CN-121997226-A - Time sequence data anomaly identification verification method and system for cognitive test

CN121997226ACN 121997226 ACN121997226 ACN 121997226ACN-121997226-A

Abstract

The application provides a time sequence data abnormality identification verification method and a time sequence data abnormality identification verification system for a cognitive test, which relate to the technical field of data processing and comprise the steps of collecting a time sequence data sample set; extracting time domain features and domain difference features from original data based on a time sequence data sample set, carrying out multi-level anomaly identification according to the time domain features and the domain difference features, inputting the obtained level anomaly detection results into a meta classifier to carry out anomaly comprehensive analysis and confidence judgment of each time point, outputting anomaly identification verification results, and carrying out verification result identification through anomaly confidence. According to the method and the device, the technical problem that in the prior art, because the standardized cognitive test depends on the macroscopic summary index, the tiny cognitive anomalies are covered up, and the confidence of the time sequence data anomaly identification result is affected is solved, and the accuracy and the reliability of the time sequence data anomaly identification are improved through multi-level anomaly characteristics and input element classifiers for anomaly analysis.

Inventors

WEI YUCHUN
ZHU JIJUN
PAN YINJIE

Assignees

南京华伟医疗设备有限公司

Dates

Publication Date: 20260508
Application Date: 20260119

Claims (10)

