Anomaly Detection Using Stacked LSTM Error Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional anomaly detection techniques in time-series data face challenges with univariate or multivariate data, requiring pre-specified time windows and extensive preprocessing, and fail to capture unpredictable patterns like abrupt vehicle braking or rapid acceleration/deceleration.
Innovation Solution
The implementation of a stacked long-short term memory (LSTM) neural network that learns to predict time-series data without a pre-specified time window, using error vectors and parameters like mu and sigma to detect anomalies by modeling normal behavior and identifying prediction errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional anomaly detection techniques use pre-specified time windows and statistical measures, then the detection process is simple, but the system performance degrades and extensive preprocessing is required
Solution Approach 1:
The patent replaces traditional mechanical/statistical methods (CUSUM, EWMA with pre-specified time windows) with a neural network-based prediction model that automatically learns temporal patterns. The neural network substitutes the manual time window selection mechanism with automated feature learning, eliminating the need for pre-specified time windows and extensive preprocessing while improving detection performance on univariate and multivariate time-series data
Solution Approach 2:
The patent changes the fundamental parameter from fixed pre-specified time windows to dynamic learned temporal patterns through neural network training. The system transforms the rigid statistical approach into a flexible adaptive model that automatically adjusts to the specific characteristics of the time-series data, improving both ease of implementation and detection performance
2Ease of operation
If traditional prediction models are used to detect anomalies, then the method is straightforward, but unpredictable patterns like abrupt braking or rapid acceleration/deceleration are not captured
Solution Approach 1:
The patent introduces dynamics by using a recurrent neural network that can adapt to changing patterns in time-series data. The model dynamically adjusts its predictions based on learned temporal dependencies, enabling it to capture unpredictable patterns such as abrupt braking or rapid acceleration/deceleration that static traditional prediction models cannot detect. The system maintains simplicity through automated learning while achieving high measurement precision
3Device complexity
If pre-specified time windows are used for temporal analysis, then the analysis process is controlled, but the requirement for extensive preprocessing increases
Solution Approach 1:
The patent implements self-service by enabling the neural network to automatically learn and determine the appropriate temporal analysis windows from the data itself. The model performs self-adjustment during training, eliminating the need for manual pre-specified time windows and reducing preprocessing requirements. The system maintains controlled analysis through the structured neural network architecture while automatically adapting to the specific characteristics of each time-series dataset
Data Source
Figure 1
Figure 2A~2B
Figure 2C
AI summary
An anomaly detection system and method is provided. The system comprising: a hardware processor; and a memory storing instructions to configure the hardware processor, wherein the hardware processor receives a first time-series data comprising a first set of points and a second time-series data comprising a second set of points, computes a first set of error vectors for each point of the first set, and a second set of error vectors for each point of the second set, each set of error vectors comprising one or more prediction errors; estimates parameters based on the first set of error vectors comprising; applies (or uses) the parameters on the second set of error vectors; and detects an anomaly in the second time-series data when the parameters are applied on the second set of error vectors.