Sensor Anomaly Detection via KDE Segmentation and KL Divergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the IoT sensor network, accurate and efficient anomaly detection is challenging due to the large number of devices and limited resource capabilities of IoT sensors, which makes it difficult to identify faulty sensors and filter spurious reports in real-time.
Innovation Solution
The use of Kernel Density Estimation (KDE) in combination with Kullback-Leibler (KL) divergence to determine a threshold value for anomaly detection, and the calculation of a minimum window size to optimize detection performance, by dividing sensor data into segments based on cyclic or non-cyclic patterns and converting them into curves for divergence analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional anomaly detection methods are used in IoT sensor networks, then the system can operate with simple algorithms, but the detection accuracy deteriorates due to the large number of devices and limited resource capabilities
Solution Approach 1:
The patent segments the sensor data time window into multiple segments and applies KDE to each segment separately. This segmentation approach allows the system to handle large volumes of data from millions of sensors by processing smaller chunks, thereby maintaining detection accuracy without overwhelming the limited computational resources of IoT devices
Solution Approach 2:
The patent transforms the anomaly detection problem from direct statistical comparison to kernel density estimation in a transformed parameter space. By converting data points into density values through KDE and then applying KL divergence, the system achieves higher detection accuracy while the computational complexity remains manageable through parameter transformation rather than complex algorithmic structures
2Quantity of substance
If the number of connected devices increases to millions or billions, then the IoT network coverage expands, but the difficulty of accurate and efficient anomaly detection increases
Solution Approach 1:
The patent implements a self-service anomaly detection mechanism where each sensor's data is independently analyzed using KDE and KL divergence against its own historical baseline. This eliminates the need for centralized processing of all sensor data, allowing the system to scale to millions of devices without proportionally increasing detection difficulty, as each device essentially detects its own anomalies using the same standardized approach
Solution Approach 2:
The transformation to kernel density estimation creates a standardized parameter space that simplifies comparison across diverse sensor types and scales. By converting raw sensor readings into density values and then measuring divergence, the system handles heterogeneous data from billions of devices uniformly, reducing the overall difficulty of anomaly detection across the entire network
3Speed
If real-time anomaly detection is implemented, then the system responds quickly to sensor failures, but the computational resources required increase
Solution Approach 1:
The patent divides the computation into segments: historical data is processed once to establish baselines using KDE, then real-time detection only requires computing KL divergence between current segments and the pre-computed baseline. This segmentation enables real-time detection speed while minimizing ongoing energy consumption, as the computationally intensive KDE is performed only during baseline establishment
Solution Approach 2:
The system performs preliminary KDE computation on historical clean sensor data to establish baseline density distributions before real-time detection begins. This preliminary action stores the computationally expensive transformations in advance, so that real-time anomaly detection only requires simpler KL divergence calculations, achieving both real-time speed and energy efficiency
Data Source
AI summary
Embodiments determine anomalies in sensor data generated by a sensor. Embodiments receive a first time window of clean sensor data generated by the sensor, the clean sensor data including anomaly free data, and determine if the clean sensor data includes a cyclic pattern. When the clean sensor data has a cyclic pattern, embodiments divide the first time window into a plurality of segments of equal length, where each equal length includes the cyclic pattern. Embodiments convert the first time window and each of the plurality of segments into corresponding curves using Kernel Density Estimation (“KDE”). For each of the plurality of segments, embodiments determine a Kullback-Leibler (“KL”) divergence value between corresponding curves of the segment and the first time window to generate a plurality of KL divergence values.