Fluid Moving Device Operational Settings via Multivariate Time Series Motif Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data center cooling systems are computationally intensive, making them impractical for real-time adaptive control, leading to inefficiencies in power consumption and increased carbon footprints due to lengthy implementation times of traditional computational fluid dynamics-based modeling approaches.
Innovation Solution
A method and analyzer that automatically determine operational settings for fluid moving devices by identifying motifs in utilization data as multivariate time series patterns, filtering through frequent episode mining, and calculating sustainability metrics to optimize energy use while meeting heat dissipation demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If computational fluid dynamics-based modeling is used to achieve adaptive cooling control, then cooling efficiency is improved, but implementation time becomes excessively long making it impractical for real-time control
Solution Approach 1:
The patent creates simplified copies of the complex CFD models by training machine learning algorithms on historical CFD simulation data. These trained models serve as lightweight replicas that can predict cooling performance instantly without requiring full CFD computations, thus resolving the contradiction between accuracy and speed.
Solution Approach 2:
The patent performs comprehensive CFD simulations and data collection in advance to train the machine learning models. This preliminary action creates a pre-computed knowledge base that enables real-time control decisions without requiring lengthy CFD runs during actual operation.
2Productivity
If traditional CFD-based adaptive control is implemented, then cooling performance is optimized, but computational intensity makes it impractical for active control processes
Solution Approach 1:
The patent replaces the mechanical CFD computation system with a machine learning-based prediction system. The trained ML models perform rapid inference using minimal computational resources, substituting the heavy CFD mechanical computation process with a lightweight algorithmic approach that enables real-time control.
Solution Approach 2:
The patent uses inexpensive, computationally lightweight ML models that can be rapidly deployed and retrained compared to expensive CFD simulations. These simpler models consume far less computational power while providing sufficient accuracy for control applications.
3Adaptability or versatility
If detailed CFD modeling is performed for each subsystem, then adaptive operation capability is achieved, but the complexity and time required make it impractical for data center scale
Solution Approach 1:
The patent develops a universal machine learning framework that can be applied across multiple subsystems (cooling, power distribution, compute racks) with a single modeling approach. This universal method maintains adaptability across different data center configurations without requiring separate complex CFD models for each subsystem.
Solution Approach 2:
The patent segments the data center into manageable zones or subsystems and applies the ML modeling approach to each segment independently. This segmentation allows adaptive control to be implemented at appropriate scales without overwhelming computational complexity, enabling parallel processing and reduced model size.
Data Source
AI summary
In a method for determining operational settings for a plurality of fluid moving devices, one or more motifs in utilization data of the plurality of fluid moving devices collected over a time series is identified as a multivariate time series of data, sustainability metric levels for each of the one or more identified motifs are calculated, and a determination as to which of the one or more identified motifs have favorable sustainability metric levels is made.