CN105787180A - Large-scale crowd behavior evolution analysis method based on Map-Reduce and multi-agent models - Google Patents
Large-scale crowd behavior evolution analysis method based on Map-Reduce and multi-agent models Download PDFInfo
- Publication number
- CN105787180A CN105787180A CN201610111329.4A CN201610111329A CN105787180A CN 105787180 A CN105787180 A CN 105787180A CN 201610111329 A CN201610111329 A CN 201610111329A CN 105787180 A CN105787180 A CN 105787180A
- Authority
- CN
- China
- Prior art keywords
- intelligent body
- map
- reduce
- key assignments
- list
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/36—Circuit design at the analogue level
- G06F30/367—Design verification, e.g. using simulation, simulation program with integrated circuit emphasis [SPICE], direct methods or relaxation methods
Abstract
The invention discloses a large-scale crowd behavior evolution analysis method based on map-reduce and multi-agent models. The method is achieved through a map-reduce programming model, and simulation operation is achieved through data processing operation; by converting agent information in multi-agent crowd simulation into entries of a key value and attribute value list and by achieving logical judgment decision and status update of agents through map and reduce operations, large-scale crowd simulation is converted into data processing operation in a key value and attribute value list mode. The method is achieved on clusters based on an ApacheHadoop frame. It is proved through tests that the simulation crowds produced through the simulation method are in accordance with existing serialization simulation results in subsection and morphology; according to the method, the scale of the crowds capable of being simulated is large, the operation speed is high, and occupation of a memory working set is greatly reduced.
Description
Technical field
The invention belongs to Simulation and Modeling Technology field, relate to a kind of crowd simulation method, be specifically related to a kind of based on
Map-reduce analyzes method with the large-scale crowd behavior evolution of multiple agent model.
Background technology
Large-scale crowd simulation study, such as emergency evacuation, at facilities planning, disaster assistance with combat terrorism
Deng the research relevant to public safety plays the effect become more and more important.Traditional crowd behaviour research is adopted mostly
By the research method extracting data based on real scene or Physical Experiment.It is high, dangerous that these methods have cost
Greatly, the shortcomings such as many, shortage interactivity are limited.Along with computer simulation technique and the development of High Performance Computing,
The method using emulation carries out the research of large-scale crowd scene and has become as possibility.Compared with traditional method,
The advantages such as crowd behaviour research tool efficiency height based on emulation, low cost and flexibility are strong.Multiple agent model
It is widely used in many evacuation to simulate.
Yet with the complexity of multiple agent model, large-scale crowd simulation is iff using based on unit
Serialization emulation technology as support, it is easy to cause analogue system overload, it is impossible to reach intended imitative
True effect.Intelligent body quantity is not only the problem of performance, and the result of small-scale intelligent body emulation cannot reflect
The dynamics of large-scale crowd.In the research work that this field is conventional, in order to support extensive intelligent body
Efficient execution, have employed various High Performance Computing, such as multi-CPU system ([document 1]), cloud meter
Calculate ([document 2]), Distributed Calculation and grid computing ([document 3-5]), graphics processing unit calculates (GPGPU)
The technology such as ([document 6]) and programmable gate array (FPGA) ([document 7]).These technology are all in various degree
On expand the scale of multiple agent crowd simulation, improve the execution efficiency of emulation.
In recent years, due to the development of data science, being born with the map-reduce programming model of Google is
The big data processing method represented.Map-reduce programming model ([document 6-7]) (whether document 8-9？
Yes) data being applicable to the long form form to key assignments-attribute-value structure carry out process rapidly and efficiently: first
First pass through map process data are expanded, single entry expansion is become some intermediate data；Then
By reduce process, the continuous item in intermediate data is merged process again, to obtain final result.Real
The Computational frame of existing map-reduce programming model, such as Apache Hadoop, it is typically deployed at extensive collection
On group, the process for mass data can be distributed on multiple node, thus realize to big data " point and
Control it " processing mode, significantly promote data process response speed.