1. The time sequence data abnormality identification verification method for the cognitive test is characterized by comprising the following steps of: collecting a time sequence data sample set, including response time, response correctness, stimulation mode and test block index; Extracting time domain features and domain difference features from original data based on the time sequence data sample set, wherein the time domain features comprise a reaction time mean value and a standard deviation calculated by taking a sliding window as a unit, and the domain-specific features comprise a block index to which the window belongs and a target stimulation error-making ratio; Carrying out multi-level anomaly identification according to the time domain characteristics and the domain difference characteristics, wherein point anomaly detection, mode anomaly detection and context anomaly detection are carried out; And inputting the obtained detection results of the anomalies of each level into a meta classifier to perform comprehensive analysis and confidence judgment on the anomalies of each time point, outputting the verification results of anomaly identification, and performing verification result identification through the anomaly confidence.
2. The recognition verification method for abnormal time series data facing cognitive test according to claim 1, wherein extracting time domain features and domain difference features from original data based on the time series data sample set comprises: Cleaning the original data to remove invalid data caused by extreme values, missing data and equipment faults; sample data segmentation is carried out on the cleaned original data according to the size of a preset sliding window, and the average value and standard deviation of the reaction time in each sliding window are calculated to obtain time domain characteristics; Positioning the sequence and the dividing rule of the test blocks to which the data belong in the sliding window according to the test block indexes of the original data, and marking the block indexes corresponding to the data; Calculating an error ratio based on the test times and correctness of all target stimuli in each sliding window, and obtaining a target stimulus error ratio; wherein the domain-differentiation feature is determined by a block index of a sliding window and the target stimulus-error ratio.
3. The recognition verification method for time series data abnormality oriented to cognitive testing according to claim 1, wherein before performing multi-level abnormality recognition according to the time domain features and domain difference features, the method comprises: Establishing a multi-level abnormality identification space, wherein the multi-level abnormality identification space comprises a point abnormality detection module, a mode abnormality detection module and a context abnormality detection module; The point anomaly detection module is used for identifying anomalies of the response time and error rate data points; The mode abnormality detection module is used for identifying abnormal modes of gradual attenuation or fluctuation frequency of the reaction time and the error rate; The context abnormality detection module is used for analyzing the context information of the cognitive test and identifying behaviors which are not in accordance with task requirements.
4. The cognitive test-oriented time series data anomaly identification verification method of claim 3, wherein establishing a multi-level anomaly identification space comprises: Performing abnormal point feature learning based on the abnormal multi-scene sample set, performing abnormal recognition detection on each data point through a local outlier factor algorithm, constructing a point abnormal detection module, and recognizing abnormal data points deviating from a normal behavior mode through analyzing density differences of each data point relative to a neighborhood of each data point; training and reconstructing the characteristics of each data window through an LSTM self-encoder based on pattern characteristic learning of time sequence data, identifying anomalies in a reaction time or error rate change pattern, and constructing a pattern anomaly detection module; Combining the task background information and the context characteristics of the time sequence data, modeling the data through a transducer model, identifying a behavior mode which is inconsistent with the task requirements, and constructing a context abnormality detection module; And the point abnormality detection module, the mode abnormality detection module and the context abnormality detection module are connected in parallel to construct the multi-level abnormality recognition space.
5. The recognition verification method for time series data abnormality oriented to cognitive test according to claim 3, wherein the meta classifier adopts a weighted decision mechanism, learns weighting rules under different types of abnormalities and user cognition difference scenes through a training historical data set, performs weighted fusion on the input multi-level abnormality recognition results based on the weighting rules, and outputs final abnormality labels and confidence degrees of the final abnormality labels at each time point.
6. The cognitive test oriented temporal data anomaly recognition verification method of claim 5, wherein the meta classifier comprises: The attention weighting layer is used for receiving the preliminary results and contribution characteristics of the point abnormality detection module, the mode abnormality detection module and the context abnormality detection module, and dynamically calculating the importance weights of the results of each level in the current context through a lightweight attention network; The increment learning unit is used for interacting expert-labeled sample data to update increment, dynamically adjusting internal weighting rules and decision boundaries, and adapting to cognitive behavior differences of different users and concept drift generated along with time; And the interpretable output layer is used for carrying out abnormal attribution analysis based on the SHAP model, calculating the contribution degree of each input feature to the final abnormal judgment result, generating a readable attribution report, and identifying key features causing the abnormality and the corresponding abnormality type.
7. The recognition test-oriented time series data anomaly identification verification method of claim 6, further comprising, after outputting the anomaly identification verification result: and visually displaying the final anomaly labels, the anomaly confidence and the attribution reports, generating an interactive time axis anomaly spectrum, wherein anomaly events of different levels and types are identified by different colors and marks, and attribution analysis is inserted into the anomaly events.
8. The recognition verification method of time series data abnormality for cognitive testing according to claim 6, wherein the weighted fusion of the input multi-level abnormality recognition results based on the weighting rule, further comprises: establishing an abnormal cooperative relationship among the point abnormality detection module, the mode abnormality detection module and the context abnormality detection module; fusion verification constraint is carried out according to the abnormal cooperative relationship; And carrying out interactive verification on the abnormal detection results of each level based on the fusion verification constraint, and inputting the multi-level abnormal identification result passing the interactive verification into the meta classifier.
9. The cognitive test-oriented time series data anomaly identification verification method of claim 8, wherein establishing an anomaly synergistic relationship between the point anomaly detection module, the pattern anomaly detection module and the context anomaly detection module comprises: Based on historical time sequence data and expert knowledge, the point abnormality detection module, the mode abnormality detection module and the context abnormality detection module are used as nodes to perform node relation analysis, wherein the node relation analysis at least comprises a fusion relation, a dependency relation and a contradiction relation, the fusion relation is used for reflecting that different modules recognize abnormal results which are adjacent in time sequence and are mutually supported semantically, and the abnormal results jointly appear to be capable of jointly pointing to a composite abnormal event with higher confidence; And establishing a connection edge attribute between nodes based on the fusion relationship, the dependency relationship and the contradiction relationship, and constructing a cooperative relationship graph structure to reflect the abnormal cooperative relationship.
10. A cognition test oriented time series data anomaly identification verification system, characterized by the steps for implementing the cognition test oriented time series data anomaly identification verification method according to any one of claims 1 to 9, comprising: The data collection module is used for collecting a time sequence data sample set, including response time, response correctness, a stimulation mode and a test block index; The characteristic extraction module is used for extracting time domain characteristics and domain difference characteristics from original data based on the time sequence data sample set, wherein the time domain characteristics comprise a reaction time mean value and a standard deviation calculated by taking a sliding window as a unit, and the domain-specific characteristics comprise a block index to which the window belongs and a target stimulation error-making ratio; the multi-level anomaly identification module is used for carrying out multi-level anomaly identification according to the time domain characteristics and the domain difference characteristics, wherein point anomaly detection, mode anomaly detection and context anomaly detection are carried out; And the anomaly analysis module is used for inputting the obtained anomaly detection results of all levels into the meta classifier to perform anomaly comprehensive analysis and confidence judgment of all time points, outputting an anomaly identification verification result, and performing verification result identification through anomaly confidence.

