CN111506627B - Target behavior clustering method and system - Google Patents
Target behavior clustering method and system Download PDFInfo
- Publication number
- CN111506627B CN111506627B CN202010317163.8A CN202010317163A CN111506627B CN 111506627 B CN111506627 B CN 111506627B CN 202010317163 A CN202010317163 A CN 202010317163A CN 111506627 B CN111506627 B CN 111506627B
- Authority
- CN
- China
- Prior art keywords
- vector
- triaxial
- value
- target
- order difference
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2462—Approximate or statistical queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Probability & Statistics with Applications (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Remote Sensing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a target behavior clustering method and a target behavior clustering system. The flow is as follows: A. extracting a track data packet to be clustered of a target; and respectively executing B-D for each track data packet: B. analyzing vehicle motion parameters of preset dimensions from the track data packet, and carrying out vectorization processing on the vehicle motion parameters of each dimension; C. constructing a feature set of the target behavior; D. constructing a target behavior entity description; E. constructing a similarity matrix according to the similarity between the descriptions of each target behavior entity; F. clustering the target behaviors according to the similarity matrix; G. and calculating the central point of each target behavior cluster and related statistics and storing the central point and the related statistics in an associated manner. According to the invention, the feature set of the target (single or group) behaviors is extracted to describe the behavior entity of each target, and the characteristics of the targets are combined, so that the behavior description is more targeted, the definition of the target behaviors is more definite, and the class cluster division is more accurate.
Description
Technical Field
The invention relates to the field of Internet of vehicles, in particular to a clustering method and a clustering system for behaviors of single or group users.
Background
For massive user track data of a time sequence, the user population behavior is abstracted from the massive user track data, and the massive user track data is applied to a real-time collision detection system and is required to be clustered. And aiming at the user/population behaviors based on the time sequence track data, the method has certain personalized characteristics, and particularly under a specific scene, great differences exist among the user/population behaviors of different clusters. The existing clustering method is mostly considered from the direction of generality, namely, all types of driving behavior data are subjected to clustering analysis by taking the same-view kernel as the same type of historical data, so that population behaviors with special scenes or personalized features are assimilated, the finally obtained clustering clusters are not obvious in individuation, and the lack of high-pertinence basis for the target behavior analysis of the special scenes is directly caused. The direct expression is a collision detection and frequent misjudgment for driving behaviors with special driving habits or different vehicle conditions from most vehicles.
Based on the background, the conventional distance measurement functions such as Euclidean distance and the like adopted by the conventional clustering method cannot achieve the optimal effect on the clustering of behaviors, and also cannot achieve an ideal result on the clustering of driving behaviors.
Disclosure of Invention
The invention aims at: aiming at the problems, a target behavior clustering method is provided. In order to make up for the problem that in the existing clustering method, no specific driving scene is considered, so that the clustering result is not strong in pertinence and the cluster classification is not accurate enough.
The technical scheme adopted by the invention is as follows:
a method of clustering target behavior, comprising the steps of:
A. extracting a track data packet to be clustered of a target;
and respectively executing B-D for each track data packet:
B. analyzing vehicle motion parameters of preset dimensions from the track data packet, and carrying out vectorization processing on the vehicle motion parameters of each dimension;
C. carrying out first preprocessing on the vectors of each dimension to respectively obtain the characteristics of each dimension, and constructing a characteristic set of the target behavior according to the characteristics of each dimension;
D. performing second preprocessing on each dimension parameter in the feature set to construct a target behavior entity description;
E. constructing a similarity matrix according to the similarity between the descriptions of each target behavior entity;
F. clustering the target behaviors according to the similarity matrix;
G. and calculating the central point and the related statistic of each target behavior cluster, and storing the central point and the related statistic of each target behavior cluster in a correlated way.
According to the clustering method, the feature set of the target (single or group) behaviors is extracted to describe the behavior entity of each target, and the characteristics of the targets are combined, so that the behavior description is more targeted, the definition of the target behaviors is more definite, and the classification of the class clusters is more accurate. For applications of occasions such as collision detection, personalized factors such as driving behaviors and vehicle conditions of targets are considered in detection probability, so that the basis in detection is more targeted, and false detection is not easy to occur. It should be noted that, for a single user, the parsed packet is the historical track packet of that user, which has a historical concept; for a user population, track data packets for a plurality of users in the same time period do not have a historical concept.
Further, the step B includes:
for targets of user groups, vehicle motion parameters analyzed from the track data packet comprise a speed, an acceleration triaxial and an angular velocity triaxial, and corresponding vectorization processing is carried out to obtain a speed vector, an acceleration triaxial vector and an angular velocity triaxial vector;
for the target of a single user, the vehicle motion parameters analyzed from the track data packet comprise speed, acceleration triaxial, angular velocity triaxial and alarm state, and the corresponding vectorization processing obtains a speed vector, acceleration triaxial, angular velocity triaxial and alarm vector.
Further, in the step C, the first preprocessing includes:
for a target of a user population, the first preprocessing includes:
calculating a non-zero speed average value and a first-order difference minimum value according to the speed vector;
correcting the acceleration triaxial respectively, and calculating the maximum value and the median of the first-order difference absolute value of each vector in the acceleration triaxial vector based on the corrected acceleration triaxial respectively;
respectively calculating the maximum value of the first-order difference absolute value and the median of the first-order difference absolute value of each vector in the triaxial vector of the angular velocity, and calculating the maximum value in the maximum value of the first-order difference absolute value of each triaxial vector of the angular velocity;
for the purpose of a single user, the first preprocessing procedure includes:
calculating a non-zero speed average value and a first-order difference minimum value according to the speed vector;
correcting the acceleration triaxial respectively, and calculating the maximum value of the first-order difference absolute value of each vector in the acceleration triaxial vector based on the corrected acceleration triaxial;
respectively calculating the maximum value of the first-order difference absolute value of each vector in the triaxial vector of the angular velocity, and calculating the maximum value of the first-order difference absolute value of each triaxial vector of the angular velocity;
the alarm vector is converted into subcode.