In existing large-scale crowd simulation study based on multiple agent, along with the increase of crowd size, imitative
Data volume during Zhen increases considerably, and for the single individuality in emulation colony, operand there is no
It is obviously improved.Computing intensive can preferably be carried out point by traditional distributed parallel Processing Algorithm
Solve and be confusingly dealt into and perform on different node, but the efficient distributed treatment of large-scale data can not be realized.Existing
Large-scale crowd based on distributed parallel mode emulation typically require the actual demand for simulating scenes
Virtual environment is carried out subregion, and realizes the parallelization of simulation process on this basis.This implementation method is not only
Reduce the versatility of simulation algorithm, and the simulated events on cross-subarea border need a large amount of cross-node communications,
Thus increase extra internodal data exchange, reduce and emulate efficiency.
In sum, it would be highly desirable to invent a kind of algorithm using big data to process and platform realizes large-scale crowd behavior
The method of emulation.
[document 1] Mao T, Jiang H, Li J, et al.Parallelizing continuum crowds [C] //Proceedings
of the 17th ACM Symposium on Virtual Reality Software and Technology.ACM,
2010:231-234.
[document 2] Vigueras G, Lozano M, Perez C, et al.A scalable architecture for crowd
simulation:Implementing a parallel action server[C]//Parallel Processing,2008.
ICPP'08.37th International Conference on.IEEE,2008:430-437.
[document 3] Chen D, Wang L, Bian C, et al.A grid infrastructure for hybrid simulations [J].
Computer Systems Science and Engineering,2011,26(1),:197-206.
[document 4] Chen D, Wang L, Wu X, et al.Hybrid modelling and simulation of huge
crowd over a hierarchical grid architecture[J].Future Generation Computer Systems,
2013,29(5):1309-1317.
[document 5] Wang Y, Lees M, Cai W.Grid-based partitioning for large-scale distributed
agent-based crowd simulation[C]//Proceedings of the Winter Simulation Conference.
Winter Simulation Conference,2012:241.
[document 6] Yilmaz E, Isler V,Y Y.The virtual marathon:parallel computing
supports crowd simulations[J].IEEE computer graphics and applications,2009,29(4):
26-33.
[document 7] Georgoudas I G, Kyriakos P, Sirakoulis G C, et al.An FPGA implemented
cellular automaton crowd evacuation model inspired by the electrostatic-induced
potential fields[J].Microprocessors and Microsystems,2010,34(7):285-300.
[document 8] Dean J, Ghemawat S.MapReduce:simplified data processing on large
clusters[J].Communications of the ACM,2008,51(1):107-113.
[document 9] Dean J, Ghemawat S.MapReduce:a flexible data processing tool [J].
Communications of the ACM,2010,53(1):72-77.
Summary of the invention
In order to solve above-mentioned technical problem, the invention provides one based on map-reduce and multiple agent mould
The large-scale crowd behavior evolution of type analyzes method.
The technical solution adopted in the present invention is: a kind of big rule based on map-reduce with multiple agent model
Mould crowd behaviour evolution analysis method, it is characterised in that comprise the following steps:
Step 1: intelligent body information is converted into intelligent body key assignments-property value entry；
Step 2:Map process；Accept the intelligent body key assignments-property value entry of input, and intelligent computing agent
The attraction of all neighborhoods, ultimately generates one by some mediants forming key assignments-property value form data
According to table；
Step 3:Reduce process；Receive the intermediate data table by map the output of process and process；
Step 4: position selects to move with individuality；
Step 5: Generation of simulating data.
As preferably, described in step 1, intelligent body information is converted into intelligent body key assignments-property value entry, described
Intelligent body information includes one group of attribute and a candidate lattices list, and attribute is used for describing the various shapes of intelligent body
State, including a key attributes and some other attribute and one group of weight；Candidate lattices list is used for intelligent body
Decision process, this list stores the attraction force value of all candidate's neighborhoods；Described intelligent body key assignments-attribute
Value entry includes property location, key assignments position and some other predefined attribute；Emulation starts each intelligence before
Intelligent body key assignments-property value entry can be converted into by body, change at by more predefined attributes and upper one
Reason intelligent body property location information and key assignments positional information intelligent body information is converted to successively intelligent body key assignments-
Property value entry.
