CN105005505B - The method for parallel processing of aerial multi-target track prediction - Google Patents
The method for parallel processing of aerial multi-target track prediction Download PDFInfo
- Publication number
- CN105005505B CN105005505B CN201510357525.5A CN201510357525A CN105005505B CN 105005505 B CN105005505 B CN 105005505B CN 201510357525 A CN201510357525 A CN 201510357525A CN 105005505 B CN105005505 B CN 105005505B
- Authority
- CN
- China
- Prior art keywords
- target
- calculate node
- data
- target data
- task
- 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
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000012545 processing Methods 0.000 title claims abstract description 39
- 238000004891 communication Methods 0.000 claims abstract description 5
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000013507 mapping Methods 0.000 claims description 12
- 238000012544 monitoring process Methods 0.000 claims description 10
- 230000005540 biological transmission Effects 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 6
- 230000008034 disappearance Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 4
- 230000001934 delay Effects 0.000 claims description 4
- 235000013399 edible fruits Nutrition 0.000 claims description 2
- 230000008569 process Effects 0.000 description 8
- 230000003139 buffering effect Effects 0.000 description 7
- 238000013461 design Methods 0.000 description 4
- 241001269238 Data Species 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000000153 supplemental effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000004064 recycling Methods 0.000 description 2
- 239000000523 sample Substances 0.000 description 2
- 230000006641 stabilisation Effects 0.000 description 2
- 238000011105 stabilization Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000010304 firing Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 230000003245 working effect Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the method for parallel processing of aerial multi-target track prediction, belong to the technical field of aerial multi-target track prediction.It builds and includes under cluster environment:Host node, the Distributed Architecture for the calculate node for predicting multi-target track for being responsible for task scheduling and logic transaction management, using MPI standard and Pthreads standard two-stages parallel form realize communication between host node, calculate node, logic affairs it is parallel, and propose a kind of Task Assigned Policy based on flight state, so that the task distribution of each calculate node is more balanced under the premise of effectively call duration time is shortened, it is low and the technical issues of system real time cannot be met to solve single computer node processing capacity.
Description
Technical field
The invention discloses the method for parallel processing of aerial multi-target track prediction, belong to aerial multi-target track prediction
Technical field.
Background technology
Aerial multi-target track prediction needs to provide instantaneous position, the speed of target into line trace to a large amount of aerial targets
With the indication information such as drop point.System must distribute within the extremely short time and calculate mass data, and requirement of real-time is high.
Multi-target track prediction task data has the characteristics that:1. lateral independence, i.e. between each batch of target mutually solely
It is vertical;2. longitudinal correlation needs to carry out data backtracking to same a collection of target orbit determination and track correct;3. single goal data stabilization
Property, i.e., it can be continuously available the target data under the premise of target does not fly away from monitoring area;4. destination number unstability, i.e., newly
Target generates, old target disappears.
Serial processing mode is used when handling multi-target track prediction task at present, i.e., is pressed in single computer
The time sequencing to arrive according to data is handled successively.The computing capability of multi-core CPU can not be both made full use of in this way, can not also be expired
Sufficient real-time demand.Especially with increasing for aerial target quantity, the task data amount of generation is increased sharply, to system processing power
Requirement it is harsher.
Utilize the cluster environment and multiple programming technology (such as MPI, Message Passing being made of multiple stage computers
Interface, the standard criterion in message sending function library), it is assigned to each computer node after task is carried out classifying rationally
On be carried out at the same time task processing and can greatly shorten execution time of overall task, so as to meet real-time demand.Multiple target rail
Lateral independence in mark prediction task feature makes overall task have higher parallel characteristics, and longitudinal correlation is again so that collection
The formulation of task scheduling approach has certain challenge under group rings border.
For multi-target track prediction task processing research be concentrated mainly on trajectory calculation algorithm, current patent and
There are no the method for parallel processing by multi-target track prediction task deployment under cluster environment in document.