Description

Time sequence data anomaly identification verification method and system for cognitive test Technical Field The application relates to the technical field of data processing, in particular to a time sequence data anomaly identification verification method and system for cognitive testing. Background Currently, the existing standardized cognitive test evaluates the cognitive processing capacity of a subject by recording the behavioral response of the subject to a specific stimulus, and generally relies on macroscopic summary indicators, such as average response time and overall accuracy, as a main analysis basis. However, the summary index compresses the fine granularity performance changing with time in the test process, masks the time sequence dynamic information of the test time level, and because the task context is completely stripped, the periodic fluctuation and the local abnormality possibly presented by the subject under different test blocks and different stimulus types are difficult to reflect, the identification accuracy of the fine abnormality is lower, so that the credibility and the interpretability of the cognitive test data are affected. In summary, in the prior art, the standardized cognitive test depends on the macroscopic summary index, so that the fine cognitive anomalies are covered, and the confidence of the time series data anomaly recognition result is affected. Disclosure of Invention The application aims to provide a time sequence data anomaly identification verification method and system for a cognitive test, which are used for solving the technical problem that in the prior art, because a standardized cognitive test depends on a macroscopic summary index, fine cognitive anomalies are covered, and the confidence of a time sequence data anomaly identification result is affected. In view of the above problems, the application provides a method and a system for identifying and verifying time sequence data anomalies for cognitive testing. The application provides a time sequence data anomaly identification verification method for a cognitive test, which is realized by a time sequence data anomaly identification verification system for the cognitive test, and comprises the steps of collecting a time sequence data sample set, extracting time domain features and field difference features from original data based on the time sequence data sample set, wherein the time domain features comprise a reaction time mean value and a standard deviation calculated by taking a sliding window as a unit, the field specificity features comprise a block index to which the window belongs and a target stimulation error rate, carrying out multi-level anomaly identification according to the time domain features and the field difference features, carrying out point anomaly detection, mode anomaly detection and context anomaly detection, inputting obtained anomaly detection results of all levels into a meta classifier, carrying out anomaly comprehensive analysis of all time points and confidence judgment, outputting anomaly identification verification results, and carrying out verification result identification through anomaly confidence. The method comprises the steps of cleaning original data, removing extreme values, missing data and invalid data caused by equipment faults, carrying out sample data segmentation on the cleaned original data according to the size of a preset sliding window, calculating the mean value and standard deviation of reaction time in each sliding window, obtaining time domain features, positioning the sequence of test blocks of the data in the sliding window and a division rule according to the test block index of the original data, marking the block index corresponding to the data, calculating error rate based on the test times and correctness of all target stimuli in each sliding window, and obtaining target stimulus error rate, wherein the field difference features are determined according to the block index of the sliding window and the target stimulus error rate. Optionally, a multi-level abnormality identification space is established, and the multi-level abnormality identification space comprises a point abnormality detection module, a mode abnormality detection module and a context abnormality detection module, wherein the point abnormality detection module is used for identifying abnormality of response time and error rate data points, the mode abnormality detection module is used for identifying an abnormality mode of gradual attenuation or fluctuation frequency of the response time and the error rate, and the context abnormality detection module is used for analyzing context information of a cognitive test and identifying behaviors which are inconsistent with task requirements. The method comprises the steps of selecting a plurality of data windows, carrying out abnormal point feature learning based on an abnormal multi-scene sample set, carrying out abnormal recognition d