CN106022368A - Incremental track anomaly detection method based on incremental kernel principle component analysis - Google Patents
Incremental track anomaly detection method based on incremental kernel principle component analysis Download PDFInfo
- Publication number
- CN106022368A CN106022368A CN201610325491.6A CN201610325491A CN106022368A CN 106022368 A CN106022368 A CN 106022368A CN 201610325491 A CN201610325491 A CN 201610325491A CN 106022368 A CN106022368 A CN 106022368A
- Authority
- CN
- China
- Prior art keywords
- track
- nucleus
- data
- newly
- nucleus lesion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an incremental track anomaly detection method based on incremental kernel principle component analysis, and belongs to the field of an incremental track anomaly detection method. The method comprises the following steps: to begin with, carrying out model initialization calculation, carrying out initial kernel feature space calculation through conventional Batch KPCA, and when M newly-increased track data comes, carrying out standardization on the M track data first; then, calculating kernel feature space of the newly-increased data through Batch KPCA; calculating average reconstruction error of the newly-increased data and training data, and if the error of the two is larger than a preset threshold value, using a follow-up kernel feature space division-merging method to update kernel feature space; then, carrying out projection on the updated kernel feature space and extracting a principal component; and finally, carrying out unsupervised learning and anomaly detection by utilizing a support vector machine. The advantages are that the method is superior to a conventional kernel principle component analysis method; computing complexity is reduced; and track anomaly detection efficiency is improved.
Description
Technical field
A kind of method that the present invention relates to increment track abnormality detection, a kind of increment based on increment core principle component analysis
The method of track abnormality detection.
Background technology
Track data comprises the various features such as geographical position coordinates, speed, direction, is considered as a kind of high dimensional data.Core master
Component analysis is a kind of nonlinear principal component analysis track method for detecting abnormality, by nonlinear mapping by track data from former
Beginning data space map, in high-dimensional feature space, then uses linear principal component analysis to carry out feature in high-dimensional feature space
Extract.But the computation complexity that core principle component analysis is when carrying out nuclear matrix feature decomposition is O (N3), has a strong impact on greatly
Application on scale data collection.Introduce incremental learning mode reducing time complexity is the key improving this type of method.
Summary of the invention
A kind of method that the invention aims to provide increment track abnormality detection based on increment core principle component analysis, solves
The problem that the computation complexity of existing core principle component analysis method is high.
The object of the present invention is achieved like this: the method:
First the initialization carrying out model calculates, and uses traditional Batch KPCA to carry out incipient nucleus feature space calculating, whenever
When having M bar to increase track data arrival newly, first this M bar track data is standardized;
Then Batch KPCA is used to calculate the nucleus lesion of newly-increased data;
Calculate newly-increased data and the average reconstruction error of training data respectively, if both errors are more than given threshold values, then perform
Follow-up nucleus lesion splits-merging method, updates nucleus lesion;
Then the nucleus lesion after updating is projected, extract principal component;
One-class support vector machine is finally utilized to carry out unsupervised learning and abnormality detection;
The method specifically comprises the following steps that
(1) it is somebody's turn to do increment track method for detecting abnormality based on increment core principle component analysis, it is necessary first to set the big of sliding window
Little P and the trace number M every time updated;P represents the size of the nucleus lesion every time needing renewal, and nucleus lesion is big
Little immobilize algorithm the term of execution;M represents the size of each increment;
(2) traditional Batch KPCA is then used to calculate incipient nucleus feature space model and the calculating of slip data window
Its average reconstruction errorThe newly-increased track data vector of Posterior circle batch processing;Processing newly-increased track data
During vector, first construct its nucleus lesion model, calculate its average reconstruction error
(3) the average reconstruction error ratio ε between itself and sliding window nucleus lesion is then calculatedratio;Work as εratioIt is higher than
During given threshold values v, use nucleus lesion framing shadowgraphy to update sliding window nucleus lesion, first use core feature empty
Between dividing method from slip data window, remove M bar track data characteristic vector the earliest, reduce nucleus lesion;Then
Use nucleus lesion to merge method to be merged in sliding window nucleus lesion by newly-increased M bar track data vector;
(4) calculate the projection of sliding window nucleus lesion simultaneously, ask for main constituent and calculate its eigenspace projection;
(5) finally use one-class support vector machine that the principal component extracted carries out unsupervised learning and abnormality detection, record inspection
The abnormal track measured;After having detected, need to recalculate the average reconstruction error of sliding window nucleus lesion
To process and increase track data newly next time.