Further, in the step D, the second preprocessing includes:
for the purpose of the user population, the second preprocessing includes:
respectively carrying out sub-coding on the non-zero speed average value, the first-order difference absolute value median of each vector in the acceleration triaxial vector and the first-order difference absolute value median of each angular speed triaxial vector after binning, constructing a second average value based on the non-zero speed average value and the first-order difference minimum value of the speed vector, and respectively carrying out average value processing on the maximum value in the first-order difference absolute value maximum value of each angular speed triaxial vector and the first-order difference absolute value maximum value of each vector in the acceleration triaxial vector based on corresponding threshold values;
for the purpose of a single user, the second preprocessing procedure includes:
and respectively carrying out mean processing on the non-zero speed mean value of the speed vector, the maximum value of the first-order difference absolute value of each vector in the acceleration triaxial vector and the maximum value of the first-order difference absolute value of each angular speed triaxial vector based on the corresponding threshold value, and constructing a second mean value based on the non-zero speed mean value and the first-order difference minimum value of the speed vector.
Further, the step E specifically includes: and calculating the similarity distance between every two target behavior entity descriptions according to specific parameters in the target behavior entity descriptions, and constructing a similarity matrix according to each calculated similarity distance.
The above selected parameters are necessarily related to the characteristics of the target behavior habit. The similarity judgment is carried out according to the specific parameters, the judgment result is more accurate, and the influence of some irrelevant parameters on the comparison result can be effectively avoided while the calculation force resource is efficiently utilized.
Further, for the targets of the user population, the parameters involved in the similarity calculation include: the non-zero velocity mean value, the first-order difference absolute value median of each vector in the acceleration triaxial vector and the first-order difference absolute value median of each angular velocity triaxial vector are sub-coded vectors after binning;
for the purposes of a single user, parameters involved in the similarity calculation include: the non-zero velocity mean value, the maximum value of the first order difference absolute value maximum values of the triaxial vectors of each angular velocity, the mean value based on the corresponding threshold, and the second mean value, the alarm vector subcode.
Further, the method for calculating the similar distance comprises the following steps:
for the targets of the user population, the similarity distance calculation method comprises the following steps:
wherein O is act-m 、O act-n Respectively two compared target entity description subsets, wherein the target entity description subsets are composed of parameters participating in similarity calculation;
for the targets of a single user, the similarity distance calculation method is as follows:
wherein O is m 、O n Two compared target entity description subsets, respectively, which are composed of parameters involved in similarity calculation, VF avg For a non-zero velocity mean value, ΔVF is based on the mean value of the corresponding threshold min As the second mean value, ΔAF g For the maximum value of the first-order difference absolute value maximum values of the triaxial vectors of each angular velocity, cr is the subcode of the alarm vector based on the average value of the corresponding threshold values.
Compared with the common similarity comparison method such as Euclidean distance, the design of the similarity comparison method and the parameter of the similarity can obtain higher profile coefficients, so that the clustering effect is more outstanding.
Further, the second mean value calculating method includes:
wherein DeltaVFmin is the second mean, vavg is the non-zero velocity mean, deltaVmin is the velocity vector first order difference minimum.
Further, in the step G, the process of storing the center point of each target behavior cluster and the relevant statistics in an associated manner includes: according to the calculated central point and related statistics of each target behavior cluster, a corresponding target behavior cluster vector is constructed, and each target behavior cluster vector is written into a target behavior database to finish storage. The target behaviors are stored in the form of vectors, the correlation among related parameters is stronger, and meanwhile, the storage network is simpler, so that the data can be conveniently queried and extracted.
In order to solve all or part of the problems, the invention also provides a target behavior clustering system which runs the target behavior clustering method.
In summary, due to the adoption of the technical scheme, the beneficial effects of the invention are as follows:
1. the clustering method of the invention extracts the characteristic set of the target behavior pertinently, fully considers the personalized characteristics of the target, and is used for describing the entity of the population or single user behavior, so that the definition of the target behavior is more definite, and the described scene is more specific. Under a specific application scene, the clustering method has more referential property for judging driving behaviors.
2. Compared with the conventional Euclidean distance and other judging methods, the similarity judging parameters and the similarity judging method designed by the invention are more in line with driving behavior characteristics of various clusters, and behavior characteristic descriptions among the categories are more independent and clear.
Drawings
The invention will now be described by way of example and with reference to the accompanying drawings in which:
FIG. 1 is a flow chart of the overall method for clustering target behaviors.
FIG. 2 is a flow chart of a user population behavior clustering method.
FIG. 3 is a flowchart of a method for clustering historical driving behavior of a user.
Detailed Description
All of the features disclosed in this specification, or all of the steps in a method or process disclosed, may be combined in any combination, except for mutually exclusive features and/or steps.
Any feature disclosed in this specification (including any accompanying claims, abstract) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise. That is, each feature is one example only of a generic series of equivalent or similar features, unless expressly stated otherwise.
Example 1
The embodiment discloses a target behavior clustering method, as shown in fig. 1, comprising the following steps:
A. and extracting the track data packet to be clustered of the target.
For the targets of the user population, the track data packets to be clustered are track data packets of all users at the same time. For the targets of a single user, the track data packets to be clustered are all historical track data packets of the user.
And respectively executing B-D for each track data packet:
B. and analyzing the vehicle motion parameters of the preset dimensions from the track data packet, and carrying out vectorization processing on the vehicle motion parameters of each dimension.
For the targets of the user group, the vehicle motion parameters analyzed from the track data packet comprise a speed, an acceleration triaxial and an angular velocity triaxial, and the corresponding vectorization processing is carried out to obtain a speed vector, an acceleration triaxial vector and an angular velocity triaxial vector.
For the target of a single user, the vehicle motion parameters analyzed from the track data packet comprise speed, acceleration triaxial, angular velocity triaxial and alarm state, and the corresponding vectorization processing obtains a speed vector, acceleration triaxial, angular velocity triaxial and alarm vector.
C. And carrying out first preprocessing on the vectors of each dimension to respectively obtain the characteristics of each dimension, and constructing a characteristic set of the target behavior according to the characteristics of each dimension.