As preferably, intermediate data table described in step 2 includes an intelligent body key assignments-property value entry, with
And several records are by the entry of candidate's neighborhood relevant parameter, described relevant parameter includes the position letter in candidate field
Breath and its attraction value；IKey is the identifier of intermediate data table clause；The candidate team of same intelligent body
The intermediate data table clause that row are formed has identical mark iKey, and therefore they will be under Hadoop framework
Reduce operation is carried out unify to submit to；After Map process completes, intermediate data table will be pressed by Hadoop framework
Attraction is ranked up.
As preferably, receive the intermediate data table by map the output of process described in step 3 and process,
In processing procedure, there is the intermediate data table clause of identical iKey merged and sort, form new description
Intelligent body key assignments-property value the entry of intelligent body information, this entry addition of the decision information of the intelligent body of correspondence；
The output of Reduce process has pro forma uniformity with the input of map process；At reduce process
The intelligent body of reason constitutes a new key assignments-list of attribute values, the intelligent body key assignments-property value in this list
Entry is with the intelligent body key assignments-property value entry one_to_one corresponding in the input list of map process: output listing is with defeated
Enter list to compare, the information of each intelligent body addition of the decision information of this simulation time sheet；Whole
The new intelligent body key assignments-property value item list of reduce the output of process according to intelligent body competitiveness from height to
Low sequence, ready for next step.
As preferably, it is characterised in that: position described in step 4 selects and individuality moves, and it implemented
The place one by one that the orderly intelligent body key assignments-property value item list of Cheng Shi: reduce the output of process is serialized
Reason, it is ensured that the intelligent body that competitiveness is higher has higher priority in position selects；Every in list
One intelligent body, in candidate list item first accessibility (not by the higher intelligence of barrier or priority
Can body take) position is just chosen as mobile target；During described candidate list is Decision-making of Agent, to intelligence
Arrived in the neighborhood of energy body position periphery is ranked up and obtains；Just moved after target is selected
Journey, is i.e. updated the location coordinate information of current intelligent body, makes the individuality described by intelligent body be moved to
Selected position, exports a new intelligent body item list, and this list has and map in step 2 equally
The input form that intelligent body key assignments-property value item list is identical of process, and directly emulated as the next one
The input of the map process of timeslice.
As preferably, Generation of simulating data described in step 5, it implements process and is: whole simulation process
The execution of asynchronization on Hadoop cluster, ceaselessly circulates using simulation time sheet as unit, when user needs
When wanting emulation data (the namely individual location distribution information of special time sheet) of particular moment, by emulation
The data demand module of system sends a necessary synchronizing signal, and this signal is whole map-reduce process system
Make a global lock, make all nodes of whole hadoop cluster all suspend at current simulation time sheet；From
The intermediate data of all simulation nodes may make up STATIC SIMULATION scene, and returns user；After this process completes,
Global lock disappears, and all nodes recover normal operating condition.
The invention has the beneficial effects as follows:
1. can increase substantially what large-scale crowd behavior evolution was analyzed by emulation based on Map-reduce
Data-handling efficiency, compared to traditional method, has had significantly in speed and treatment effeciency
Improve.
2. by the process of Hadoop cluster, the space complexity of the present invention is greatly reduced, and enables place
The data scale of reason is significantly larger, and more saves internal memory.
3. by the improvement of multiple agent model, it is possible to the behaviour decision making made a distinction for different individualities,
The ability being allowed to response environment change is higher.
Accompanying drawing explanation
Accompanying drawing 1: the simulation process flow chart of the embodiment of the present invention.
Accompanying drawing 2: in the embodiment of the present invention, intelligent body information is converted into the process of key assignments-attribute value table table entry.
Accompanying drawing 3: the process that intelligent body is processed by the employing Map-reduce programming model of the embodiment of the present invention.
Accompanying drawing 4: the experiment of the embodiment of the present invention and comparison diagram.
Detailed description of the invention
Understand and implement the present invention for the ease of those of ordinary skill in the art, below in conjunction with the accompanying drawings and embodiment pair
The present invention is described in further detail, it will be appreciated that enforcement example described herein is merely to illustrate reconciliation
Release the present invention, be not intended to limit the present invention.
