Time Series Pattern Detection Using Derivative Angle Quantization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional approaches for detecting patterns in time series data, such as sudden changes in altitude for engines or machine components, are inefficient due to difficulties in constructing transition probabilities and lack timely determination, leading to potential system failures and increased costs.
Innovation Solution
A processor-implemented method that computes a first-order derivative and gradient of change in time series data, derives an angle of change, quantizes the data, and uses a weighted finite state transducers diagram (WFSTD) based on domain knowledge to detect anomalies by comparing occurrences within a specific time range against a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional amplitude-matching approaches are used to detect patterns in time series data, then the solution can identify patterns matching user-given amplitudes, but the approach fails to detect patterns with varying amplitudes and requires difficult construction of transition probabilities
Solution Approach 1:
The patent transforms the time series data by computing first-order derivatives to convert amplitude variations into directional changes (upward/downward). This parameter transformation allows the system to detect patterns based on direction sequences rather than amplitude matching, eliminating the need for complex transition probability construction while improving detection reliability for patterns with varying amplitudes
Solution Approach 2:
The patent creates a simplified representation of the original time series data by converting it into a sequence of directional symbols (upward/downward). This copied representation captures the essential pattern characteristics without retaining the complexity of original amplitude values, enabling efficient pattern matching without requiring complex transition probabilities
2Reliability
If traditional pattern detection methods are used, then the system can identify patterns, but timely determination is delayed leading to shutdown of machines or abrupt failures
Solution Approach 1:
The patent segments the continuous time series data into discrete time intervals and converts each interval into a symbolic representation (upward/downward direction). This segmentation enables real-time processing of pattern occurrences as they happen, allowing timely detection without waiting for complete data sets, thus preventing machine shutdowns and failures
Solution Approach 2:
The patent pre-defines the pattern of interest (e.g., specific sequences of upward and downward movements) before monitoring begins. By having the target pattern predetermined, the system can immediately detect when it occurs in the incoming data stream without requiring complex analysis or probability construction at detection time, enabling timely determination and preventing system failures
Data Source
AI summary
State-of-the-art approaches have concentrated on building solution(s) to match the amplitude of a time series with a user given one. However, these have failed to implement solution(s) which enables searching for pattern(s) that can depict human vision psychology. Embodiments of the present disclosure determine occurrence of pattern of interest in time series data for anomaly detection, wherein time series data is obtained, and first order derivative is computed. Further an angle of change in direction is derived based on a gradient of change in value of the time series data. This angle is further converted to a measurement unit. The time series data is quantized into bins and a weighted finite state transducers diagram (WFSTD) is obtained based on domain knowledge which is then converted to specific pattern. The specific pattern is searched in the bins to determine occurrence/count of the specific pattern for anomaly detection.