Invention content
The technical problems to be solved by the invention are the deficiencies for above-mentioned background technology, provide aerial multi-target track
The method for parallel processing of prediction, it is contemplated that the lateral independence of multi-target track prediction, longitudinal correlation propose a kind of based on meter
The parallel processing plan of calculation machine cluster, and propose a kind of Task Assigned Policy based on flight state, shorten communication effective
So that the task distribution of each calculate node is more balanced under the premise of time, it is low to solve single computer node processing capacity
And the technical issues of system real time cannot be met.
The present invention adopts the following technical scheme that for achieving the above object:
The method for parallel processing of aerial multi-target track prediction, includes the following steps:
It builds and includes under cluster environment:It is responsible for host node, the prediction multiple target rail of task scheduling and logic transaction management
The system of the calculate node of mark, each calculate node is mutual indepedent, and the system uses MPI standard and Pthreads standards two-stage simultaneously
Capable mode realize communication between host node, calculate node, logic affairs it is parallel;
Host node is stored in map mapping tables after target data is classified by lot number, will be sorted according to task scheduling strategy
Target data is sent to calculate node, the volume of calculate node being assigned to Taskassign token records every batch of target data
Number, the task amount of each calculate node is assigned to Proctasknum token records, each calculate is saved according to goal task disappearance probability
The task amount that point generates is estimated to obtain the task amount distributed to each calculate node;
Target data receiving thread and trajectory calculation thread parallel perform in calculate node, mark to record with partial map
Target data receiving thread receives and by the target data that lot number is classified, and it is pre- that trajectory calculation thread completes transmission track after calculating
Result is surveyed to host node.
Method for parallel processing as the aerial multi-target track prediction advanced optimizes scheme, by expression formula:Obtain being assigned to the task amount P of calculate node ii, numbers of the j for calculate node i current goal tasks, mi
For the number of calculate node i current goal tasks,For target data disappear probability function, wherein,J-th goal task arrival time, initial position, current state in respectively i-th of calculate node, Δ t tables
Show that new task does not disappear within the time period.
Further, the construction method of map mapping tables is in the method for parallel processing of the aerial multi-target track prediction:
According to the vector of the lot number generation record target data of initial target data, the vector structure of each batch of target data is recorded
Into map mapping tables,
When there is new target data to arrive, corresponding vector is appended to for the fresh target data for having recorded lot number
In, vector corresponding with its lot number is generated for the fresh target data for not recording lot number,
After detecting that target leaves monitoring range, the record in relation to the target data in map mapping tables is deleted.
It further, will according to task scheduling strategy in the method for parallel processing of the aerial multi-target track prediction
The method that sorted target data is sent to calculate node is:After new target data arrives, when the length of fresh target data
When degree is more than target data packet sending threshold value, fresh target data are sent to the calculate node of task amount minimum by host node.
Further, the method for parallel processing of the aerial multi-target track prediction, using N × T+a as time threshold
Whether detection target leaves monitoring range, and N is target data packet sending threshold value, and T occurs for target between disappearing for the first time
Time interval, a are the Network Transmission Delays upper limit.
Method for parallel processing as the aerial multi-target track prediction advanced optimizes scheme, and target data receives
Thread and trajectory calculation thread are interacted using Semaphore Mechanism:When calculate node receives data, semaphore adds 1, track
Semaphore subtracts 1 after computational threads copy data from partial map tables.
Method for parallel processing as the aerial multi-target track prediction advanced optimizes scheme, records each batch of target
The vector of data includes:Target location coordinate, warp-wise speed, broadwise speed, observation moment.
