Energy Accounting Function for ML Network Energy Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML-based network energy saving strategies in radio access networks often fail to achieve significant energy savings due to infrequent decision-making and inefficient energy consumption across various lifecycle phases of ML operations, leading to potential net energy losses.
Innovation Solution
Implement an Energy Accounting Function (EAF) to track and log energy consumption across multiple phases of ML operations, including model training, inference, and data processing, calculating net energy balance (Q3) to identify and correct energy inefficiencies through network orchestration and management actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If ML-based network energy saving strategies are implemented, then network energy consumption is reduced, but the system complexity increases due to multiple lifecycle phases of ML operations
Solution Approach 1:
The patent segments the ML operations into distinct lifecycle phases (model training, model inference, data collection, data processing) and tracks energy consumption for each phase separately. This segmentation allows the system to identify which specific phases contribute most to energy consumption, enabling targeted optimization strategies that reduce overall network energy consumption without requiring complete system redesign.
Solution Approach 2:
The patent introduces an Energy Accounting Function (EAF) as an intermediary component that mediates between the complex ML operations and the network energy management system. The EAF collects, aggregates, and analyzes energy consumption data from various ML phases, translating complex operational data into actionable energy management insights, thereby simplifying the overall system architecture while maintaining comprehensive energy tracking.
2Loss of energy
If energy consumption tracking across all ML phases is implemented, then net energy balance can be calculated, but the measurement and logging complexity increases
Solution Approach 1:
The Energy Accounting Function (EAF) is designed as a universal component that handles multiple functions: collecting energy consumption data from different ML phases, aggregating data from multiple sources, calculating net energy balance, and generating optimization recommendations. This multi-functional design consolidates what would otherwise require multiple separate measurement and logging systems, reducing overall complexity while enabling comprehensive energy tracking.
3Reliability
If ML operations are performed frequently to improve decision-making, then network management quality improves, but energy consumption at ML model side increases
Solution Approach 1:
The patent implements a feedback mechanism where the Energy Accounting Function continuously monitors energy consumption across ML phases and provides this information back to the network management system. This feedback enables dynamic adjustment of ML operation frequency and intensity based on actual energy consumption patterns and network conditions, allowing the system to maintain high decision-making quality while optimizing energy usage by performing ML operations only when necessary and most effective.
Data Source
AI summary
A network node transmits at least one measurement configuration to one or more devices for carrying out one or more measurements for determining at least one energy consumption estimate for at least one operation in a plurality of inter-related machine learning operations for network management. The plurality of inter-related machine learning operations comprises (i) one or more operations for collecting data, (ii) one or more data transfer operations to transmit the collected data to the network node, and (iii) one or more data storage operations to store the collected data and (iv) one or more data processing operations to process the collected data. At least one energy consumption estimate is determined using the one or more measurements. At least one network management related configuration is determined for carrying out one or more of the plurality of inter-related machine learning operations on the basis of the at least one energy consumption estimate for controlling energy consumption.