Adaptive Meter Tampering Detection via Personalized Sensor Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting the unauthorized removal or tampering of utility meters, such as using mechanical tilt switches, vibration sensors, and accelerometers, are prone to false positives due to noise and lack a globally effective threshold, making it difficult to accurately distinguish between meter removal and other conditions.
Innovation Solution
A method and system that construct a model of normal sensor values for each utility meter based on continuous raw sensor data, including external data like temperature and outage information, using automated machine learning to detect abnormal conditions indicative of removal or tampering, eliminating the need for a priori thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed threshold is applied to sensor data to detect meter removal, then the detection method is simple to implement, but it produces many false positives due to noise and varied field conditions
Solution Approach 1:
The system transitions from static fixed thresholds to dynamic adaptive thresholds that automatically adjust based on learned normal behavior patterns for each specific meter. The machine learning model continuously updates the threshold based on new sensor data, allowing the detection system to adapt to changing field conditions and meter-specific characteristics, thereby reducing false positives while maintaining implementation simplicity.
Solution Approach 2:
The system performs self-learning and self-adjustment by automatically analyzing sensor data to establish normal behavior patterns for each meter. The machine learning algorithm processes sensor readings and autonomously determines appropriate detection thresholds without requiring manual calibration or external intervention, enabling the system to improve its own detection accuracy over time while keeping the implementation straightforward.
2Adaptability or versatility
If a globally correct threshold is applied to detect all meter removals, then detection coverage is maximized, but false positives increase due to lack of adaptation to local conditions
Solution Approach 1:
The system replaces a single global threshold with individualized detection thresholds for each meter, tailored to that specific meter's normal behavior patterns and local field conditions. Each meter receives customized detection parameters based on its unique characteristics and environment, allowing the system to maintain broad detection coverage across diverse installations while significantly reducing false positives at each local location.
Solution Approach 2:
The system divides the detection problem into individual meter-specific detection tasks rather than using a unified global approach. Each meter is processed independently with its own learned normal behavior model and detection threshold, allowing the system to handle the diversity of field conditions at each location without compromising overall detection coverage or increasing false positives.
3Measurement precision
If vibration sensors and accelerometers are used to detect meter removal, then detection sensitivity is improved, but the signals become constant and require complex threshold analysis to distinguish removal from normal operation
Solution Approach 1:
The system performs preliminary learning of normal vibration patterns for each meter during an initialization phase before actual detection begins. The machine learning model analyzes sensor data and establishes baseline normal behavior characteristics in advance, creating a reference framework that simplifies subsequent detection. This preliminary action allows the system to maintain high detection sensitivity while avoiding complex real-time analysis by comparing current signals against pre-learned normal patterns.
Solution Approach 2:
The system continuously monitors sensor data and uses feedback from actual meter behavior to refine and update the normal pattern models. The machine learning algorithm processes ongoing sensor readings and adjusts the detection thresholds based on observed patterns, creating a dynamic feedback loop that simplifies signal analysis by continuously adapting to real-world conditions. This feedback mechanism maintains sensitivity while reducing the complexity of real-time decision-making.
Data Source
AI summary
Methods and system for detecting tampering of a meter. A continuous stream of raw sensor values can be received from one or more meters among a group of meters. In response to receiving the continuous stream of raw sensor values from the meter(s), a model of normal sensor values can be automatically constructed for each meter among the group of meters based on the raw sensor values obtained from the meter(s) and based on data obtained through an ongoing development of the meter(s) or through automated machine learning by the meter(s). The model of normal sensor values can be used to detect abnormal conditions with respect to the meter(s). The abnormal conditions detected with respect to the meter(s) are potentially indicative of a removal of the meter(s) or of an attempt to physically tamper with the meter(s).