CN108462708A - A kind of modeling of the behavior sequence based on HDP-HMM and detection method - Google Patents
A kind of modeling of the behavior sequence based on HDP-HMM and detection method Download PDFInfo
- Publication number
- CN108462708A CN108462708A CN201810218284.XA CN201810218284A CN108462708A CN 108462708 A CN108462708 A CN 108462708A CN 201810218284 A CN201810218284 A CN 201810218284A CN 108462708 A CN108462708 A CN 108462708A
- Authority
- CN
- China
- Prior art keywords
- hdp
- state
- hmm
- data
- model
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/02—Protocols based on web technology, e.g. hypertext transfer protocol [HTTP]
Landscapes
- Engineering & Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Security & Cryptography (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Biology (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Computing Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of modeling of behavior sequence based on HDP HMM and detection methods, the logarithm transition probability of use state track is as judgement index, it is simpler, and establish two models and codetermine testing result from positive and negative two angles, solve the problems, such as that index parameter is adjudicated in previous detection method chooses and calculate excessively complexity.Compared with prior art, the present invention also achieves better detection result making up HMM while model is defined with defect in parameter Estimation, and the average detected rate by modeling and detection method in the present invention known to experiment is 95.3%.
Description
Technical field
The present invention relates to Behavior Pattern Analysis field, the modeling and inspection of especially a kind of behavior sequence based on HDP-HMM
Survey method.
Background technology
HMM is used widely in many fields, and has good performance, but itself still has some shortcomings.HMM
It is either all very restricted in the definition of model structure or in the standard method of estimation of model parameter, for much existing
There are many incomplete places in the solution of real problem.For example, maximum Likelihood does not fully take into account model
Complexity, this just very likely results in the over-fitting of parameter.In addition, the model structure of conventional HMM must be previously determined, i.e. mould
The observation and hidden state of type need to predefine, this makes model lack adaptability.
For these deficiencies existing for HMM, Beal et al. is in the literature layering Di Li Cray processes (Hierarchical
Dirichlet Process, HDP) theory is applied to hidden Markov model, and thus proposes and a kind of extended based on HMM
Nonparametric Bayes model is layered Di Li Crays process-hidden Markov model (HDP-HMM), also referred to as unlimited hidden Ma Erke
Husband's model (infinite Hidden Markov Model, iHMM).Compared with conventional HMM, HDP-HMM is a kind of data-driven
Learning algorithm, it has the structure without pre- primary data, can automatic Iterative do well the advantages such as number, thus can more reflect data
Real features.HDP-HMM is a kind of nonparametric Bayes model, can with the status number in estimating timing data, so it
The numerous areas such as visual scene identification, abnormality detection have extraordinary application prospect.
Invention content
The invention aims to solve the deficiencies in the prior art, a kind of behavior sequence based on HDP-HMM is provided
The modeling of row and detection method.
In order to achieve the above objectives, the present invention is implemented according to following technical scheme：
A kind of modeling method of the behavior sequence based on HDP-HMM, models behavior sequence using HDP-HMM models
And go out model parameter using Beam Sampling algorithm automatic Iteratives, it is as follows：
S1, HDP-HMM models are established, choose the object sequence data of the HTTP request for using user orientation server as observed quantity
To describe the behavior of user；
S2, data are pre-processed, and the object sequence data in step S1 is divided into training data and test number
According to；
S3, by the set expression of all HTTP request sequences in training data be y={ y1,...yT}；
S4, a priori value S according to a preliminary estimate is assigned to model parameterp, then use Beam Sampling methods pair
The parameter of HDP-HMM models is trained：Algorithm introduces auxiliary variable u, passes through judgementIt marks off a limited number of hidden
Tibetan state, the conditional probability for the transfer that then done well using Dynamic Programming iterative calculation, and all hidden state tracks are sampled out,
Assuming that its dependent variable is it is known that iteratively sample out auxiliary variable u, hidden state s, transfer matrix π, shared parameter β join with other
Number.