Beneficial effect, owing to have employed such scheme, should increment track method for detecting abnormality based on increment core principle component analysis
Nucleus lesion framing shadowgraphy is used to update nucleus lesion data model.Maintain the slip number of a fixed size
According to window, whenever having M bar to increase track arrival newly, first from slip data window nucleus lesion model, remove M the earliest
Bar track data, more newly-increased M bar track data is merged in nucleus lesion;The core having only to calculate M bar track is special
Levy space, on the basis of original slip data window nucleus lesion, incrementally update nucleus lesion, it is to avoid the most more
The deficiency of nucleus lesion will be recalculated time new, solve high the asking of computation complexity of existing core principle component analysis method
Topic, has reached the purpose of the present invention.
Advantage: the method is better than traditional core principle component analysis method, reduces computational complexity, improves the abnormal inspection of track
The efficiency surveyed.
Accompanying drawing explanation
Fig. 1 is the flow chart of present invention increment based on increment core principle component analysis track method for detecting abnormality framework.
Fig. 2 is the contrast effect figure of the present invention and increment track method for detecting abnormality based on tradition KPCA.
Fig. 3 is abnormality detection design sketch of the present invention.
Fig. 4 is abnormality detection design sketch of the present invention.
Detailed description of the invention
Below in conjunction with the accompanying drawings and embodiment the invention will be further described.
Embodiment 1: the initialization first carrying out model calculates, and uses traditional Batch KPCA to carry out incipient nucleus feature empty
Between calculate, whenever have M bar increase newly track data arrive time, first this M bar track data is standardized;Then use
Batch KPCA calculates the nucleus lesion of newly-increased data.Calculate newly-increased data and the average reconstruction error of training data respectively,
If both errors are more than given threshold values, then perform follow-up nucleus lesion and split-merging method, update nucleus lesion;
Then the nucleus lesion after updating is projected, extract principal component;One-class support vector machine is finally utilized to carry out without prison
Educational inspector practises and abnormality detection.
Shown in Figure 1, a kind of increment track method for detecting abnormality based on increment core principle component analysis, comprise the following steps:
(1) initial track data, track standardization;
(2) determine the slip data window of a fixed size, calculate incipient nucleus feature space and reconstruction error;
(3) newly-increased average reconstruction error ratio between data and sliding window is calculated, if both errors are more than given valve
Value, then perform nucleus lesion and split-merging method, update nucleus lesion;
(4) calculate sliding window nucleus lesion projection after renewal, extract principal component;
(5) one-class support vector machine is utilized to carry out unsupervised learning and abnormality detection;
Concrete grammar is as follows:
It is somebody's turn to do increment track method for detecting abnormality based on increment core principle component analysis, every initially with Min-max methodological standardization
Bar track, size P setting sliding window and the trace number M every time updated.P represents the core feature every time needing to update
The size in space, nucleus lesion size immobilizes algorithm the term of execution;M represents the size of each increment;
Then traditional Batch KPCA is used to calculate the incipient nucleus feature space model of slip data window and calculate it and put down
All reconstruction errorsThe input vector t of one n dimension is mapped as the vectorial φ (t) of a l dimension by kernel function;Rebuild by mistake
Difference ε is exactlyAnd its project in nucleus lesion between squared-distance, whereinIt it is the map vector after centralization
φ(t);The newly-increased track data vector of Posterior circle batch processing;When processing newly-increased track data vector, first construct its core
Feature space model, calculates its average reconstruction errorAverage reconstruction error ratio εratioIt it is newly-increased M bar track data
Average reconstruction error and slip data window in ratio between training data ensemble average reconstruction error.Specific formula for calculation is
Then the average reconstruction error ratio ε between itself and sliding window nucleus lesion is calculatedratio;Work as εratioHigher than given valve
During value v, use nucleus lesion framing shadowgraphy to update sliding window nucleus lesion, first use nucleus lesion segmentation
Method removes M bar track data characteristic vector the earliest from slip data window, reduces nucleus lesion, nucleus lesion
Dividing method i.e. based on original input space feature space dividing method, carries out coring to it and obtains being applicable to the main one-tenth of increment core
The nucleus lesion dividing method analyzed;Then nucleus lesion is used to merge method by newly-increased M bar track data vector
Being merged in sliding window nucleus lesion, nucleus lesion merges i.e. after nucleus lesion is split, from slip data window
Mouthful nucleus lesion is split and obtains the nucleus lesion model Ω that is made up of residual track=(U, Φx, α, Λ, N), newly-increased M
The nucleus lesion model that bar track is constituted is Θ=(V, Φy, β, Δ, M), merge the nucleus lesion after Ω and Θ is updated
Model Q=(W, Φz,τ,∏,P)。
Calculate the projection of sliding window nucleus lesion simultaneously, ask for main constituent and calculate its eigenspace projection;
Finally use one-class support vector machine that the principal component extracted is carried out unsupervised learning and abnormality detection, travel through track set,
For every track sample in data set, utilize decision function to judge abnormal track, the abnormal track that record detects, and set
Put corresponding track label;After having detected, need to recalculate the average reconstruction error of sliding window nucleus lesion
To process and increase track data newly next time.