The present invention is had the advantages that using above-mentioned technical proposal:
1st, since task data has longitudinal correlation, i.e., same a collection of target orbit determination and track correct are needed to carry out data
Backtracking, the present invention is by the way of task and calculate node binding, i.e., the data of same target can only be sent to same calculating
It is calculated on node, historical datas all in this way are stored on local node, without being passed with other nodes when data are recalled
Transmission of data reduces call duration time significantly;
2nd, under the mode bound in task and calculate node, a kind of task scheduling strategy based on flight state is proposed,
The probability to be disappeared using target estimates the task amount that it is generated, when task is distributed as reference standard so that each
The task distribution of a calculate node is more balanced;
3rd, since task data arrives in real time, calculate node sThread0:The data that Master nodes are sent are received,
Data are reclassified, revert to the form of map tables in Master nodes, are denoted as partial map, sThread0 is receiving number
According to when first using MPI asynchronous probe function MPI_IProbe detection MPI buffering areas, MPI synchronizations are reused when having data arrival
Receiver function receives data, avoids the situation that CPU is occupied always;
4th, Semaphore Mechanism is used in the interaction for designing sThread0 and sThread1, concrete operations are as follows:When
When sThread0 receives data, semaphore adds 1;Data are copied in sThread1 to partial map tables, semaphore subtracts 1, so
After calculated, calculate after the completion of directly return result to host node, can be to avoid searching loop using Semaphore Mechanism
Partial map tables, the situation that CPU is caused to dally without data, save computing resource.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
It obtains significantly or is recognized by the practice of the present invention.
Description of the drawings
Fig. 1 predicts parallel processing frame for aerial multi-target track;
Fig. 2 builds real-time process for map tables;
Fig. 3 (1), Fig. 3 (2) are respectively calculate node sThread0 and sThread1 process flow;
Fig. 4 takes turns process chart for Master nodes Thread1 mono-;
Fig. 5 (1) is initial map tables;
Fig. 5 (2) is the map tables after target ph5 arrives;
Fig. 6 is Taskassign tables;
Fig. 7 is Proctasknum tables.
Specific embodiment
Embodiments of the present invention are described below in detail, the embodiment below with reference to attached drawing description is exemplary
, it is only used for explaining the present invention, and be not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singulative " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that is used in the specification of the present invention arranges
Diction " comprising " refers to there are the feature, integer, step, operation, element and/or component, but it is not excluded that presence or addition
Other one or more features, integer, step, operation, element, component and/or combination thereof.It should be understood that when we claim
Element is " connected " or during " coupled " to another element, it can be directly connected or coupled to other elements or can also deposit
In intermediary element.In addition, " connection " used herein or " coupling " can include wireless connection or coupling.Wording used herein
"and/or" includes any cell of one or more associated list items and all combines.
It will be understood to those skilled in the art that unless otherwise defined, all terms used herein are (including technical term
With scientific terminology) have the ordinary technical staff in the technical field of the invention's to be commonly understood by identical meaning.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, will not be with idealizing or the meaning of too formal be explained.
Ph is target lot number, and system is to the unique mark of every batch of target;T be judge target disappear time threshold, i.e., from
Target occurs then thinking that target disappears to no appearance after T seconds for the first time;N be every batch of target data packet sending threshold value, i.e., for
Every batch of target data retransmits to reduce network communication after accumulating N packets.
(1) architecture design
The overall architecture of trajectory predictions task processing is built using master slave mode (Master/Slave), by computer node
It is divided into two classes:Task scheduling and logic transaction management node (Master nodes, only there are one) and calculate node (Slave nodes,
Have multiple).Wherein Master nodes are responsible for receiving initial data, classification of task, task distribution scheduling and result of calculation recycling
Deng Slave nodes are responsible for trajectory calculation.The Computational frame using MPI (Massage Passing Interface)+
Pthreads two-stage parallel modes.MPI is responsible for communicating between upper layer node, and Pthread is responsible in bottom layer node each logic affairs simultaneously
Row.It, can be flexible and convenient by way of increasing calculate node when number of tasks exceeds system load using this distributed structure/architecture
Be extended.Its parallel processing frame is shown in attached drawing 1.
