AMI Usage Anomaly Detection With LSTM-Based Meter Data Remediation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced metering infrastructure (AMI) systems face disruptions due to anomalies such as theft, cyber-attacks, meter malfunctions, and data corruption, leading to inaccurate utility usage data and billing, which existing monitoring platforms struggle to detect and remediate effectively.
Innovation Solution
A machine learning model, specifically a deep learning long short-term memory (LSTM) encoder, is trained on real and noise-added data to predict power usage values, identifying anomalies by comparing predicted and actual values, and ranking them for appropriate remediation actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional monitoring platforms are used to detect anomalies in utility usage data, then the system structure is simple, but the detection accuracy and ability to identify anomalies such as theft, cyber-attacks, and meter malfunctions is insufficient
Solution Approach 1:
The patent replaces traditional rule-based anomaly detection mechanisms with a machine learning model (LSTM encoder) that automatically learns patterns from historical data. This substitution enables the system to detect complex anomalies such as theft, cyber-attacks, and meter malfunctions with higher accuracy, while the model handles the complexity internally, presenting a simplified interface to users.
Solution Approach 2:
The patent introduces an intermediary layer consisting of the LSTM encoder model and anomaly detection engine that sits between the raw utility usage data and the final anomaly identification. This intermediary processes and interprets the data, transforming complex raw measurements into meaningful anomaly detections, thereby improving measurement precision without requiring users to directly manage system complexity.
2Measurement precision
If machine learning models are implemented to improve anomaly detection accuracy, then detection precision improves, but computational resources and processing time increase
Solution Approach 1:
The patent performs preliminary action by training the LSTM encoder model offline on historical utility usage data before deployment. This pre-training phase captures patterns and relationships in the data, allowing the model to make accurate anomaly detections during operation with minimal real-time computational resources. The heavy computational work is done in advance, reducing energy consumption during actual utility monitoring.
Solution Approach 2:
The patent applies partial action by using the machine learning model selectively - it processes historical data to establish baseline patterns, then uses these patterns to detect anomalies in real-time usage data. The model doesn't need to reprocess all historical data continuously, but rather applies learned patterns to current data streams, reducing computational overhead while maintaining detection accuracy.
3Reliability
If real-time monitoring of all utility data points is performed, then complete anomaly detection is achieved, but data processing time and system complexity increase
Solution Approach 1:
The patent extracts key features and patterns from historical utility usage data during the training phase, storing them in the trained LSTM model. During real-time operation, only the essential current usage data needs to be processed against these pre-extracted patterns, rather than analyzing all historical data points. This extraction approach maintains detection completeness while significantly reducing real-time processing time.
Solution Approach 2:
The system performs preliminary analysis of historical data to establish normal usage patterns and anomaly characteristics before real-time monitoring begins. This preliminary action creates a reference framework that enables faster real-time detection, as the system only needs to compare current data against pre-established patterns rather than performing comprehensive analysis on every data point in real-time.
Data Source
AI summary
Techniques for detecting and remediating anomalous intervals in time-series data of a monitored device are disclosed. A system trains a machine learning model on a combination of real data obtained from a monitoring device and false data generated by adding noise to the real data. The model predicts operating values for the device at individual intervals of a time-series data set. The system identifies anomalies in the time-series data based on differences between the predicted values and the real values. If the difference between a predicted value generate by the machine learning model and the real value exceeds a threshold, the system identifies a particular data point, such as a meter reading, as anomalous. The system ranks anomalies to perform remediation operations based on the ranking.