffic — critical or consumer.
In this way, AI can perform precise predictions to balance energy savings, network performance and customer experience requirements, keeping the required network performance and the savings windows in sync to ensure that network KPIs are not violated.
