AI Power Prediction Using Digital Twins for Mixed-Spec Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models face challenges in providing power consumption predictions in complex network arrangements with non-homogenous or unknown device specifications, due to the lack of available training data and the complexity of dynamically changing variables.
Innovation Solution
The system generates a device profile repository that includes both known and virtual specifications, creating a 'digital twin' of devices to complete incomplete data sets and train AI models for power consumption predictions, using network mappings and virtual specifications to infer relationships and specifications for devices with unknown specifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If artificial intelligence models are used to detect and make predictions on dynamically changing variables in complex network arrangements, then power consumption predictions at application-specific level can be achieved, but the lack of available training data and incomplete data on non-standardized hardware components prevents proper model training
Solution Approach 1:
The patent creates virtual device profiles that copy and simulate the characteristics of physical devices with unknown specifications. These virtual profiles serve as digital twins that replicate hardware behavior patterns, allowing the AI model to train on synthesized data that mirrors real device performance without requiring actual physical devices or their proprietary specification data.
Solution Approach 2:
The system performs preliminary actions by proactively creating virtual device profiles and synthesizing training data before the AI model needs to make predictions. This advance preparation involves inferring device characteristics from available network data, generating realistic power consumption patterns, and building a comprehensive training dataset that enables the model to function immediately upon deployment.
2Measurement precision
If traditional measurement methods are used to monitor power consumption, then current power consumption can be measured for specific devices or entire data centers, but it is difficult or impossible to measure power consumption at the application level in complex network arrangements
Solution Approach 1:
The patent segments the network monitoring system into hierarchical layers: network level, device level, and application level. By dividing the complex network arrangement into these manageable segments, the system can apply different measurement and inference techniques at each level, ultimately achieving application-specific power consumption measurements through the aggregation and analysis of segmented data.
Solution Approach 2:
The system introduces virtual device profiles and AI models as intermediary layers between the physical network infrastructure and the application layer. These intermediaries translate complex network-level measurements into application-specific power consumption data, bridging the gap between measurable network parameters and the desired application-level metrics.
3Adaptability or versatility
If device profiles with known specifications are used for training, then standardized hardware components can be accurately modeled, but custom or specialized hardware components with non-homogenous specifications cannot be properly represented
Solution Approach 1:
The system dynamically changes device profile parameters by creating virtual specifications that adapt to custom and specialized hardware. Instead of using fixed, pre-defined device profiles, the system generates parameter sets that reflect the actual characteristics of diverse hardware components, allowing the same framework to accurately represent both standardized and custom devices through parameter variation rather than structural change.
Data Source
AI summary
Systems and methods are described herein for novel uses and/or improvements to artificial intelligence applications in an environment with limited or no available data. In particular, systems and methods are described herein for providing power consumption predictions for selected applications within network arrangements featuring devices with non-homogenous or unknown specifications.