For the most methodical deficiency, it is a kind of based on many that the present invention uses map-reduce programming model to devise
The large-scale crowd emulation mode of intelligent body method, and realize the method on Apache Hadoop framework.Intelligence
Key assignments-attribute value table can be converted into by body information list；The timing of simulation process is by the machine of timeslice with circulation
System realizes emulation, and each timeslice correspondence once circulates, and circulation includes intelligent body information key assignments-attribute every time
Map of value table and a reduce operation.Carry out the most serialized grid every time after circulation and select behaviour
Making, this operating process completes the renewal of the positional information to each intelligent body simultaneously, and generates current time
Sheet complete after scene, in order to the response data request operation that may carry out of user.
Ask for an interview Fig. 1, a kind of based on map-reduce Yu multiple agent model large-scale crowd that the present invention provides
Behavior evolution analyzes method, comprises the following steps:
Step 1: intelligent body information is converted into intelligent body key assignments-property value entry；Key assignments-attribute value table wraps
Containing a lot of key assignments-property value entry, each key assignments-property value entry describes intelligent body information；
Intelligent body information includes one group of attribute and a candidate lattices list, such as Fig. 2；Attribute is used for describing intelligence
The various states of body, including a key attributes and some other attribute (as speed, health degree, front several times
Moving recording) and one group of weight；Candidate lattices list is for the decision process of intelligent body, and this list stores
The attraction force value of all candidate's neighborhoods；Described intelligent body key assignments-property value entry includes property location, key assignments position
Put and some other predefined attribute；Before emulation starts, each intelligent body is converted into intelligent body key assignments-genus
Property value entry, change by more predefined attributes and conversion by more predefined attributes and upper one at
Reason intelligent body property location information and key assignments positional information intelligent body information is converted to successively intelligent body key assignments-
Property value entry.
Intelligent body position is that (positional information is during each cycle calculations for its nature static as the reason of key assignments
Constant) with uniqueness (a corresponding intelligent body in position in grid).Intelligence after these formattings
Body entry will give mapping process.Candidate list is ranked up by Hadoop framework according to the health degree of intelligent body.
Step 2:Map process；Accept the intelligent body key assignments-property value entry of input, and intelligent computing agent
The attraction of all neighborhoods, ultimately generates one by some mediants forming key assignments-property value form data
According to table；
Such as Fig. 3, intermediate data table includes an intelligent body key assignments-property value entry, and several records by
The entry of candidate's neighborhood relevant parameter, described relevant parameter includes positional information and its attraction in candidate field
Value；IKey is the identifier of intermediate data table clause；The mediant that the candidate queue of same intelligent body is formed
Having identical mark iKey according to table clause, therefore they will be carried out in reduce operation under Hadoop framework
Unified submission；Map process needs the environmental data formatted, and in order to accelerate simulation velocity, therefore opens in calculating
Before beginning, environmental data constant in whole simulation process is pre-entered in analogue system.Map process completes
After, intermediate data table will be ranked up by attraction by Hadoop framework.
Step 3:Reduce process；Receive the intermediate data table by map the output of process and process；
In processing procedure, there is the intermediate data table clause of identical iKey merged and sort, formed new
The intelligent body key assignments-property value entry describing intelligent body information, this entry addition of the intelligent body of correspondence certainly
Plan information；The output of Reduce process has pro forma uniformity with the input of map process；Through reduce
The intelligent body that process processes constitutes a new key assignments-list of attribute values, and the intelligent body key assignments in this list-
Property value entry and the intelligent body key assignments-property value entry one_to_one corresponding in the input list of map process: output row
Table, compared with input list, addition of the decision information of this simulation time sheet in the information of each intelligent body；
The new intelligent body key assignments-property value item list of whole reduce the output of process according to intelligent body competitiveness from
High to Low sequence, ready for next step.
Step 4: position selects to move with individuality；
During map process and reduce, it is possible to comprise orderly opening for the generation of each intelligent body and put
The decision information of queue, but this process the most directly changes individual physical location.This is because mobile crowd
In body one by one, the motion state of other individualities may be affected, and if such interactional process
Complete parallelization performs then to there is a possibility that specific individuality exists ambiguity, thus causes the logicality of simulation process
Mistake.Therefore, perform during individual mobile needs are serialized at one.