Specifically, the S4 is as follows：
S41, auxiliary variable u is sampled：
S42, hidden state track s is sampled：Because when providing auxiliary variable u, only for all t state trajectories s
MeetingWhen probability non-zero, and because cutting a property for rod construction, state trajectory is Finite Number, therefore using dynamic
State is planned to sample state, using following equation, is sampled to state trajectory s using Dynamic Programming：
S43, parameter π, φ, β are sampled：N is defined firstijFor in state trajectory s, the transfer of mono- steps of state i is state j's
Number, i, j ∈ { 1 ..., K }, K are the known state number in state trajectory s.
In addition, the present invention also provides a kind of detection method of the behavior sequence based on HDP-HMM, include the following steps：
Step 1: the object sequence data of the HTTP request of server opposite first are pre-processed, training data is built
Collection and test data set；
Step 2: determining judgement index：Since user behavior pattern has multiple renaturation, the property of regularity and contingency
The normal behaviour of matter, most users has same or analogous operation, and because state trajectory is a Markov chain, determines
The transition probability matrix of adjacent states in state trajectory is π, therefore the logarithmic mean transfer for defining determination state trajectory s is general
Rate isThen using | η | adjudicate index as behavior pattern；
Step 3: pre-defining an observation sequence length threshold value Ω0, when sequence of observations length reaches threshold value Ω0
When, calculate corresponding index parameter；
Step 4: two models HDP-HMM1 and HDP-HMM2 are established to the data object in step 1, wherein by HDP-
HMM1 is used to describe the normal behaviour pattern of validated user, more sensitive to abnormal behaviour pattern；And HDP-HMM2 is for describing
The abnormal behaviour pattern of invader or validated user, it is more sensitive to normal behaviour pattern, when being detected, input normal instruction
Practice data HDP-HMM1 model parameters are trained, input exception training data HDP-HMM2 model parameters are trained；
Step 5: optimum state track is sampled out by the model trained to two mode input detection datas respectively, by
The state transition probability of HDP-HMM1 models obtains state trajectory η1Average log transition probability, by HDP-HMM2 models
State transition probability obtains state trajectory η2Average log transition probability, then take absolute value respectively and and decision threshold value
ε1And ε2It is compared, works as satisfaction | η1| ＞ ε1And | η2|≤ε1When output abnormality, be otherwise considered as normal.
Compared with prior art, beneficial effects of the present invention are：
1, behavior sequence modeling and detection method proposed by the present invention based on HDP-HMM, the logarithm of use state track
Transition probability is used as judgement index, simpler, and establishes two models and codetermine testing result, solution from positive and negative two angles
Index parameter is adjudicated in previous detection method of having determined chooses and calculate excessively complicated problem.
2, great effort need not be spent to analyse in depth data before modeling again, the model ginseng of estimated data collection
Number.
3, better detection result is also achieved making up HMM while model is defined with defect in parameter Estimation,
Average detected rate by modeling and detection method in the present invention known to experiment is 95.3%.
Description of the drawings
Fig. 1 is the modeling process figure of the behavior sequence based on HDP-HMM of the present invention.
Fig. 2 is the flow chart of the detection method of the behavior sequence based on HDP-HMM of the present invention.
Fig. 3 is two HDP-HMM models in specific embodiments of the present invention | η1| and | η2| value chart.
Specific implementation mode
With reference to specific embodiment, the invention will be further described, in the illustrative examples and explanation of the invention
For explaining the present invention, but it is not as a limitation of the invention.
As shown in Figure 1, a kind of modeling method of behavior sequence based on HDP-HMM of the present embodiment, uses HDP-HMM moulds
Type to behavior sequence model and goes out model parameter using Beam Sampling algorithm automatic Iteratives, is as follows：
S1, HDP-HMM models are established, choose the object sequence data of the HTTP request for using user orientation server as observed quantity
To describe the behavior of user；
S2, data are pre-processed, and the object sequence data in step S1 is divided into training data and test number
According to；
S3, by the set expression of all HTTP request sequences in training data be y={ y1,...yT}；
S4, a priori value S according to a preliminary estimate is assigned to model parameterp, then use Beam Sampling methods pair
The parameter of HDP-HMM models is trained：Algorithm introduces auxiliary variable u, passes through judgementIt marks off a limited number of hidden
Tibetan state, the conditional probability for the transfer that then done well using Dynamic Programming iterative calculation, and all hidden state tracks are sampled out,
Assuming that its dependent variable is it is known that iteratively sample out auxiliary variable u, hidden state s, transfer matrix π, shared parameter β join with other
Number；
S41, auxiliary variable u is sampled：
S42, hidden state track s is sampled：Because when providing auxiliary variable u, only for all t state trajectories s
MeetingWhen probability non-zero, and because cutting a property for rod construction, state trajectory is Finite Number, therefore using dynamic
State is planned to sample state, using following equation, is sampled to state trajectory s using Dynamic Programming：
S43, parameter π, φ, β are sampled：N is defined firstijFor in state trajectory s, the transfer of mono- steps of state i is state j's
Number, i, j ∈ { 1 ..., K }, K are the known state number in state trajectory s.