(2) Master design of node
Master nodes are mainly responsible for initial data reception, data classification, task distribution, result recycling and other logics
Property processing affairs.A MPI process is opened in Master nodes, is created at three data using pthread in the process
Lineation journey.In data processing, these threads are concurrent workings.Three data processing threads are as follows:
1) Thread0 receives original target data (out of order, each batch of target data is mixed in together), and data are directly deposited
Enter buffering area buffer;
2) there are three tasks as shown in Figure 4 by Thread1:1. reading data from buffering area, divided according to target lot number ph
Class, and map mapping tables are stored in, the data classified 2. are sent to by calculating section according to task scheduling strategy (emphasis description below)
3. point detects target whether also in monitoring range, if target does not have new data arrival in T seconds, then it is assumed that and the target disappears,
Then it discharges corresponding data space and removes record;
3) Thread2 is responsible for receiving the calculating knot that calculate node returns:
Thread1 is thread most complicated in Master nodes and the critical thread for realizing task scheduling;For
Out of order data in buffer, Thread1 needs are classified according to ph follow-up work to be facilitated to distribute, and sorted data are deposited
Enter in a map mapping table, the key values in map mapping tables are ph, identical with the ph values of a collection of target data, in map mapping tables
Value values be the initial data received, by vector vector in the form of store;Each item data is known as a packet in value,
The information such as the target location coordinate comprising a certain moment, warp-wise speed, broadwise speed, observation moment, because of single goal data stabilization
Property and destination number unstability, so being built in real time to map tables, construction method is as follows, and flow chart is shown in attached drawing 2,
New data data arrives:
1. calculate the ph ' values of data;
2. searching ph ' in map tables, if finding, perform 3., otherwise perform 4.;
3. being expert at supplemental data in ph ', while count the length len of value values;If len is more than threshold value N, to calculating
Node is sent, and is performed 5.;
4. map tables increase a line, ph ' and data is stored in;It performs 5.;
5. terminate.
(3) calculate node designs
Can several MPI calculation procedures be opened up (for common according to the computing capability decision for calculating basic body in calculate node
Computer generally only opens one, below give tacit consent to each calculate node in method description and open a calculation procedure), it is each to calculate
Two threads are created in node, the data for being responsible for receiving the distribution of Master nodes are responsible for for one calculating and returning the result, be located
Reason flow is shown in Fig. 3 (1) and Fig. 3 (2).
1)sThread0:The data that Master nodes are sent are received, data are reclassified, are reverted in Master nodes
The form of map tables is denoted as partial map;Because data are real-time, sThread0 is used first when receiving data
The asynchronous probe function MPI_IProbe detections MPI buffering areas of MPI, reuse MPI and synchronize receiver function reception when having data arrival
Data avoid the situation that CPU is occupied always.
2)sThread1:In the interaction for designing sThread0 and sThread1 using Semaphore Mechanism, concrete operations are such as
Under:When sThread0 receives data, semaphore adds 1;Data, semaphore are copied in sThread1 to partial map tables
Subtract 1, then calculated.Host node is directly returned result to after the completion of calculating, it can be to avoid cycle using Semaphore Mechanism
Traversal partial map tables, the situation that CPU is caused to dally without data, save computing resource.