During basis, the orderly intelligent body key assignments-property value item list of reduce the output of process is serialized
Process one by one, it is ensured that the intelligent body that competitiveness is higher has higher priority in position selects；For list
In each intelligent body, in candidate list item first accessibility (not by barrier or priority more
High intelligent body takies) position is just chosen as mobile target；During described candidate list is Decision-making of Agent,
Arrived in the neighborhood of intelligent body position periphery is ranked up and obtains；Just move after target is selected
Process, is i.e. updated the location coordinate information of current intelligent body, makes the individuality described by intelligent body be moved
To selected position；After this step terminates, simulated virtual environment just completes the renewal of a timeslice.This
The intelligent body item list that the output of process one is new, this list has and the input of map process in step 2 equally
The form that list is identical, and directly by the input of the map process as next simulation time sheet.This step is complete
Jump to step 2 after one-tenth and repeat this circulation, completing the simulation process of next timeslice.
Step 5: Generation of simulating data；
Whole simulation process is the execution of asynchronization on Hadoop cluster, does not stops using simulation time sheet as unit
Ground circulation, when user needs emulation data (the namely individual position distribution letter of special time sheet of particular moment
Breath) time, sending a necessary synchronizing signal by the data demand module of analogue system, this signal is whole
Map-reduce process manufactures a global lock, makes all nodes of whole hadoop cluster all in current emulation
Timeslice is suspended；Intermediate data from all simulation nodes may make up STATIC SIMULATION scene, and returns user；
After this process completes, global lock disappears, and all nodes recover normal operating condition.
The performance of correctness and test emulation in order to separately verify emulation logic, the present embodiment has carried out two altogether
Emulation experiment.The experiment porch of two experiments be one comprise a host node and eight from node based on
The small-sized cluster of Hadoop framework and a contrast test work station, all nodes of cluster all with test
Work station has identical hardware configuration, but software merit rating is otherwise varied, as shown in table 1.
Table 1 test platform configures
Map and the reduce process of emulation all uses C Plus Plus to realize and in Hadoop streaming pattern
Lower operation.Simulated environment information generated by patterned map_editor and embed when compiling map and
Inside reduce process procedures.
Experiment one: the logical correctness checking of emulation
In emulation logic verification of correctness is tested, in order to reduce the gap of cluster and unit as far as possible, operate in collection
Emulation based on map-reduce on group has simply used two and has carried out computing from node.This experiment uses
Several scenes and random initial crowd branch.The simulating scenes carried out on cluster is that 10,000 people attempt
From a floor space 6400 square metres, have in 8 square buildings exported and withdraw；Serialized contrast
The simulating scenes of test is then that 2000 people withdraw from the building of same structure, this is because same using
On the premise of simulation model, single work station is used to carry out serialization emulation, the evacuation emulation scene of 2000 people
Close to its disposal ability upper limit.Serialized contrast experiment is based on map-reduce imitative for (1) checking
Correctness the most logically；(2) average each individual, expense of each timeslice in measuring and calculating serialization emulation,
And make contrast with the average overhead of map-reduce mode.
Emulation experiment is shown with initial Crowds Distribute such as Fig. 4 (a) and Fig. 4 (b) of contrast test.Both
It is being uniformly distributed of individual stochastic generation.Two groups of Crowds Distribute testing the tenth timeslice are respectively such as Fig. 4 (c)
With Fig. 4 (b) Suo Shi.In two groups of experiments, all occur in that individuality is divided into 8 groups and is respectively facing 8 different going out
The phenomenon of mouth motion.Test shown in Crowds Distribute such as Fig. 4 (e) and Fig. 4 (f) of the 30th timeslice for two groups.
In two groups of experiments, 8 outlets all become crowded to capacity in this moment as can be seen from Figure.
By contrasting the Crowds Distribute in two groups of different experiments, it appeared that the same phase of emulation experiment, people
The branch of group is about close；This demonstrate that emulation of based on map-reduce programming model is logically
Correctness.