In addition, as described in Figure 2, the present embodiment also provides a kind of detection method of the behavior sequence based on HDP-HMM, including
Following steps：
Step 1: the object sequence data of the HTTP request of server opposite first are pre-processed, training data is built
Collection and test data set；
Step 2: determining judgement index：Since user behavior pattern has multiple renaturation, the property of regularity and contingency
The normal behaviour of matter, most users has same or analogous operation, and because state trajectory is a Markov chain, determines
The transition probability matrix of adjacent states in state trajectory is π, therefore the logarithmic mean transfer for defining determination state trajectory s is general
Rate isThen using | η | adjudicate index as behavior pattern；
Step 3: pre-defining an observation sequence length threshold value Ω0, when sequence of observations length reaches threshold value Ω0
When, calculate corresponding index parameter；
Step 4: two models HDP-HMM1 and HDP-HMM2 are established to the data object in step 1, wherein by HDP-
HMM1 is used to describe the normal behaviour pattern of validated user, more sensitive to abnormal behaviour pattern；And HDP-HMM2 is for describing
The abnormal behaviour pattern of invader or validated user, it is more sensitive to normal behaviour pattern, when being detected, input normal instruction
Practice data HDP-HMM1 model parameters are trained, input exception training data HDP-HMM2 model parameters are trained；
Step 5: optimum state track is sampled out by the model trained to two mode input detection datas respectively, by
The state transition probability of HDP-HMM1 models obtains state trajectory η1Average log transition probability, by HDP-HMM2 models
State transition probability obtains state trajectory η2Average log transition probability, then take absolute value respectively and and decision threshold value
ε1And ε2It is compared, works as satisfaction | η1| ＞ ε1And | η2|≤ε1When output abnormality, be otherwise considered as normal.
Emulation experiment
Experimental data
The data set for testing use herein comes from NASA-HTTP (ftp://ita.ee.lb.gov/trace/NASA_
Access_log_Aug15.gz) data set.4 to 31 users of August in 2015 are contained in data set to Florida State
All HTTP request datas of NASA Kennedy's space server on base.Which part data are randomly selected herein, by processing
Obtain two datasets, including positive sample collection, i.e. normal request data about 5500 and negative sample collection, i.e. exception request data
2300, it is denoted as DS1 and DS2 respectively, is time series data.
Experimental setup
DataSet1 and DataSet2 are roughly divided into 70% training data and 30% test data respectively.Using normal and
Abnormal training dataset is respectively trained HDP-HMM1 and HDP-HMM2.It is divided into two parts in model measurement, one
Normal and abnormality test data set is detected respectively using model, is in addition to normal and abnormal data set blended data
Collection is tested.Model HDP-HMM1 and HDP-HMM2 are respectively detected same group of data during this exports η respectively1
And η2, then take absolute value respectively and with decision threshold value ε1And ε2It is compared, works as satisfaction | η1| ＞ ε1And | η2|≤ε1When it is defeated
Go out exception, is otherwise considered as normal.
The decision threshold value ε of two models1, ε2It is most important to model inspection performance quality, herein first in detection |
η1| and | η2| value counted, obtain ε1And ε2Preliminary value range, and then using multigroup experiment compare again final
Determine optimum decision threshold value.
Experimental result and analysis
In abnormality detection, omission factor (false positive rate, FP rate) and rate of false alarm (true
Negative rate, TN rate) it is two important indicators for weighing detecting system performance.The two in the present embodiment model refer to
Mark and two decision threshold value ε1And ε2It is related, it needs further to analyze the relationship between parameter to obtain a preferable judgement
Thresholding, to obtain preferable system performance.
