IoT Sensor Anomaly Period Detection Using AutoEncoder Recovery Error
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems in IoT networks face challenges with manual systems' high manpower and time costs, and automatic systems struggle with feature extraction limitations and reliability due to the imbalance between normal and anomaly data, hindering accurate and timely detection.
Innovation Solution
Anomaly period detection technique integrating a statistical-based AI model with an AutoEncoder, utilizing statistical analysis and unsupervised learning to automate labeling and improve detection accuracy, reducing manpower and time costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual anomaly detection systems are used, then detection reliability can be maintained through human judgment, but manpower and time costs increase significantly
Solution Approach 1:
The system performs self-service by automatically detecting anomalies through the AutoEncoder model without requiring manual intervention. The neural network autonomously processes sensor data, identifies anomalies, and generates detection reports, eliminating the need for human operators to manually analyze data while maintaining reliable detection through continuous automated operation.
Solution Approach 2:
The patent replaces manual mechanical detection processes with an automated neural network-based system. The AutoEncoder model substitutes human judgment with computational algorithms that process data through encoding and decoding operations, achieving both time efficiency and reliable anomaly detection through mathematical modeling rather than manual analysis.
2Loss of time
If automatic anomaly detection systems are used, then time costs are reduced, but detection accuracy deteriorates due to limitations in feature extraction and reliability
Solution Approach 1:
The system changes the parameter of detection methodology by transitioning from traditional statistical methods to deep learning-based AutoEncoder models. This parameter change enables the system to automatically learn complex feature representations from sensor data, improving detection accuracy while maintaining automated operation and time efficiency through neural network computations.
Solution Approach 2:
The patent applies phase transitions in the context of data processing by transforming raw sensor data through multiple encoding and decoding phases in the AutoEncoder model. The system progresses through distinct computational phases (encoding, latent space transformation, decoding) to extract meaningful features and identify anomalies, thereby improving accuracy while maintaining automated processing.
3Device complexity
If statistical analysis methods are used for anomaly detection, then detection process is simple, but detection precision is insufficient due to inability to handle complex patterns
Solution Approach 1:
The system implements nesting by embedding multiple processing layers within the AutoEncoder architecture. The model contains nested encoding layers that progressively extract features from raw data, with each layer handling more complex patterns. This nested structure allows the system to maintain relatively simple overall architecture while achieving high detection precision through hierarchical feature extraction.
Solution Approach 2:
The patent applies dimensionality change by transforming sensor data through the AutoEncoder's encoding and decoding operations. The system maps data from the original input space to a compressed latent space and back, creating a transformed dimensional representation that captures complex patterns and relationships, thereby improving detection precision without significantly increasing process complexity.
4Speed
If AI models are used for anomaly detection, then detection speed improves, but reliability decreases due to data imbalance between normal and anomaly samples
Solution Approach 1:
The system performs preliminary action by pre-training the AutoEncoder model on normal data patterns before actual anomaly detection. The model learns normal behavior characteristics during the training phase, enabling it to quickly and reliably identify deviations during operation. This preliminary learning process establishes the foundation for both fast detection and high reliability by establishing what normal behavior looks like.
Solution Approach 2:
The patent implements feedback mechanisms where the system continuously monitors detection results and can retrain the AutoEncoder model with new data. The feedback loop allows the model to adapt to changing conditions and improve its reliability over time, while maintaining fast detection speeds through the efficient neural network architecture that processes data in real-time.
Data Source
AI summary
A server device is proposed. The device may support anomaly period detection. The server device may receive sensor data from a sensor and produce smoothing period data by performing preprocessing on the sensor data based on an anomaly period detection model corresponding to the sensor. The server device may also acquire recovery data by inputting the smoothing period data into the anomaly period detection model. The server device may further determine a normal period pattern or an anomaly period pattern by comparing an error calculated between the smoothing period data and the recovery data with a predetermined threshold.