The Time & Space Complexity of two groups of experiments is as shown in table 2.Table shows two groups experiment in first 50
Time & Space Complexity when timeslice is run.Experiment have recorded total CPU time, total internal memory usage amount and
Run the internal memory working set size that 50 timeslices are consumed；According to above-mentioned data, calculate each individuality
Average EMS memory occupation size and each timeslice the average CPU time consume.
Experimental result shows, in terms of time complexity, emulation based on map-reduce is more imitative than serialized
Proper program is more time-consuming: the former completes the workload of five times with the CPU time being equivalent to the latter 19%.
On the one hand this be because emulation based on map-reduce is highly-parallel, is on the other hand because
Map-reduce programming model substituted for the logical operation of complexity with relatively simple data processing operation.
In terms of space complexity, for each individuality, serialized simulated program takies relatively small number of interior
Deposit (being about as much as the 81% of map-reduce emulation), but each individuality of emulation based on map-reduce
The 34% of the internal memory working set size taken only serialization emulation.This is because in map and reduce process
The logical operation in serialization emulation is instead of with data processing operation.In this case, same section of program pin
Being repeatedly executed at predetermined intervals different data sets, the internal memory working set size of each data set consumption is greatly decreased.
The Time & Space Complexity of table 2 emulation
Experiment two: the performance verification of emulation
Simulation performance experiment employs in cluster all of eight and from node and uses more multiple than a upper experiment
Miscellaneous scene and larger crowd: the scene of emulation is a floor space 256,000 square metre, has
16 outlets, internal structure is more complicated, has the square building of relatively multi-obstacle avoidance.The experiment of this group includes two
Individual different emulation experiment process, has 20,000 people (experiment A) and 100,000 people (experiment B) examination respectively
Figure is withdrawn from this is built.In experiment A, crowd density is relatively small, and in experiment B, crowd density is very big, people
Number close to building institute energy galleryful the upper limit.The initial distribution of two crowds is still that uniformly dividing of stochastic generation
Cloth state.Emulation performs 50 timeslices altogether, to the time complexity emulated and sky in two experiments
Between complexity added up, and calculate time of each timeslice consumption and internal memory that each individuality takies is empty
Between, as shown in table 3.
The Time & Space Complexity of table 3 emulation
It should be appreciated that the part that this specification does not elaborates belongs to prior art.
It should be appreciated that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered
Restriction to scope of patent protection of the present invention, those of ordinary skill in the art is under the enlightenment of the present invention, not
Depart under the ambit that the claims in the present invention are protected, it is also possible to make replacement or deformation, each fall within this
Within bright protection domain, the scope that is claimed of the present invention should be as the criterion with claims.
Claims (6)
1. large-scale crowd behavior evolution based on map-reduce with multiple agent model analyzes a method, its
It is characterised by, comprises the following steps:
Step 1: intelligent body information is converted into intelligent body key assignments-property value entry；
Step 2:Map process；Accept the intelligent body key assignments-property value entry of input, and intelligent computing agent
The attraction of all neighborhoods, ultimately generates one by some mediants forming key assignments-property value form data
According to table；
Step 3:Reduce process；Receive the intermediate data table by map the output of process and process；
Step 4: position selects to move with individuality；
Step 5: Generation of simulating data.
Large-scale crowd row based on map-reduce Yu multiple agent model the most according to claim 1
For evolution analysis method, it is characterised in that: described in step 1, intelligent body information is converted into intelligent body key assignments-
Property value entry, described intelligent body information includes one group of attribute and a candidate lattices list, and attribute is used for describing
The various states of intelligent body, including a key attributes and some other attribute and one group of weight；Candidate lattices
List is for the decision process of intelligent body, and this list stores the attraction force value of all candidate's neighborhoods；Described intelligence
Body key assignments-property value entry can include property location, key assignments position and some other predefined attribute；Emulation
Before beginning, each intelligent body is converted into intelligent body key assignments-property value entry；Conversion is by more predefined
Intelligent body information is turned by attribute and the upper intelligent body property location information processed and key assignments positional information successively
It is changed to intelligent body key assignments-property value entry.
