Time Series Threshold Selection Using Anomaly Degree and Duration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
It is challenging to set an appropriate threshold value for detecting anomalous states in time series data due to the difficulty in balancing detection rate and false positive rate, especially when anomalous events are infrequent.
Innovation Solution
A time series data processing method that extracts a combination of parameters from normal period data to determine a threshold value, using a combination of 'anomaly degree' and 'duration' to identify a normal period maximum value and an anomalous period maximum value, and calculates a margin value to select an appropriate threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the threshold value is set to the lowest possible value to securely detect an anomalous state, then the detection rate is improved, but the false positive rate increases
Solution Approach 1:
The patent transitions from using a single threshold value to using a combination of multiple threshold values (first threshold for anomaly degree, second threshold for duration). This dimensional expansion allows the system to evaluate anomalies based on both magnitude and persistence, resolving the contradiction between detection sensitivity and false positive reduction.
Solution Approach 2:
The patent changes the parameters used for anomaly detection from a single threshold to multiple parameters (anomaly degree threshold and duration threshold). By requiring both conditions to be met simultaneously, the system achieves better balance between detecting true anomalies and avoiding false positives.
2Object-generated harmful factors
If the threshold value is set to the upper limit to avoid false positives, then the false positive rate is reduced, but the detection rate decreases
Solution Approach 1:
The patent adds a temporal dimension (duration) to the anomaly detection process. By requiring the anomaly to persist beyond a second threshold in addition to exceeding a first threshold, the system can filter out transient false positives while maintaining sensitivity to genuine anomalies that persist over time.
Solution Approach 2:
The patent introduces dynamic evaluation by considering the duration of anomaly states. Instead of a static single-threshold approach, the system dynamically evaluates both the magnitude and the temporal persistence of anomalies, allowing flexible detection that adapts to the nature of different anomaly patterns.
3Adaptability or versatility
If multiple candidates for threshold values are extracted from anomaly degree graphs, then more options are available, but it becomes difficult to determine which threshold value is appropriate
Solution Approach 1:
The patent transforms the threshold selection problem by introducing a second parameter (duration) in addition to anomaly degree. This parameter change creates a two-dimensional threshold space that naturally filters candidates, making it easier to select appropriate thresholds by evaluating both magnitude and temporal characteristics of anomalies.
Solution Approach 2:
The patent adds the duration dimension to the threshold evaluation process. By plotting and analyzing anomalies in a two-dimensional space (anomaly degree vs. duration), the system provides a more structured approach to threshold selection, reducing the complexity of choosing from multiple single-dimensional candidates.
Data Source
AI summary
A time series data processing apparatus according to the present invention includes an extracting unit configured to extract, from normal period time series data that is time series data of a period during which a measurement target is determined to be in a normal state of time series data including a plurality of parameters based on data measured from the measurement target, a combination of the plurality of parameters in which a value of another parameter with respect to a value of a predetermined parameter is maximum among combinations of the plurality of parameters, as a normal period maximum value.