For a target of a user population, the first preprocessing includes:
calculating a non-zero speed average value and a first-order difference minimum value according to the speed vector;
correcting the acceleration triaxial respectively, and calculating the maximum value and the median of the first-order difference absolute value of each vector in the acceleration triaxial vector based on the corrected acceleration triaxial respectively;
and respectively calculating the maximum value of the first-order difference absolute value and the median of the first-order difference absolute value of each vector in the triaxial vectors of the angular velocity, and calculating the maximum value in the maximum value of the first-order difference absolute value of each triaxial vector of the angular velocity.
For the purpose of a single user, the first preprocessing procedure includes:
calculating a non-zero speed average value and a first-order difference minimum value according to the speed vector;
correcting the acceleration triaxial respectively, and calculating the maximum value of the first-order difference absolute value of each vector in the acceleration triaxial vector based on the corrected acceleration triaxial;
respectively calculating the maximum value of the first-order difference absolute value of each vector in the triaxial vector of the angular velocity, and calculating the maximum value of the first-order difference absolute value of each triaxial vector of the angular velocity;
the alarm vector is converted into subcode.
D. And carrying out second preprocessing on the parameters of each dimension in the feature set to construct the target behavior entity description.
For the purpose of the user population, the second preprocessing includes:
and respectively carrying out sub-coding on the non-zero speed average value, the first-order difference absolute value median of each vector in the acceleration triaxial vector and the first-order difference absolute value median of each angular speed triaxial vector after binning, constructing a second average value based on the non-zero speed average value and the first-order difference minimum value of the speed vector, and respectively carrying out average value processing on the maximum value of the first-order difference absolute value maximum value of each angular speed triaxial vector and the first-order difference absolute value maximum value of each vector in the acceleration triaxial vector based on corresponding threshold values.
For the purpose of a single user, the second preprocessing procedure includes:
and respectively carrying out mean processing on the non-zero speed mean value of the speed vector, the maximum value of the first-order difference absolute value of each vector in the acceleration triaxial vector and the maximum value of the first-order difference absolute value of each angular speed triaxial vector based on the corresponding threshold value, and constructing a second mean value based on the non-zero speed mean value and the first-order difference minimum value of the speed vector.
E. And constructing a similarity matrix according to the similarity between the descriptions of the target behavior entities.
And calculating the similarity distance between every two target behavior entity descriptions according to specific parameters in the target behavior entity descriptions, and constructing a similarity matrix according to each calculated similarity distance.
For the targets of the user population, the parameters involved in the similarity calculation include: the non-zero velocity mean value, the first order difference absolute value median of each vector in the acceleration triaxial vector and the first order difference absolute value median of each angular velocity triaxial vector are sub-coded vectors after binning.
For the purposes of a single user, parameters involved in the similarity calculation include: the non-zero velocity mean value, the maximum value of the first order difference absolute value maximum values of the triaxial vectors of each angular velocity, the mean value based on the corresponding threshold, and the second mean value, the alarm vector subcode.
F. And clustering the target behaviors according to the similarity matrix.
Clustering the target behaviors by using a kmeans++ algorithm according to the similarity matrix, and searching parameters through grids to obtain target behavior class clusters of a plurality of classes with optimal targets.
G. And calculating the central point and the related statistic of each target behavior cluster, and storing the central point and the related statistic of each target behavior cluster in a correlated way.
And calculating the central point and related statistics of each target behavior cluster, and constructing a corresponding target behavior cluster vector. And writing each target behavior cluster vector into a target behavior database for subsequent use.
Example two
Taking a user population target as an example, the embodiment discloses a process for constructing a target behavior entity description (i.e. the steps B-D) in the embodiment:
the user population behavior entity description construction process is described in detail below.
1. Track data packets with the same time uploaded by each equipment end are obtained, track segments with the time interval of S seconds and the length of N are respectively analyzed from each track data packet, and the speed, the acceleration triaxial and the angular speed triaxial in each point location data are extracted. A velocity vector V, an acceleration triaxial X, Y, Z, and an angular velocity triaxial H, T, K are formed.
2. Calculating the characteristics of the speed, acceleration and angular velocity dimensions respectively:
calculating the mean value V of non-0 velocity from the velocity vector V avg First order difference minimum DeltaV min 。
According to the triaxial acceleration coordinate system component correction method, X, Y, Z is corrected to obtain X x ,Y y ,Z z . Calculate X x First order difference X of (2) x ' obtain the absolute value of the first order difference |X x ' maximum value DeltaX is calculated x ＝max(|X x ' I). The same applies to find the maximum value delta Y of the first-order difference absolute value of the acceleration of the other two axes y 、ΔZ z . Further calculate the absolute value of the first order difference |X x Median Δx of' | x-median ＝median(|X x ' I), and similarly, find the median delta Y of the first-order difference absolute values of the accelerations of the other two axes y-median 、ΔZ z-median 。
The first-order differential absolute values (H ', (T ', (K ')) are similarly obtained from the angular velocity triaxial vector, the maximum values (delta H, delta T, delta K) of the first-order differential absolute values are obtained, and the maximum value (delta A) of the maximum values of the first-order differential absolute values of the angular velocity triaxial is further obtained g =max (Δh, Δt, Δk), and then the median Δa of the first order difference absolute value is obtained g-median ＝median(|H’|,|T’|,|K’|)。
Final composition of feature set F for constructing user population behavior act ＝[V avg ,ΔV min ，ΔX x ，ΔY y ，ΔZ z ,ΔX x-median ，ΔY y-median ,ΔZ z-median ，ΔA g ，ΔA g-median ]。
3. From feature set F act Constructing entity descriptions O of user population behaviors act ＝[V box ,ΔVF min ，ΔX box ，ΔY box ，ΔZ box ,ΔXF x ，ΔYF y ，ΔZF z ，ΔA box ，ΔAF g ]Wherein:
V box is the velocity mean value V avg Sub-coded vectors after binning. And (3) enabling the speed binning result to have N bins, wherein the sub-coding vector of the speed binning is a vector with the length of N and is initialized to 0. V (V) avg If the bin division result is M number bin, then V box Is marked 1 at the mth position of (c).
ΔX box ，ΔY box ，ΔZ box Respectively delta X x-median ，ΔY y-median ,ΔZ z-median Sub-coded vectors of the binned result.
ΔA box Is delta A g-median Sub-coded vectors of the binned result.
ΔAF g ＝ΔA g /A g0 ，A g0 Is a preset threshold.