Large-scale crowd row based on map-reduce Yu multiple agent model the most according to claim 1
For evolution analysis method, it is characterised in that: intermediate data table described in step 2 includes an intelligent body key assignments
-property value entry, and several records are by the entry of candidate's neighborhood relevant parameter, described relevant parameter includes
The positional information in candidate field and its attraction value；IKey is the identifier of intermediate data table clause；Same
The intermediate data table clause that the candidate queue of individual intelligent body is formed has identical mark iKey, and therefore they will
Reduce operation is carried out unifying to submit under Hadoop framework；After Map process completes, intermediate data table will
It is ranked up by attraction by Hadoop framework.
Large-scale crowd row based on map-reduce Yu multiple agent model the most according to claim 3
For evolution analysis method, it is characterised in that: receive described in step 3 by the intermediate data of map the output of process
Table also processes, and in processing procedure, has the intermediate data table clause of identical iKey merged and arrange
Sequence, forms the new intelligent body key assignments-property value entry describing intelligent body information, and this entry addition of correspondence
The decision information of intelligent body；The output of Reduce process has pro forma uniformity with the input of map process；
The intelligent body processed through reduce process constitutes a new key assignments-list of attribute values, in this list
Intelligent body key assignments-property value entry is with the intelligent body key assignments-property value entry in the input list of map process one by one
Corresponding: output listing, compared with input list, addition of this simulation time sheet in the information of each intelligent body
Decision information；The new intelligent body key assignments-property value item list of whole reduce the output of process is according to intelligence
The competitiveness of body sorts from high to low, ready for next step.
Large-scale crowd row based on map-reduce Yu multiple agent model the most according to claim 1
For evolution analysis method, it is characterised in that: position described in step 4 selects and individuality moves, and it implements
Process is: the orderly intelligent body key assignments-property value item list of reduce the output of process is serialized one by one
Process, it is ensured that the intelligent body that competitiveness is higher has higher priority in position selects；For in list
Each intelligent body, in candidate list item, first accessible position is just chosen as mobile target；Described candidate
During list is Decision-making of Agent, arrived in the neighborhood of intelligent body position periphery is ranked up and obtains
Arrive；Just move process after target is selected, i.e. the location coordinate information of current intelligent body be updated,
Make the individuality described by intelligent body be moved to selected position, export a new intelligent body item list, should
List has and the input shape that intelligent body key assignments-property value item list is identical of map process in step 2 equally
Formula, and directly by the input of the map process as next simulation time sheet.
Large-scale crowd row based on map-reduce Yu multiple agent model the most according to claim 1
For evolution analysis method, it is characterised in that: Generation of simulating data described in step 5, it implements process and is:
Whole simulation process is the execution of asynchronization on Hadoop cluster, ceaselessly follows using simulation time sheet as unit
Ring, when user needs the emulation data of particular moment, sends one by the data demand module of analogue system
Necessary synchronizing signal, this signal is that whole map-reduce process manufactures a global lock, makes whole hadoop
All nodes of cluster all suspend at current simulation time sheet；Intermediate data from all simulation nodes may make up
STATIC SIMULATION scene, and return user；After this process completes, global lock disappears, and all nodes are just recovering
Often running status.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610111329.4A CN105787180A (en) | 2016-02-29 | 2016-02-29 | Large-scale crowd behavior evolution analysis method based on Map-Reduce and multi-agent models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610111329.4A CN105787180A (en) | 2016-02-29 | 2016-02-29 | Large-scale crowd behavior evolution analysis method based on Map-Reduce and multi-agent models |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105787180A true CN105787180A (en) | 2016-07-20 |