To being trained respectively for HDP-HMM1 and HDP-HMM2 parameters, in training data final convergent status number namely
Normal behaviour number of modes is 32.Status number in abnormal data finally converges to 9.It can be intuitively by state transfer trajectory diagram
The status number for going out abnormal behaviour data is less, and state transfer track is more regular.
The selection of decision threshold threshold value has a significant impact to the performance of system.Preliminary decision threshold value in order to obtain, I
It is right using training data as mode input | η1| and | η2| value counted, as shown in figure 3, Fig. 3 can be seen that in the presence of a small amount of
Outlier, that is, there may be a small amount of abnormal behaviour in DataSet1.Equally, it is also likely to be present in DataSet2 few
Measure normal behaviour.It determines that decision threshold value will exclude these a small amount of outliers to ensure verification and measurement ratio, judgement can determine by figure
Threshold value ε1And ε2It takes respectively near 5.30 and 3.25.
Model wrong report number TN and missing inspection number FP is counted using test data as mode input, according to following public affairs
Formula calculates verification and measurement ratio, and wherein T and F are respectively positive and negative sample number.
Test data is divided into the proper testing data comprising 1000 datas, the abnormality test data of 600 datas with
And 900 normal and 100 exceptions three groups of test datas, it is detected respectively, and to ε1, ε2Different values are compared.Root
Data according to the experiment work as ε1And ε2Take detection result when near 5.25 and 3.30 preferable respectively, repeated detection calculates average detected rate and is
95.3%, reach preferable detection result.
Technical scheme of the present invention is not limited to the limitation of above-mentioned specific embodiment, every to do according to the technique and scheme of the present invention
The technology deformation gone out, each falls within protection scope of the present invention.
Claims (3)
1. a kind of modeling method of the behavior sequence based on HDP-HMM, which is characterized in that using HDP-HMM models to behavior sequence
Row model and go out model parameter using Beam Sampling algorithm automatic Iteratives, are as follows：
S1, HDP-HMM models are established, the object sequence data of the HTTP request of selection user orientation server are retouched as observed quantity
State the behavior of user；
S2, data are pre-processed, and the object sequence data in step S1 is divided into training data and test data；
S3, by the set expression of all HTTP request sequences in training data be y={ y1,...yT}；
S4, a priori value S according to a preliminary estimate is assigned to model parameterp, then use Beam Sampling methods to HDP-HMM
The parameter of model is trained：Algorithm introduces auxiliary variable u, passes through judgementA limited number of hidden state is marked off,
Then the conditional probability for the transfer that does well is iterated to calculate using Dynamic Programming, and samples out all hidden state tracks, it is assumed that its
Dependent variable is it is known that iteratively sample out auxiliary variable u, hidden state s, transfer matrix π, shared parameter β and other parameters.
2. the modeling method of the behavior sequence according to claim 1 based on HDP-HMM, it is characterised in that：The S4 tools
Steps are as follows for body：
S41, auxiliary variable u is sampled：
S42, hidden state track s is sampled：Because when providing auxiliary variable u, for all t state trajectories s only full
FootWhen probability non-zero, and because cut rod construction a property, state trajectory is Finite Number, thus using dynamic advise
It draws to be sampled to state, using following equation, state trajectory s is sampled using Dynamic Programming：
S43, parameter π, φ, β are sampled：N is defined firstijFor time that in state trajectory s, the transfer of mono- steps of state i is state j
Number, i, j ∈ { 1 ..., K }, K are the known state number in state trajectory s.