ΔXF x＝ ΔX x /G 0 ，G 0 Is a preset threshold.
ΔYF y＝ ΔY y /G 0 ，G 0 Is a preset threshold.
ΔZF z＝ ΔZ z /G 0 ，G 0 Is preset toA threshold value.
Example III
Based on the second embodiment, the present embodiment discloses a process of constructing a similarity matrix by the targets of the user population according to the similarity between the descriptions of the target behavioral entities (i.e., the above step E):
for a user population as a target, the feature involved in similarity calculation is a subset of entity description vectors [ V box ，ΔX box ，ΔY box ，ΔZ box ，ΔA box ]Let the subset of the population entity description vectors of a track packet be O act-m ＝[V box-m ，ΔX box-m ，ΔY box-m ，ΔZ box-m ，ΔA box-m ]The population entity description subset of one track packet is O act-n ＝[V box-n ，ΔX box-n ，ΔY box-n ，ΔZ box-n ，ΔA box-n ]The distance function between them is:
finally, the population behavior similarity matrix of the N track data packets of all the users is obtained through calculation:
example IV
In this embodiment, taking the goal of a single user as an example, a process of constructing a description of a target behavior entity (i.e., the above steps B to D) is disclosed:
1. and analyzing track segments with the time interval of S seconds and the length of N from all the historical track data packets uploaded by the equipment end, and extracting the speed, the acceleration triaxial, the angular speed triaxial and the alarm state from each point location data. A velocity vector V, an acceleration triaxial vector X, Y, Z, an angular velocity triaxial vector H, T, K, and an alarm vector C are formed.
2. Calculating the characteristics of speed, acceleration, angular velocity and alarm dimension respectively:
calculating the mean value V of non-0 velocity from the velocity vector V avg First order difference minimum DeltaV min 。
According to the triaxial acceleration coordinate system component correction method, X, Y, Z is corrected to obtain X x ,Y y ,Z z . Calculate X x First order difference X of (2) x ' obtain the absolute value of the first order difference |X x ' maximum value DeltaX is calculated x ＝max(|X x ' I). The same applies to find the maximum value delta Y of the first-order difference absolute value of the acceleration of the other two axes y 、ΔZ z 。
The first-order differential absolute values (H ', (T ', (K ')) are similarly obtained from the angular velocity triaxial vector, the maximum values (delta H, delta T, delta K) of the first-order differential absolute values are obtained, and the maximum value (delta A) of the maximum values of the first-order differential absolute values of the angular velocity triaxial is further obtained g ＝max(ΔH，ΔT，ΔK)。
The alarm vector C is converted into subcode Cr.
Final composition of feature set F for constructing user driving behavior act ＝[V avg ,ΔV min ，ΔX x ，ΔY y ，ΔZ z ,ΔA g ，Cr]。
3. From feature set F act Building entity description O of user driving behavior act ＝[VF avg ,ΔVF min ，ΔXF x ，ΔYF y ，ΔZF z ,ΔAF g ，Cr]Wherein:
VF avg ＝V avg /V 0 ，V 0 is a preset threshold.
ΔAF g ＝ΔA g /A g0 ，A g0 Is a preset threshold.
ΔXF x＝ ΔX x /G 0 ，G 0 Is a preset threshold.
ΔYF y＝ ΔY y /G 0 ，G 0 Is a preset threshold.
ΔZF z＝ ΔZ z /G 0 ，G 0 Is a preset threshold.
Example five
The fourth embodiment discloses a process of constructing a similarity matrix by using the targets of the single user according to the similarity between the descriptions of the target behavioral entities (i.e. the above step E):
the feature involved in similarity calculation is the entity description vector subset VF avg ,ΔVF min ，ΔAF g ，Cr]Let entity description vector subset of a track data packet be O m ＝[VF avg-m ,ΔVF min-m ，ΔAF g-m ，Cr m ]The entity description vector subset of the other track data packet is O n ＝[VF avg-n ,ΔVF min-n ，ΔAF g-n ，Cr n ]The distance between them is:
finally, the historical behavior similarity matrix of the single user with N historical track data packets is obtained by calculation:
example seven
The present embodiment discloses a process of storing the center point of the target behavior class cluster and the relevant statistics in an associated manner (i.e., the step G above).
For the user population targets, based on the second embodiment, the center point and the related statistics of each population behavior cluster are calculated to form a binned population behavior cluster vector c= [ CID, count,<terminal，icount>,CV box ，ΔCX box ，ΔCY box ，ΔCZ box ，ΔCA box ，Q v ,Q x ，Q y ，Q z ，Q a ，QD v ，QD x ，QD y ，QD z ，QD a ]. Wherein:
CID is the group category identification.
count is the total number of the group track packets
< terminal, icount > is the binary group of N devices in the population < device number, the total number of track packets of the device >
CV box V in the description of all entities for group behavior belonging to the category box Bitwise and operation result of the vector.
ΔCX box DeltaX in all entity descriptions for group behaviors belonging to that class box Bitwise and operation result of the vector.
ΔCY box ΔY in all entity descriptions for population behavior belonging to that class box Bitwise and operation result of the vector.
ΔCZ box ΔZ in all entity descriptions for population behavior belonging to that class box Bitwise and operation result of the vector.
ΔCA box ΔA in all entity descriptions for population behavior belonging to that class box Bitwise and operation result of the vector.
Q v DeltaVF in all entity descriptions for population behavior belonging to that category min Default to 80% quantiles.
Q x ΔFX in all entity descriptions for group behaviors belonging to the category x Default to 80% quantiles.
Q y ΔFY in all entity descriptions for population behavior belonging to that class y Default to 80% quantiles.
Q z ΔFZ in all entity descriptions for population behavior belonging to that class z Default to 80% quantiles.
Q a ΔAF in all entity descriptions for population behavior belonging to that class g Default to 80% quantiles.
QD v For a population belonging to the categoryFor all entity descriptions ΔVF min By default, 75% quantiles-25% quantiles are taken.
QD x ΔFX in all entity descriptions for group behaviors belonging to the category x By default, 75% quantiles-25% quantiles are taken.
QD y ΔFY in all entity descriptions for historic behavior belonging to the category y By default, 75% quantiles to 25% quantiles are taken.