Family
ID=56403112
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610111329.4A Pending CN105787180A (en) | 2016-02-29 | 2016-02-29 | Large-scale crowd behavior evolution analysis method based on Map-Reduce and multi-agent models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105787180A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108259608A (en) * | 2018-01-19 | 2018-07-06 | 中国科学院生态环境研究中心 | A kind of multiple agent cluster operation method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101710361A (en) * | 2009-11-13 | 2010-05-19 | 北京航空航天大学 | Distributed traffic simulating platform based on Service-Oriented Architecture (SOA) and simulating method thereof |
| CN104424339A (en) * | 2013-09-11 | 2015-03-18 | 腾讯科技（深圳）有限公司 | Method, device and system for data analysis |
| CN105117224A (en) * | 2015-08-21 | 2015-12-02 | 西安电子科技大学 | Software module clustering method based on multi-agent evolution |
-
2016
- 2016-02-29 CN CN201610111329.4A patent/CN105787180A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101710361A (en) * | 2009-11-13 | 2010-05-19 | 北京航空航天大学 | Distributed traffic simulating platform based on Service-Oriented Architecture (SOA) and simulating method thereof |
| CN104424339A (en) * | 2013-09-11 | 2015-03-18 | 腾讯科技（深圳）有限公司 | Method, device and system for data analysis |
| CN105117224A (en) * | 2015-08-21 | 2015-12-02 | 西安电子科技大学 | Software module clustering method based on multi-agent evolution |
Non-Patent Citations (3)
| Title |
|---|
| JEFFREY DEAN .ETC: ""mapReduce:A flexible data processing tool"", 《COMMUNICATIONS OF THE ACM》 * |
| LIZHE WANG .ETC: ""G-Hadoop: MapReduce across distributed data centers for data-intensive computing"", 《FUTURE GENERATION COMPUTER SYSTEMS》 * |
| TAO YU .ETC: ""A data parallel approach to modelling and simulation of large crowd"", 《CLUSTER COMPUT》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108259608A (en) * | 2018-01-19 | 2018-07-06 | 中国科学院生态环境研究中心 | A kind of multiple agent cluster operation method |
| CN108259608B (en) * | 2018-01-19 | 2020-07-07 | 中国科学院生态环境研究中心 | Multi-agent cluster operation method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ashish et al. | Parallel bat algorithm-based clustering using mapreduce | |
| CN105787227B (en) | More GPU molecular dynamics simulations of structural material irradiation damage | |
| Plagge et al. | Nemo: A massively parallel discrete-event simulation model for neuromorphic architectures | |
| CN112433819A (en) | Heterogeneous cluster scheduling simulation method and device, computer equipment and storage medium | |
| Hu et al. | A swarm intelligent method for traffic light scheduling: application to real urban traffic networks | |
| Ali et al. | Large scale data analysis using MLlib | |
| CN102945198A (en) | Method for characterizing application characteristics of high performance computing | |
| Chen et al. | Massively parallel modelling & simulation of large crowd with GPGPU | |
| Saprykin et al. | Large-scale multi-agent mobility simulations on a GPU: towards high performance and scalability | |
| CN107229234A (en) | The distributed libray system and method for Aviation electronic data | |
| Vu et al. | High-performance mesoscopic traffic simulation with GPU for large scale networks | |
| CN105787180A (en) | Large-scale crowd behavior evolution analysis method based on Map-Reduce and multi-agent models | |
| Yu et al. | A data parallel approach to modelling and simulation of large crowd | |
| CN103177473B (en) | A kind of large scale scene synthetic method of Case-based Reasoning | |
| Cai et al. | Online data-driven surrogate-assisted particle swarm optimization for traffic flow optimization | |
| Miao et al. | A cellular automata based evacuation model on GPU platform | |
| Youssef | Parallelization of a bio-inspired computational model for the simulation of 3-D multicellular tissue growth | |
| Vasileska et al. | Particle-in-cell code for gpu systems | |
| Tang et al. | A job scheduling algorithm based on parallel workload prediction on computational grid | |
| Zhu et al. | Efficient Gaussian Kernel Microcluster Real-Time Clustering Method for Industrial Internet of Things (IIoT) Streams | |
| Zhao et al. | PaCS: A Parallel Computation Framework for Field-Based Crowd Simulation | |
| Nesmachnow et al. | Large-scale multithreading self-gravity simulations for astronomical agglomerates | |
| Petkova | Network Partitioning in Distributed Agent-Based Models | |
| Yimu et al. | A cloud computing service architecture of a parallel algorithm oriented to scientific computing with CUDA and monte carlo | |
| Santucci et al. | Supporting Computer Domains |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160720 |