3. a kind of detection method of the behavior sequence based on HDP-HMM as claimed in claim 1 or 2, which is characterized in that including with
Lower step：
Step 1: the object sequence data of the HTTP request of server opposite first are pre-processed, structure training dataset and
Test data set；
Step 2: determining judgement index：It is more since user behavior pattern has multiple renaturation, the property of regularity and contingency
The normal behaviour of number user has same or analogous operation, and because state trajectory is a Markov chain, determines state
The transition probability matrix of adjacent states in track is π, therefore the logarithmic mean transition probability for defining determination state trajectory s isThen using | η | adjudicate index as behavior pattern；
Step 3: pre-defining an observation sequence length threshold value Ω0, when sequence of observations length reaches threshold value Ω0When,
Calculate corresponding index parameter；
Step 4: two models HDP-HMM1 and HDP-HMM2 are established to the data object in step 1, wherein by HDP-HMM1
Normal behaviour pattern for describing validated user, it is more sensitive to abnormal behaviour pattern；And HDP-HMM2 is invaded for describing
The abnormal behaviour pattern of person or validated user, it is more sensitive to normal behaviour pattern, when being detected, input normal training number
HDP-HMM2 model parameters are trained according to being trained to HDP-HMM1 model parameters, inputting abnormal training data；
Step 5: optimum state track is sampled out by the model trained to two mode input detection datas respectively, by HDP-
The state transition probability of HMM1 models obtains state trajectory η1Average log transition probability, by the state of HDP-HMM2 models
Transition probability obtains state trajectory η2Average log transition probability, then take absolute value respectively and with decision threshold value ε1With
ε2It is compared, works as satisfaction | η1| ＞ ε1And | η2|≤ε1When output abnormality, be otherwise considered as normal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810218284.XA CN108462708B (en) | 2018-03-16 | 2018-03-16 | HDP-HMM-based behavior sequence detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810218284.XA CN108462708B (en) | 2018-03-16 | 2018-03-16 | HDP-HMM-based behavior sequence detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108462708A true CN108462708A (en) | 2018-08-28 |
| CN108462708B CN108462708B (en) | 2020-12-08 |
Family
ID=63236746
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810218284.XA Active CN108462708B (en) | 2018-03-16 | 2018-03-16 | HDP-HMM-based behavior sequence detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108462708B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111209942A (en) * | 2019-12-27 | 2020-05-29 | 广东省智能制造研究所 | Multi-mode sensing abnormity monitoring method for foot type robot |
| CN111222319A (en) * | 2019-11-14 | 2020-06-02 | 电子科技大学 | Document information extraction method based on novel HDP model |
| CN111353352A (en) * | 2018-12-24 | 2020-06-30 | 杭州海康威视数字技术股份有限公司 | Abnormal behavior detection method and device |
| CN111614695A (en) * | 2020-05-29 | 2020-09-01 | 华侨大学 | Network intrusion detection method and device of generalized inverse Dirichlet mixed HMM model |
| CN113642682A (en) * | 2021-10-14 | 2021-11-12 | 北京理工大学 | Trajectory primitive extraction and analysis method and system under multi-vehicle interaction environment |
| CN117079669A (en) * | 2023-10-17 | 2023-11-17 | 博上(山东)网络科技有限公司 | Feature vector extraction method for LSB audio steganography with low embedding rate |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101557506A (en) * | 2009-05-19 | 2009-10-14 | 浙江工业大学 | Intelligent detecting device for violent behavior in elevator car based on computer vision |
| US8010341B2 (en) * | 2007-09-13 | 2011-08-30 | Microsoft Corporation | Adding prototype information into probabilistic models |
| CN103473555A (en) * | 2013-08-26 | 2013-12-25 | 中国科学院自动化研究所 | Horrible video scene recognition method based on multi-view and multi-instance learning |
| CN103530603A (en) * | 2013-09-24 | 2014-01-22 | 杭州电子科技大学 | Video abnormality detection method based on causal loop diagram model |
| CN103678346A (en) * | 2012-09-07 | 2014-03-26 | 阿里巴巴集团控股有限公司 | Man-machine recognition method and system |
-
2018
- 2018-03-16 CN CN201810218284.XA patent/CN108462708B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8010341B2 (en) * | 2007-09-13 | 2011-08-30 | Microsoft Corporation | Adding prototype information into probabilistic models |