Embodiment 2: the present invention and the comparison of increment track Outlier Detection Algorithm based on tradition KPCA (Batch KPCA);
In order to verify effectiveness of the invention, choose 7270 tracks between nineteen ninety to 2006 in Atlantic hurricane data
Point amounts to 221 tracks and verifies as experimental data set.Increment track based on tradition KPCA is different as can be seen from Figure 2
Often the time that performs of detection method is along with the increase of slip data window is in quickly increase.And based on nucleus lesion segmentation-conjunction
And execution time of increment track Outlier Detection Algorithm increase as well as the increase of slip data window, but amplitude is not
Calculate big.And in the case of identical slip data window size, increment track based on nucleus lesion segmentation-merging is different
Often detection algorithm is more less than the calculating time used by increment track method for detecting abnormality based on tradition KPCA, and along with slip
The increase of data window, this species diversity can be increasing.
Embodiment 3: the abnormality detection design sketch of the present invention;
Choose the inventive method the 4th iteration and testing result of the 6th iteration on Atlantic hurricane data set, such as Fig. 3
Shown in Fig. 4.Lines thicker in figure represent the abnormal track detected, thinner lines represent normal trace.From figure
It can be seen that abnormality detection effect is fine, the track of a lot of behaviors abnormality all is checked out out.And from twice iteration
Result is it can be seen that being continuously added along with newly-increased track, and some new abnormal tracks are detected, again due to core feature
Space is being constantly updated, and some previous abnormal tracks are left in the basket.
Claims (2)
1. a method for increment track abnormality detection based on increment core principle component analysis, is characterized in that: the method:
First the initialization carrying out model calculates, and uses traditional Batch KPCA to carry out incipient nucleus feature space calculating, whenever
When having M bar to increase track data arrival newly, first this M bar track data is standardized;
Then Batch KPCA is used to calculate the nucleus lesion of newly-increased data;
Calculate newly-increased data and the average reconstruction error of training data respectively, if both errors are more than given threshold values, then perform
Follow-up nucleus lesion splits-merging method, updates nucleus lesion;
Then the nucleus lesion after updating is projected, extract principal component;
One-class support vector machine is finally utilized to carry out unsupervised learning and abnormality detection.
The method of a kind of increment track abnormality detection based on increment core principle component analysis the most according to claim 1, its
Feature is: the method specifically comprises the following steps that
(1) it is somebody's turn to do increment track method for detecting abnormality based on increment core principle component analysis, it is necessary first to set the big of sliding window
Little P and the trace number M every time updated;P represents the size of the nucleus lesion every time needing renewal, and nucleus lesion is big
Little immobilize algorithm the term of execution;M represents the size of each increment;
(2) traditional Batch KPCA is then used to calculate incipient nucleus feature space model and the calculating of slip data window
Its average reconstruction errorThe newly-increased track data vector of Posterior circle batch processing;Processing newly-increased track data
During vector, first construct its nucleus lesion model, calculate its average reconstruction error
(3) the average reconstruction error ratio ε between itself and sliding window nucleus lesion is then calculatedratio;Work as εratioIt is higher than
During given threshold values v, use nucleus lesion framing shadowgraphy to update sliding window nucleus lesion, first use core feature empty
Between dividing method from slip data window, remove M bar track data characteristic vector the earliest, reduce nucleus lesion;Then
Use nucleus lesion to merge method to be merged in sliding window nucleus lesion by newly-increased M bar track data vector;
(4) calculate the projection of sliding window nucleus lesion simultaneously, ask for main constituent and calculate its eigenspace projection;
(5) finally use one-class support vector machine that the principal component extracted carries out unsupervised learning and abnormality detection, record inspection
The abnormal track measured;After having detected, need to recalculate the average reconstruction error of sliding window nucleus lesion
To process and increase track data newly next time.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610325491.6A CN106022368B (en) | 2016-05-17 | 2016-05-17 | A method of the increment track abnormality detection based on increment core principle component analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610325491.6A CN106022368B (en) | 2016-05-17 | 2016-05-17 | A method of the increment track abnormality detection based on increment core principle component analysis |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106022368A true CN106022368A (en) | 2016-10-12 |
CN106022368B CN106022368B (en) | 2019-04-05 |
Family
ID=57097096
