Root Cause Identification for Smart Grid Time-Series Events

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current machine learning-based anomaly detection methods for time series data in smart grid systems require manual labeling by domain experts, are time-consuming, and lack automatic identification of root causes, making them inefficient and error-prone.

Innovation Solution

A method and system that automatically identify the root cause of events in smart grid systems by acquiring time series data, using an ensemble of secondary detection algorithms to generate statistical labels, and performing queries in a semantic database to determine causal factors and root causes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual labeling by domain experts is used to train machine learning-based anomaly detection methods, then the detection accuracy can be improved, but the time consumption and cost increase significantly

Engineering Contradiction:
Improveanomaly detection accuracyVSAvoidtime consumption for manual labeling
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-service by automatically generating statistical labels through ensemble detection algorithms and self-supervised learning, eliminating the need for manual expert labeling. The system uses its own detection results to train and improve its performance autonomously.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Statistical labels serve as an intermediary between raw time series data and final anomaly detection results. These automatically generated labels act as a bridge that enables machine learning models to be trained without direct human intervention, reducing time consumption while maintaining detection accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If manual labeling by domain experts is used, then the detection system can be trained, but the process becomes error-prone and contingent on varying expert understandings

Engineering Contradiction:
Improvedetection system training reliabilityVSAvoidcomplexity of manual labeling process
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system changes the parameter of label generation from human-dependent (expert opinion) to algorithm-dependent (statistical computation). By transforming labels into quantifiable statistical metrics generated by ensemble algorithms, the system eliminates variability between different experts and reduces errors associated with manual labeling.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The manual mechanical process of expert labeling is replaced with an automated computational system. The ensemble of detection algorithms and statistical label generation mechanisms substitute the human expert's manual work, providing consistent and reproducible results without human error or variability.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Ease of operation

If traditional anomaly detection methods are used, then events can be detected, but automatic identification of root causes and causal factors is not provided

Engineering Contradiction:
Improveease of event detectionVSAvoidloss of causal explanation information
Core Design Contradiction:
Ease of operationVSLoss of information

Solution Approach 1:

The system segments the anomaly detection process into distinct functional modules: event detection, statistical label generation, causal factor identification, and root cause analysis. This segmentation allows each component to specialize in its function while maintaining overall system coherence, enabling both easy operation and comprehensive information retention.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system adds a new dimension to anomaly detection by incorporating temporal and causal analysis. Instead of only detecting events in time series data, the system extends into the causal dimension by identifying causal factors and root causes, providing a more comprehensive understanding of detected events without complicating the basic detection operation.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

4Productivity

If machine learning methods are trained offline using labeled data, then event detection capability is achieved, but automatic coupling to root cause analysis tools is not possible

Engineering Contradiction:
Improveevent detection productivityVSAvoidautomation of root cause analysis coupling
Core Design Contradiction:
ProductivityVSExtent of automation

Solution Approach 1:

The system achieves multi-functionality by integrating event detection, statistical label generation, and root cause analysis into a single unified platform. The statistical labels generated serve multiple purposes: training detection models and providing input for causal analysis, enabling automatic coupling between detection and analysis functions without requiring separate systems.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system merges previously separate functions (anomaly detection and root cause analysis) into an integrated workflow. By combining the detection algorithms, statistical label generation, and causal factor identification into one system, automatic coupling is achieved, improving both productivity and automation extent simultaneously.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4328687B1Method and system for identifying a root cause of an event
Publication Date: 2025.03.05 SIEMENS AG OESTERR
  • EP4328687B1 patent drawingFigure 1~2
  • EP4328687B1 patent drawingFigure 3A~3D
  • EP4328687B1 patent drawingFigure 4

AI summary

The present invention concerns a method and a system (110) for automatically identifying a root cause of an event occurring within a monitored system (120), wherein time series data representing temporal evolutions of a physical quantity value measured for one or several components (121-129) of said monitored system (120) are acquired and processed for detecting said event, and queries are successively and automatically generated for searching in a semantic database, and using statistical properties of the time series data, causal factors for the detected event, wherein said causal factors are then used for determining, from said semantic database, the root cause of the detected event.