QD z ΔFZ in all entity descriptions for historic behavior belonging to the category z By default, 75% quantiles to 25% quantiles are taken.
QD a ΔAF in all entity descriptions for population behavior belonging to that class g By default, 75% quantiles-25% quantiles are taken.
And then writing the group behavior cluster vector C into a user group behavior database.
For the target of a single user, based on the fourth embodiment, calculating the center point and related statistics of each historical behavior cluster, and forming a warehouse-in historical behavior cluster vector C= [ Terminal, CID, CVF avg ,ΔCVF min ，ΔCAF g ，CCr，Q x ，Q y ，Q z ，QD x ，QD y ，QD z ]
Terminal is the device number.
CID is the device history behavior class identification.
CVF avg VF in a description of historical behavioral entities belonging to that category avg Is a mean value of (c).
ΔCVF min For DeltaVF in a historical behavioural entity description belonging to that category min Is a mean value of (c).
ΔCAF g ΔAF in the entity description for historic behaviors belonging to the category g Is a mean value of (c).
CCr is the result of bitwise AND of Cr in the historical behavioral entity description belonging to that category.
Q x For delta in historical behavioral entity descriptions belonging to that categoryFX x Default to 80% quantiles.
Q y For ΔFY in historical behavioral entity descriptions belonging to that category y Default to 80% quantiles.
Q z For ΔFZ in historical behavioral entity descriptions belonging to that category z Default to 80% quantiles.
QD x ΔFX in the description of historical behavioral entities belonging to that category x By default, 75% quantiles-25% quantiles are taken.
QD y For ΔFY in historical behavioral entity descriptions belonging to that category y By default, 75% quantiles to 25% quantiles are taken.
QD z For ΔFZ in historical behavioral entity descriptions belonging to that category z By default, 75% quantiles to 25% quantiles are taken.
And writing the historical behavior cluster vector C into a user historical behavior database.
Example eight
The embodiment discloses a user population behavior clustering method, as shown in fig. 2, comprising the following steps:
s001, extracting track data packets of all users at the same time from a historical database.
S002, analyzing a track segment with the time interval of S seconds and the length of N from each track data packet, and extracting the speed, the acceleration triaxial and the angular velocity triaxial of each point location data. A velocity vector V, an acceleration triaxial X, Y, Z, and an angular velocity triaxial H, T, K are formed.
And S003, calculating corresponding data characteristics according to each vector.
Calculating the mean value V of non-0 velocity from the velocity vector V avg First order difference minimum DeltaV min
According to the triaxial acceleration coordinate system component correction method, X, Y, Z is corrected to obtain X x ,Y y ,Z z . Calculate X x First order difference |X of (1) x ' I, obtaining the absolute value of the first order difference, and obtaining the maximum valueΔX x ＝max(|X x ' I). Delta Y is obtained by the same method y ，ΔZ z . Further calculate the absolute value of the first order difference |X x Median Δx of' | x-median ＝median(|X x ' I), and the same theory finds ΔY y-median ,ΔZ z-median 。
The first-order differential absolute values (H ', (T ', (K ')) are similarly obtained from the angular velocity triaxial vector, the maximum values (delta H, delta T, delta K) of the first-order differential absolute values are obtained, and the maximum value (delta A) of the angular velocity triaxial is further obtained g =max (Δh, Δt, Δk), and then the median Δa of the first order difference absolute value is obtained g-median ＝median(|H’|,|T’|,|K’|)。
Final composition of feature set F for constructing user population behavior act ＝[V avg ,ΔV min ，ΔX x ，ΔY y ，ΔZ z ,ΔX x-median ，ΔY y-median ,ΔZ z-median ，ΔA g ，ΔA g-median ]。
S004, from the feature set F act Constructing entity descriptions O of user population behaviors act ＝[V box ,ΔVF min ，ΔX box ，ΔY box ，ΔZ box ,ΔXF x ，ΔYF y ，ΔZF z ，ΔA box ，ΔAF g ]Wherein:
V box is the velocity mean value V avg Sub-coded vectors after binning. And (3) enabling the speed binning result to have N bins, wherein the sub-coding vector of the speed binning is a vector with the length of N and is initialized to 0. V (V) avg If the bin division result is M number bin, then V box Is marked 1 at the mth position of (c).
ΔVF min ＝(V avg -ΔV min )/V avg If V avg ＝0ΔVF min ＝-1。
ΔX box ，ΔY box ，ΔZ box Respectively delta X x-median ，ΔY y-median ,ΔZ z-median Sub-coded vectors of the binned result.
ΔA box Is delta A g-median Sub-coded vectors of the binned result.
ΔAF g ＝ΔA g /A g0 ，A g0 For a preset threshold, the default value is 180 0 /s。
ΔXF x＝ ΔX x /G 0 ，G 0 The default value is 980mg for the preset threshold.
ΔYF y＝ ΔY y /G 0 ，G 0 The default value is 980mg for the preset threshold.
ΔZF z＝ ΔZ z /G 0 ，G 0 The default value is 980mg for the preset threshold.
S005, calculating the driving behavior similarity between every two track data packets of all users according to the user-defined population behavior similarity function, and generating a user population behavior similarity matrix.
The feature involved in similarity calculation is a subset of entity description vectors [ V box ，ΔX box ，ΔY box ，ΔZ box ，ΔA box ]Let the subset of the population entity description vectors of a track packet be O act-m ＝[V box-m ，ΔX box-m ，ΔY box-m ，ΔZ box-m ，ΔA box-m ]The population entity description subset of one track packet is O act-n ＝[V box-n ，ΔX box-n ，ΔY box-n ，ΔZ box-n ，ΔA box-n ]。
The distance between them is:
finally, the population behavior similarity matrix of the N track data packets of all the users is obtained through calculation:
s006, clustering the population behaviors of the user by using a kmeans++ algorithm according to the population behavior similarity matrix, and searching parameters through grids to obtain the optimal M-class population behavior clusters of the user.