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610325491.6A Active CN106022368B (en) | 2016-05-17 | 2016-05-17 | A method of the increment track abnormality detection based on increment core principle component analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106022368B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109165550A (en) * | 2018-07-13 | 2019-01-08 | 首都师范大学 | A kind of multi-modal operation track fast partition method based on unsupervised deep learning |
CN109726737A (en) * | 2018-11-27 | 2019-05-07 | 武汉极意网络科技有限公司 | Trajectory-based anomaly detection method and device |
WO2021254413A1 (en) * | 2020-06-19 | 2021-12-23 | 南京大学 | Isolation distribution kernel construction method and apparatus, and anomaly data detection method and apparatus |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103020643A (en) * | 2012-11-30 | 2013-04-03 | 武汉大学 | Classification method based on kernel feature extraction early prediction multivariate time series category |
CN104915671A (en) * | 2015-06-23 | 2015-09-16 | 中国矿业大学 | FGAK (Fast Global Alignment Kernels) based abnormal trajectory detection method |
-
2016
- 2016-05-17 CN CN201610325491.6A patent/CN106022368B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103020643A (en) * | 2012-11-30 | 2013-04-03 | 武汉大学 | Classification method based on kernel feature extraction early prediction multivariate time series category |
CN104915671A (en) * | 2015-06-23 | 2015-09-16 | 中国矿业大学 | FGAK (Fast Global Alignment Kernels) based abnormal trajectory detection method |
Non-Patent Citations (2)
Title |
---|
TAT-JUN CHIN AND DAVID SUTER: "Incremental Kernel Principal Component Analysis", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
鲍苏宁 等: "基于核主成分分析的异常轨迹检测方法", 《计算机应用技术》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109165550A (en) * | 2018-07-13 | 2019-01-08 | 首都师范大学 | A kind of multi-modal operation track fast partition method based on unsupervised deep learning |
CN109165550B (en) * | 2018-07-13 | 2021-06-29 | 首都师范大学 | Multi-modal surgery track rapid segmentation method based on unsupervised deep learning |
CN109726737A (en) * | 2018-11-27 | 2019-05-07 | 武汉极意网络科技有限公司 | Trajectory-based anomaly detection method and device |
CN109726737B (en) * | 2018-11-27 | 2020-11-10 | 武汉极意网络科技有限公司 | Track-based abnormal behavior detection method and device |
WO2021254413A1 (en) * | 2020-06-19 | 2021-12-23 | 南京大学 | Isolation distribution kernel construction method and apparatus, and anomaly data detection method and apparatus |
Also Published As
Publication number | Publication date |
---|---|
CN106022368B (en) | 2019-04-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111651937B (en) | Method for diagnosing faults of in-class self-adaptive bearing under variable working conditions | |
CN105738109B (en) | Bearing fault classification diagnosis method based on rarefaction representation and integrated study | |
WO2022141669A1 (en) | Bearing fault diagnosis method for dynamic joint distribution alignment network under variable working conditions | |
CN102592268B (en) | Method for segmenting foreground image | |
WO2017024691A1 (en) | Analogue circuit fault mode classification method | |
CN103810288B (en) | Method for carrying out community detection on heterogeneous social network on basis of clustering algorithm | |
CN106384092A (en) | Online low-rank abnormal video event detection method for monitoring scene | |
CN106022368A (en) | Incremental track anomaly detection method based on incremental kernel principle component analysis | |
CN106528417A (en) | Intelligent detection method and system of software defects | |
CN102708294B (en) | Self-adaptive parameter soft measuring method on basis of semi-supervised local linear regression | |
CN104899448A (en) | Adaptive compensation method for static localization scheme of ensemble Kalman filter | |
CN109657693B (en) | Classification method based on correlation entropy and transfer learning | |
Liu et al. | Kd-lightnet: A lightweight network based on knowledge distillation for industrial defect detection | |
CN113553245B (en) | Log anomaly detection method combining bidirectional slice GRU and gate control attention mechanism | |
Fu et al. | MCA-DTCN: A novel dual-task temporal convolutional network with multi-channel attention for first prediction time detection and remaining useful life prediction | |
JP2022140294A (en) | System and method for training neural network having auto-encoder architecture to recover missing data | |
CN107169299B (en) | Tracking method for formation target in decentralized maneuvering mode | |
CN107292265B (en) | A kind of target trajectory rapid extracting method based on motor-driven detection | |
Li et al. | Fault transfer diagnosis of rolling bearings across different devices via multi-domain information fusion and multi-kernel maximum mean discrepancy | |
CN115880505B (en) | Low-order fault intelligent identification method for target edge detection neural network | |
CN115620082B (en) | Model training method, head posture estimation method, electronic device, and storage medium | |
CN107480647B (en) | Method for detecting abnormal behaviors in real time based on inductive consistency abnormality detection | |
CN115859826A (en) | Integrated learning-based shield equipment fault fusion prediction method | |
CN115936003A (en) | Software function point duplicate checking method, device, equipment and medium based on neural network | |
CN114021469A (en) | Method for monitoring one-stage furnace process based on mixed sequence network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |