Page 363 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Density-Based Spatial Clustering of Applications with Noise, KMeans,
and LightGBM realize profiling analysis of tide, scenario, coverage, and 4.3 - 5G
power evaluation.
LSTM and CNN predict time-series KPIs (loads, traffic, etc.) using data
Model Training and graphs.
Fine-Tuning Reinforcement Learning techniques such as Q-learning and Butterfly
Optimisation Algorithm, optimize threshold and detection upgrades by
fine-tuning parameters based on rewards.
LLM utilizing CoT, Agent, and RAG frameworks with the DeepSeek model
enables smart Q&A, automatic data analysis, and automated operations.
The use case has been deployed in 4/5G Base stations and equipment
Testbeds or Pilot rooms within China Telecom. This use case is part of a large-scale deploy-
Deployments ment and continuous optimization initiative across China Telecom’s
national network.

2 Use Case Description

2�1 Description

The rapid expansion of 4G/5G base stations, data centers, and equipment rooms has led
to soaring energy demands, which are exacerbated by fragmented, multi-vendor device
management, low digitalization levels, and unpredictable service loads. Traditional energy-
saving methods often struggle with high fault rates, slow emergency response times, and
limited operational efficiency, hindering progress toward sustainable digital infrastructure.

To overcome these challenges, a unified AI-powered platform has been deployed to achieve
real-time energy optimization and intelligent operations and maintenance (O&M) across
complex network environments. The platform integrates a centralized decision-making engine
with edge control systems, enabling adaptive, automated control of diverse infrastructure
devices, including base stations, servers, and cooling systems.
The key innovations include:

Adaptive Energy Control: The platform continuously aggregates real-time data, enabling
dynamic adjustment of device operating thresholds and cooling levels. It automatically identifies
high-load scenarios and implements energy-saving measures based on service demands.

Intelligent O&M and Automation: AI-powered agents, supported by modular tools and a
domain-specific knowledge base, deliver predictive maintenance, automated inspection,
anomaly detection, and real-time reporting [3]. This reduces fault resolution time from days
to minutes and increases operational efficiency by 90%.

Edge-Cloud Collaborative Control: Edge control engines and cloud decision systems
coordinate multi-device operations, optimizing energy consumption while ensuring service
continuity and stability. This overcomes the limitations of manual operations and static controls,
delivering consistent performance even under variable load conditions.

327