Sliding Window Data Sample Selection for Power Management

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing power consumption management systems face challenges in efficiently handling concept drifts and anomaly detection due to computing resource limitations, leading to performance degradation and excessive computation costs, particularly in managing energy consumption data streams.

Innovation Solution

A device and method for selecting data samples using a sliding time window approach, where past samples are categorized into two shares: a first share of recent samples kept irrespective of similarity criteria and a second share subject to elimination based on similarity criteria and current samples, allowing for a balanced tradeoff between time forgetting and similarity forgetting to adapt to concept drifts and anomalies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a growing amount of data samples is gathered over time for ML processing, then the relevance of predictions and failure detections should be enhanced, but computing resource limitations (processing time and memory) force the introduction of significant approximations and lead to performance degradation

Engineering Contradiction:
Improveprediction relevanceVSAvoidcomputing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts only the most relevant data samples from the complete dataset for ML processing. By identifying and selecting a subset of samples that are most informative for detecting concept drifts and anomalies, the system avoids processing the entire dataset, thereby reducing computing resource consumption while maintaining prediction relevance.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different processing strategies to different portions of the data based on their relevance. Recent data samples are processed with higher priority and retained longer, while older samples are processed differently or discarded. This local differentiation in data handling optimizes the balance between memory usage and detection accuracy.

Inventive Principle:
Principle #3Local quality

2Stability of the object's composition

If too much past information is remembered in the ML model, then historical patterns are preserved, but outdated data interferes with presently relevant ones and distorts behavioral interpretation

Engineering Contradiction:
Improvehistorical pattern preservationVSAvoidconcept drift adaptation
Core Design Contradiction:
Stability of the object's compositionVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic data retention where the amount of historical data kept in the ML model varies over time. Recent data samples are retained longer to capture current behavioral patterns, while older samples are progressively discarded. This dynamic approach allows the system to adapt to concept drifts while maintaining stability through selective memory of historical patterns.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent proactively identifies and removes outdated data samples before they can negatively impact model performance. By monitoring data age and relevance metrics, the system预先 eliminates samples that are likely to become outdated, preventing distortion of behavioral interpretation rather than reacting after damage occurs.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If forgetting techniques are applied to dispose of old data, then computing resources are saved, but rare events that need consideration are lost

Engineering Contradiction:
Improvecomputing resource efficiencyVSAvoidanomaly detection capability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent employs feedback mechanisms to monitor the effectiveness of data forgetting. By tracking detection performance metrics and comparing expected versus actual anomaly detections, the system adjusts its data retention and forgetting strategies to ensure rare events are not lost while maintaining computing efficiency.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent dynamically adjusts parameters controlling data retention and forgetting based on detected patterns. When rare events or anomalies are detected, the system modifies forgetting rates to preserve such critical data. When normal patterns prevail, more aggressive forgetting is applied to save computing resources.

Inventive Principle:
Principle #35Parameter changes

4Adaptability or versatility

If the ML model continuously learns on newly received data without fully forgetting the past, then adaptation to concept drifts is enabled, but catastrophic interference occurs where previously learned information is suddenly and massively forgotten

Engineering Contradiction:
Improveconcept drift detectionVSAvoidlearned information retention
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the data processing into distinct phases: a training phase where the ML model learns from historical data, and a detection phase where it monitors for concept drifts. During the training phase, comprehensive data is retained; during detection, selective forgetting is applied. This segmentation prevents catastrophic interference by separating learning from adaptation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements periodic retraining cycles where the ML model is fully trained on accumulated data, then operates in detection mode with selective forgetting until the next retraining cycle. This periodic alternation between full learning and selective forgetting prevents catastrophic interference while maintaining adaptability to concept drifts.

Inventive Principle:
Principle #19Periodic action

Data Source

PatentEP3985582A1Devices and processes for data sample selection for power consumption management
Publication Date: 2022.04.20 CRAFT AI
  • EP3985582A1 patent drawingFigure 1
  • EP3985582A1 patent drawingFigure 2A~2D
  • EP3985582A1 patent drawingFigure 3~4

AI summary

Data samples are selected for power consumption management. This comprises receiving samples associated with respective times, distributed in a sliding time window as current samples (23) and in a past period as past samples (25). Selected past samples are determined by keeping a first share (26) of the past samples, consisting in most recent ones, and a second share (29) through eliminating among the past samples deprived from the first share, called a complementary share (27), part of the past samples in function of at least some of the current samples and of elimination conditions (32) depending on similarity criteria (31) applied to at least the first and complementary shares. The selected past samples are provided with the current samples for power consumption management. Applications to power failure detection and power consumption dynamic adaptation.