S007, calculating the center point and the related statistics of each group behavior cluster, forming a group behavior cluster vector c= [ CID, count,<terminal，icount>,CV box ，ΔCX box ，ΔCY box ，ΔCZ box ，ΔCA box ，Q v ,Q x ，Q y ，Q z ，Q a ，QD v ，QD x ，QD y ，QD z ，QD a ]。
CID is the group category identification.
count is the total number of the group track packets
< terminal, icount > is the binary group of N devices in the population < device number, the total number of track packets of the device >
CV box V in the description of all entities for group behavior belonging to the category box Bitwise and operation result of the vector.
ΔCX box DeltaX in all entity descriptions for group behaviors belonging to that class box Bitwise and operation result of the vector.
ΔCY box ΔY in all entity descriptions for population behavior belonging to that class box Bitwise and operation result of the vector.
ΔCZ box ΔZ in all entity descriptions for population behavior belonging to that class box Bitwise and operation result of the vector.
ΔCA box ΔA in all entity descriptions for population behavior belonging to that class box Bitwise and operation result of the vector.
Q v DeltaVF in all entity descriptions for population behavior belonging to that category min Default to 80% quantiles.
Q x Tracing all entities of the population behavior belonging to the categoryDeltaFX of the middle x Default to 80% quantiles.
Q y ΔFY in all entity descriptions for population behavior belonging to that class y Default to 80% quantiles.
Q z ΔFZ in all entity descriptions for population behavior belonging to that class z Default to 80% quantiles.
Q a ΔAF in all entity descriptions for population behavior belonging to that class g Default to 80% quantiles.
QD v DeltaVF in all entity descriptions for population behavior belonging to that category min By default, 75% quantiles-25% quantiles are taken.
QD x ΔFX in all entity descriptions for group behaviors belonging to the category x By default, 75% quantiles-25% quantiles are taken.
QD y ΔFY in all entity descriptions for historic behavior belonging to the category y By default, 75% quantiles to 25% quantiles are taken.
QD z ΔFZ in all entity descriptions for historic behavior belonging to the category z By default, 75% quantiles to 25% quantiles are taken.
QD a ΔAF in all entity descriptions for population behavior belonging to that class g By default, 75% quantiles-25% quantiles are taken.
And S008, writing the result of the S007 into a user population behavior database.
Example nine
The embodiment discloses a user driving behavior clustering method, as shown in fig. 3, comprising the following steps:
s001, extracting all track data packets of a single user from a historical database.
S002, analyzing a track section with the time interval of S seconds and the length of N from each track data packet, and extracting the speed, the acceleration triaxial, the angular velocity triaxial and the alarm state in each point location data. A velocity vector V, an acceleration triaxial vector X, Y, Z, an angular velocity triaxial vector H, T, K, and an alarm vector C are formed.
S003, calculating the data characteristics corresponding to each track data packet.
Calculating the mean value V of non-0 velocity from the velocity vector V avg First order difference minimum DeltaV min
According to the triaxial acceleration coordinate system component correction method, X, Y, Z is corrected to obtain X x ,Y y ,Z z . Calculate X x First order difference X of (2) x ' obtaining the maximum value DeltaX of the absolute value of the first order difference x ＝max(|X x ' I). By the same thing, ΔY y ，ΔZ z 。
The first order difference is obtained from the three axial vectors of the angular velocity, the maximum values DeltaH, deltaT, deltaK of the absolute values of the first order difference are obtained, and the maximum value DeltaA of the three axial vectors of the angular velocity is further obtained g ＝max(ΔH，ΔT，ΔK)。
The alarm vector C is converted into subcode Cr.
Final composition of feature set F for constructing user driving behavior act ＝[V avg ,ΔV min ，ΔX x ，ΔY y ，ΔZ z ,ΔA g ，Cr]。
S004, from the feature set F act Building an entity description O of the driving behavior of each track packet of the user history act ＝[VF avg ,ΔVF min ，ΔXF x ，ΔYF y ，ΔZF z ,ΔAF g ，Cr]Wherein:
VF avg ＝V avg /V 0 ，V 0 the default value is 15km/h for a preset threshold.
ΔVF min ＝(V avg -ΔV min )/V avg If V avg ＝0ΔVF min ＝-1。
ΔAF g ＝ΔA g /A g0 ，A g0 For a preset threshold, the default value is 180 0 /s。
ΔXF x＝ ΔX x /G 0 ，G 0 The default value is 980mg for the preset threshold.
ΔYF y＝ ΔY y /G 0 ，G 0 The default value is 980mg for the preset threshold.
ΔZF z＝ ΔZ z /G 0 ，G 0 The default value is 980mg for the preset threshold.
S005, calculating the driving behavior similarity between every two track data packets of the user history according to the self-defined driving behavior similarity function, and generating a user driving behavior similarity matrix.
The feature involved in similarity calculation is the entity description vector subset VF avg ,ΔVF min ，ΔAF g ，Cr]Let entity description vector subset of a track data packet be O m ＝[VF avg-m ,ΔVF min-m ，ΔAF g-m ，Cr m ]The entity description vector subset of the other track data packet is O n ＝[VF avg-n ,ΔVF min-n ，ΔAF g-n ，Cr n ]The distance between them is:
finally, the historical behavior similarity matrix of the single user with N historical track data packets is obtained by calculation:
s006, clustering the historical behaviors of the user by using a kmeans++ algorithm according to the historical behavior similarity matrix, and searching parameters through grids to obtain the optimal M types of historical behavior clusters of the user.
S007, calculating the central point and the related statistics of each historical behavior class cluster to form the historical behavior of the warehouse-inCluster-like vector c= [ Terminal, CID, CVF avg ,ΔCVF min ，ΔCAF g ，CCr，Q x ，Q y ，Q z ，QD x ，QD y ，QD z ]
Terminal is the device number.
CID is the device history behavior class identification.
CVF avg VF in a description of historical behavioral entities belonging to that category avg Is a mean value of (c).
ΔCVF min For DeltaVF in a historical behavioural entity description belonging to that category min Is a mean value of (c).
ΔCAF g ΔAF in the entity description for historic behaviors belonging to the category g Is a mean value of (c).
CCr is the result of bitwise AND of Cr in the historical behavioral entity description belonging to that category.
Q x ΔFX in the description of historical behavioral entities belonging to that category x Default to 80% quantiles.
Q y For ΔFY in historical behavioral entity descriptions belonging to that category y Default to 80% quantiles.