| CN101557506A (en) * | 2009-05-19 | 2009-10-14 | 浙江工业大学 | Intelligent detecting device for violent behavior in elevator car based on computer vision |
| CN103678346A (en) * | 2012-09-07 | 2014-03-26 | 阿里巴巴集团控股有限公司 | Man-machine recognition method and system |
| CN103473555A (en) * | 2013-08-26 | 2013-12-25 | 中国科学院自动化研究所 | Horrible video scene recognition method based on multi-view and multi-instance learning |
| CN103530603A (en) * | 2013-09-24 | 2014-01-22 | 杭州电子科技大学 | Video abnormality detection method based on causal loop diagram model |
Non-Patent Citations (2)
| Title |
|---|
| 岳猛: "基于环路LDA-HMM模型的视频异常事件检测", 《中国优秀硕士学位论文全文数据库（电子期刊），信息科技辑》 * |
| 崔永艳: "基于多示例学习的异常行为检测方法研究", 《中国优秀硕士学位论文全文数据库（电子期刊），信息科技辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353352A (en) * | 2018-12-24 | 2020-06-30 | 杭州海康威视数字技术股份有限公司 | Abnormal behavior detection method and device |
| CN111222319A (en) * | 2019-11-14 | 2020-06-02 | 电子科技大学 | Document information extraction method based on novel HDP model |
| CN111209942A (en) * | 2019-12-27 | 2020-05-29 | 广东省智能制造研究所 | Multi-mode sensing abnormity monitoring method for foot type robot |
| CN111209942B (en) * | 2019-12-27 | 2023-12-19 | 广东省智能制造研究所 | Multi-mode sensing abnormality monitoring method for foot robot |
| CN111614695A (en) * | 2020-05-29 | 2020-09-01 | 华侨大学 | Network intrusion detection method and device of generalized inverse Dirichlet mixed HMM model |
| CN111614695B (en) * | 2020-05-29 | 2022-04-22 | 华侨大学 | Network intrusion detection method and device of generalized inverse Dirichlet mixed HMM model |
| CN113642682A (en) * | 2021-10-14 | 2021-11-12 | 北京理工大学 | Trajectory primitive extraction and analysis method and system under multi-vehicle interaction environment |
| CN117079669A (en) * | 2023-10-17 | 2023-11-17 | 博上(山东)网络科技有限公司 | Feature vector extraction method for LSB audio steganography with low embedding rate |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108462708B (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108462708A (en) | A kind of modeling of the behavior sequence based on HDP-HMM and detection method | |
| Peel et al. | Detecting change points in the large-scale structure of evolving networks | |
| CN108960303B (en) | Unmanned aerial vehicle flight data anomaly detection method based on LSTM | |
| Li et al. | Interval-valued reliability analysis of multi-state systems | |
| CN109685314B (en) | Non-intrusive load decomposition method and system based on long-term and short-term memory network | |
| US20190034518A1 (en) | Target class feature model | |
| CN111767930A (en) | Method for detecting abnormal time series data of Internet of things and related equipment thereof | |
| CN109214349A (en) | A kind of object detecting method based on semantic segmentation enhancing | |
| CN111122162B (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
| CN108664690A (en) | Long-life electron device reliability lifetime estimation method under more stress based on depth belief network | |
| CN110757510A (en) | Method and system for predicting remaining life of robot | |
| CN112381171A (en) | Multi-sensor node missing data filling method based on combined model | |
| 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 | |
| CN113312696A (en) | Bridge health condition dynamic prediction alarm method based on ARIMA algorithm | |
| CN110489602B (en) | Knowledge point capability value estimation method, system, device and medium | |
| Blackledge et al. | A surface inspection machine vision system that includes fractal texture analysis | |
| CN111833375B (en) | Method and system for tracking animal group track | |
| CN116680639A (en) | Deep-learning-based anomaly detection method for sensor data of deep-sea submersible | |
| CN111241629A (en) | Intelligent prediction method for performance change trend of airplane hydraulic pump based on data driving | |
| Wu et al. | Monitoring reliability for three-parameter Frechet distribution using control charts | |
| CN113868597B (en) | Regression fairness measurement method for age estimation | |
| Vera et al. | Optimizing a dental milling process by means of Soft Computing Techniques | |
| CN117675230A (en) | Knowledge-graph-based oil well data integrity identification method | |
| CN111625525B (en) | Environment data repairing/filling method and system | |
| Kumar | Measuring Pearson's correlation coefficient of fuzzy numbers with different membership functions under weakest t-norm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |