Cognitive Anomaly Detection System for Energy Consumption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting energy consumption anomalies in buildings and refrigeration systems rely on unsupervised machine learning, which often results in false alarms due to contaminated data and the inability to distinguish between data anomalies and operational anomalies, especially in noisy environments with high data volume and speed.
Innovation Solution
A cognitive system using Internet of Things (IoT) sensors and active learning to detect and diagnose energy consumption anomalies by training machine learning models with uncontaminated data and interacting with users to label new data points, distinguishing between data and operational anomalies through scenario-based simulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If unsupervised machine learning is used for anomaly detection, then automation is improved, but measurement precision deteriorates due to false alarms and inability to distinguish data anomalies from operational anomalies
Solution Approach 1:
The system implements a feedback loop where users can review and label anomaly detections. The labeled data is then fed back to retrain the machine learning model, improving its precision over time while maintaining automated operation. This resolves the contradiction by allowing the system to start with high automation and progressively improve measurement precision through iterative feedback.
Solution Approach 2:
The patent introduces an intermediary layer between automated detection and final anomaly classification. This intermediary involves user review and labeling of detected anomalies, which helps distinguish between data anomalies and operational anomalies. The intermediary process improves measurement precision without completely eliminating automation, as the system handles initial detection automatically while using human judgment for critical classification.
2Device complexity
If machine learning models are trained on contaminated data, then device complexity is reduced, but measurement precision deteriorates due to false alarms
Solution Approach 1:
The system performs preliminary data cleaning and validation before training machine learning models. By preprocessing the data to remove contaminants and anomalies beforehand, the system maintains relatively simple model architecture while achieving high measurement precision. The preliminary action of data purification prevents false alarms without requiring complex modeling approaches.
Solution Approach 2:
The system implements self-service mechanisms where the machine learning model automatically identifies and handles contaminated data through feature engineering and anomaly detection in the training process itself. This allows the system to maintain simplicity while improving precision, as the model learns to distinguish between valid training data and contaminated data without requiring extensive external preprocessing.
3Measurement precision
If scenario-based simulation is used to generate training data, then measurement precision is improved by providing uncontaminated data, but device complexity increases
Solution Approach 1:
The system creates synthetic copies of real-world energy consumption scenarios through simulation. These simulated copies provide uncontaminated training data that mirrors real operational conditions without the noise and errors present in actual sensor data. By copying realistic scenarios in a controlled simulation environment, the system achieves high measurement precision while managing complexity through virtual rather than physical experimentation.
Data Source
AI summary
Embodiments for detection of energy consumption anomalies in one or more energy consumption systems in a cloud computing environment by a processor. Energy consumption anomalies associated with a selected facility, a selected object, or a combination thereof may be detected and diagnosed according to feedback and simulation data generated from a scenario-based simulation operation.