Q z For ΔFZ in historical behavioral entity descriptions belonging to that category z Default to 80% quantiles.
QD x ΔFX in the description of historical behavioral entities belonging to that category x By default, 75% quantiles-25% quantiles are taken.
QD y For ΔFY in historical behavioral entity descriptions belonging to that category y By default, 75% quantiles to 25% quantiles are taken.
QD z For ΔFZ in historical behavioral entity descriptions belonging to that category z By default, 75% quantiles to 25% quantiles are taken.
And S008, writing the result of the S007 into a user history behavior database.
Examples ten
The embodiment discloses a target behavior clustering system, which runs the clustering method in any of the embodiments.
The invention is not limited to the specific embodiments described above. The invention extends to any novel one, or any novel combination, of the features disclosed in this specification, as well as to any novel one, or any novel combination, of the steps of the method or process disclosed.
Claims (8)
1. A method for clustering target behaviors, comprising the steps of:
A. extracting a track data packet to be clustered of a target;
and B-D is respectively executed for each track data packet:
B. analyzing vehicle motion parameters of preset dimensions from the track data packet, and carrying out vectorization processing on the vehicle motion parameters of each dimension; comprising the following steps:
for targets of user groups, vehicle motion parameters analyzed from the track data packet comprise a speed, an acceleration triaxial and an angular velocity triaxial, and corresponding vectorization processing is carried out to obtain a speed vector, an acceleration triaxial vector and an angular velocity triaxial vector;
for the target of a single user, the vehicle motion parameters analyzed from the track data packet comprise speed, acceleration triaxial, angular velocity triaxial and alarm state, and the corresponding vectorization processing obtains speed vectors, acceleration triaxial vectors, angular velocity triaxial vectors and alarm vectors
C. Carrying out first preprocessing on the vectors of each dimension to respectively obtain the characteristics of each dimension, and constructing a characteristic set of the target behavior according to the characteristics of each dimension; the first preprocessing includes:
for a target of a user population, the first preprocessing includes:
calculating a non-zero speed average value and a first-order difference minimum value according to the speed vector;
correcting the acceleration triaxial respectively, and calculating the maximum value and the median of the first-order difference absolute value of each vector in the acceleration triaxial vector based on the corrected acceleration triaxial respectively;
respectively calculating the maximum value of the first-order difference absolute value and the median of the first-order difference absolute value of each vector in the triaxial vector of the angular velocity, and calculating the maximum value in the maximum value of the first-order difference absolute value of each triaxial vector of the angular velocity;
for the purpose of a single user, the first preprocessing procedure includes:
calculating a non-zero speed average value and a first-order difference minimum value according to the speed vector;
correcting the acceleration triaxial respectively, and calculating the maximum value of the first-order difference absolute value of each vector in the acceleration triaxial vector based on the corrected acceleration triaxial;
respectively calculating the maximum value of the first-order difference absolute value of each vector in the triaxial vector of the angular velocity, and calculating the maximum value of the first-order difference absolute value of each triaxial vector of the angular velocity;
converting the alarm vector into subcode;
D. performing second preprocessing on each dimension parameter in the feature set to construct a target behavior entity description;
E. constructing a similarity matrix according to the similarity between the descriptions of each target behavior entity;
F. clustering the target behaviors according to the similarity matrix;
G. and calculating the central point and the related statistic of each target behavior cluster, and storing the central point and the related statistic of each target behavior cluster in a correlated way.
2. The target behavior clustering method as set forth in claim 1, wherein in the step D, the second preprocessing includes:
for the purpose of the user population, the second preprocessing includes:
respectively carrying out sub-coding on the non-zero speed average value, the first-order difference absolute value median of each vector in the acceleration triaxial vector and the first-order difference absolute value median of each angular speed triaxial vector after binning, constructing a second average value based on the non-zero speed average value and the first-order difference minimum value of the speed vector, and respectively carrying out average value processing on the maximum value in the first-order difference absolute value maximum value of each angular speed triaxial vector and the first-order difference absolute value maximum value of each vector in the acceleration triaxial vector based on corresponding threshold values;
for the purpose of a single user, the second preprocessing procedure includes:
and respectively carrying out mean processing on the non-zero speed mean value of the speed vector, the maximum value of the first-order difference absolute value of each vector in the acceleration triaxial vector and the maximum value of the first-order difference absolute value of each angular speed triaxial vector based on the corresponding threshold value, and constructing a second mean value based on the non-zero speed mean value and the first-order difference minimum value of the speed vector.
3. The target behavior clustering method according to any one of claims 1-2, wherein the step E specifically includes: and calculating the similarity distance between every two target behavior entity descriptions according to specific parameters in the target behavior entity descriptions, and constructing a similarity matrix according to each calculated similarity distance.
4. The method of object behavior clustering as claimed in claim 3, wherein for the objects of the user population, the parameters involved in the similarity calculation include: the non-zero velocity mean value, the first-order difference absolute value median of each vector in the acceleration triaxial vector and the first-order difference absolute value median of each angular velocity triaxial vector are sub-coded vectors after binning;
for the purposes of a single user, parameters involved in the similarity calculation include: the non-zero velocity mean value, the maximum value of the first order difference absolute value maximum values of the triaxial vectors of each angular velocity, the mean value based on the corresponding threshold, and the second mean value, the alarm vector subcode.
5. The method for clustering target behaviors according to claim 4, wherein the method for calculating the similarity distance is as follows:
for the targets of the user population, the similarity distance calculation method comprises the following steps:
wherein O is act-m 、O act-n Respectively two compared target entity description subsets, wherein the target entity description subsets are composed of parameters participating in similarity calculation;
for the targets of a single user, the similarity distance calculation method is as follows:
wherein O is m 、O n Two compared target entity description subsets, respectively, which are composed of parameters involved in similarity calculation, VF avg For a non-zero velocity mean value, ΔVF is based on the mean value of the corresponding threshold min As the second mean value, ΔAF g For the maximum value of the first-order difference absolute value maximum values of the triaxial vectors of each angular velocity, cr is the subcode of the alarm vector based on the average value of the corresponding threshold values.
6. The method for clustering target behaviors according to any one of claims 2, 4 and 5, wherein the method for calculating the second average value is as follows:
wherein DeltaVFmin is the second mean, vavg is the non-zero velocity mean, deltaVmin is the velocity vector first order difference minimum.
7. The method of clustering target behaviors as set forth in claim 1, wherein in the step G, the process of storing the center point of each target behavior class cluster and the related statistics in association includes: according to the calculated central point and related statistics of each target behavior cluster, a corresponding target behavior cluster vector is constructed, and each target behavior cluster vector is written into a target behavior database to finish storage.
8. A target behavior clustering system, characterized in that it operates the target behavior clustering method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010317163.8A CN111506627B (en) | 2020-04-21 | 2020-04-21 | Target behavior clustering method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010317163.8A CN111506627B (en) | 2020-04-21 | 2020-04-21 | Target behavior clustering method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111506627A CN111506627A (en) | 2020-08-07 |
| CN111506627B true CN111506627B (en) | 2023-05-30 |
Family
ID=71864895
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010317163.8A Active CN111506627B (en) | 2020-04-21 | 2020-04-21 | Target behavior clustering method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111506627B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116562598B (en) * | 2023-07-07 | 2023-09-19 | 成都花娃网络科技有限公司 | Distribution scheduling method, device and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105681089A (en) * | 2016-01-26 | 2016-06-15 | 上海晶赞科技发展有限公司 | Network user behavior clustering method, device and terminal |
| CN109784970A (en) * | 2018-12-13 | 2019-05-21 | 交控科技股份有限公司 | It is a kind of to be ridden the service recommendation method and device of data based on AFC passenger |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006021641A (en) * | 2004-07-08 | 2006-01-26 | Nissan Motor Co Ltd | Vehicle periphery display device and its display method |
| CN103218906B (en) * | 2013-04-23 | 2016-04-13 | 中国科学院深圳先进技术研究院 | To fall data collection and analysis platform |
| CN103729550B (en) * | 2013-12-18 | 2016-08-17 | 河海大学 | Multiple-model integration Flood Forecasting Method based on propagation time cluster analysis |
| CN106778532B (en) * | 2016-11-28 | 2019-05-31 | 合肥工业大学 | Based on the driving posture feature classification method for removing differentiation size parameter |
| CN106815312A (en) * | 2016-12-21 | 2017-06-09 | 东软集团股份有限公司 | A kind of driver's evaluation method and device |
| CN107585164B (en) * | 2017-09-04 | 2019-11-22 | 交通运输部公路科学研究所 | A kind of method and device for the driver that classifies |
| CN108280415A (en) * | 2018-01-17 | 2018-07-13 | 武汉理工大学 | Driving behavior recognition methods based on intelligent mobile terminal |
| CN108470146B (en) * | 2018-02-11 | 2022-07-08 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Similar track identification method of classic track |
| CN109101567A (en) * | 2018-07-17 | 2018-12-28 | 杭州电子科技大学 | A kind of distributed text approximate KNN semantic search calculation method |
| CN110490264A (en) * | 2019-08-23 | 2019-11-22 | 中国民航大学 | Multidimensional distance cluster method for detecting abnormality and system based on time series |
| CN110825826A (en) * | 2019-11-07 | 2020-02-21 | 深圳大学 | Clustering calculation method, device, terminal and storage medium |
-
2020
- 2020-04-21 CN CN202010317163.8A patent/CN111506627B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105681089A (en) * | 2016-01-26 | 2016-06-15 | 上海晶赞科技发展有限公司 | Network user behavior clustering method, device and terminal |
| CN109784970A (en) * | 2018-12-13 | 2019-05-21 | 交控科技股份有限公司 | It is a kind of to be ridden the service recommendation method and device of data based on AFC passenger |
Non-Patent Citations (1)
| Title |
|---|
| 大陆游客境外旅游景区内时空行为模式研究――以香港海洋公园为例;黄潇婷;张晓珊;赵莹;;资源科学(第11期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111506627A (en) | 2020-08-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111211994A (en) | Network traffic classification method based on SOM and K-means fusion algorithm | |
| Guedalia et al. | An on-line agglomerative clustering method for nonstationary data | |
| CN104951764B (en) | Hot-short Activity recognition method based on secondary spectral clustering and HMM-RF mixed models | |
| Cassidy et al. | Calculating feature importance in data streams with concept drift using Online Random Forest | |
| CN111506627B (en) | Target behavior clustering method and system | |
| US20240070388A1 (en) | Lexical analyzer for a neuro-linguistic behavior recognition system | |
| CN104331523B (en) | A kind of question sentence search method based on conceptual object model | |
| CN109490838A (en) | A kind of Recognition Method of Radar Emitters of data base-oriented incompleteness | |
| CN102393910A (en) | Human behavior identification method based on non-negative matrix decomposition and hidden Markov model | |
| CN107729952A (en) | A kind of traffic flow classification method and device | |
| Diop et al. | Testing Parameter Change in General Integer‐Valued Time Series | |
| CN107153837A (en) | Depth combination K means and PSO clustering method | |
| Shafronenko et al. | Fuzzy clustering of distorted observations based on optimal expansion using partial distances | |
| CN111506692B (en) | Collision detection method based on target behaviors | |
| Ashok Kumar et al. | A transfer learning framework for traffic video using neuro-fuzzy approach | |
| CN106878073B (en) | Network multimedia business semisupervised classification method based on t Distribution Mixed Model | |
| CN115879018A (en) | Cabin equipment state perception and fault diagnosis method based on K-means algorithm and BP neural network | |
| Li et al. | Boosting imbalanced data learning with Wiener process oversampling | |
| CN113343079A (en) | Attack detection robust recommendation method based on random forest and target item identification | |
| Wang et al. | On-line signature verification using graph representation | |
| CN111476321B (en) | Air flyer identification method based on feature weighting Bayes optimization algorithm | |
| Hervieu et al. | A HMM-based method for recognizing dynamic video contents from trajectories | |
| de Souza et al. | Clustering of interval-valued data using adaptive squared Euclidean distances | |
| Cristiani et al. | A fuzzy multi-class novelty detector for data streams under intermediate latency | |
| Maschler et al. | Towards Deep Industrial Transfer Learning: Clustering for Transfer Case Selection |
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 |