(4) Master nodes are synchronous with calculate node
Since destination number has unstability, Master nodes are added in task with needing to protect during revocation with calculate node
It holds consistent.The present invention takes the mode for omitting control information, and detailed process is as follows:
1) goal task adding procedure is relatively simple, has realized that Master node tasks add in map table developing algorithms, has removed
Update Taskassign tables and Proctasknum tables (being referred in following task scheduling strategy), calculate node are also needed except this
Partial map tables are with reference to map table construction methods, and task adding procedure is same, therefore Master is saved during fresh target arrival
Point completes tasks synchronization with calculate node;
2) target revocation is relative complex, and the present invention sets timestamp one in the map of Master nodes, for recording most
The time that latter secondary data reach, calculating current time and the difference of timestamp, difference are then recognized more than threshold value T when traverse map tables
Target has flown away from monitoring area thus, the delete target information in map tables, at the same update Taskassign tables and
Proctasknum tables safeguard partial map tables, but threshold value at this time is at least N in calculate node using same method
× T+a, wherein, a is the Network Transmission Delays upper limit.Control information is not just needed in this way, and Master nodes and calculate node are reflected
Firing table can keep synchronizing substantially;
(5) Task Assigned Policy
For every a collection of target, Master nodes are sent to calculate node after receiving an appropriate number of data packet (N packets),
In order to realize system load balancing, the present invention devises two tables and carrys out management role distribution condition:Taskassign token records are every
The calculate node number that target is assigned to is criticized, Proctasknum tokens record the allocated task amount of each calculate node,
In actual process, bound for single goal data stability, destination number unstability and task and calculate node
Design, the present invention establish following task models, according to target flight state estimations task amount, propose under computer cluster environment
Task Assigned Policy.
Task model:1. the speed that each target data reaches is identical, every group of data processing time is c;2. target with
Some probability disappears, and constant;3. different target disappearance probability is different, and with certain information-related, such as initial position, flight shape
State etc..It is determined assuming that each target is expected the probability to disappear by function f (t, X, S).Wherein, t is the time, and X is target initial bit
It puts, S is dbjective state (current speed, acceleration);4. new task does not disappear in time Δt;5. without departing from system most
Big processing capacity.
Under the task model, system delay includes two parts, and respectively data processing delay (is calculated and passed including data
It is defeated) and queueing delay, queueing delay is depending on data calculation time and present node number of tasks, when k-th of node tasks number increases
Add 1, queueing delay upper limit linear increase, i.e. Δ G=c × (2mk+ 1), wherein, mkFor k-th of node current goal number of tasks, profit
With queuing model, with the minimum target of time Δt system queuing Delay bound increment, selectMinimum calculating
Node.When there is new task TASK arrivals, processing procedure is as follows:
1. according to Taskassign tables, times of each calculate node in Proctasknum tables is updated using formula f (t, X, S)
Business amountObtain being assigned to the task amount P of calculate node ii, j is calculate node i current goal tasks
Number, miFor the number of calculate node i current goal tasks,For target data disappear probability function, wherein,J-th goal task arrival time, initial position, current state in respectively i-th of calculate node, Δ t tables
Showing hypothesis, new task does not disappear within the time period;
2. inquiring Proctasknum tables, the calculate node P of task amount minimum is foundmin;
3. increasing a line in Taskassign tables, the lot number and P of TASK are recordedmin;
4. TASK is assigned to PminNode.
The task scheduling approach directly uses index of the number of tasks of each calculate node as judge node load,
It is because the time of each target Continuous may be different, such as certain targets are longer because of the time for being in monitoring area, corresponding
Task amount is big;And the target of other calculate node distribution may just leave monitoring range, corresponding task amount within a very short time
Small, calculate node data volume difference to be treated actually each in this way is very big, and each node computational load can be caused uneven,
The advantages of program, is to estimate its task amount generated using the probability that target disappears so that each calculate node
Task distribution is more balanced, to function f (t, X0, S) selection, be defined generally according to the characteristics of monitoring objective.
The computation complexity of above-mentioned allocation strategy is very low, is O (n+m), wherein, n is calculate node number, and m is number of targets.
For ease of the understanding to the embodiment of the present invention, done further by taking several specific embodiments as an example below in conjunction with attached drawing
Explanation, and each embodiment does not form the restriction to the embodiment of the present invention.
One of ordinary skill in the art will appreciate that:Attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Flow is not necessarily implemented necessary to the present invention.
Assuming that:1. there are one Master nodes, three Slave nodes, number is followed successively by 1,2,3, on each Slave nodes
Open up a calculation procedure;2. the middle record of current map tables is shown in attached drawing 5 (1), that is, there are 5 batches of targets, respectively ph1,
ph2,ph3,ph4,ph5;3. current Taskassign tables are shown in attached drawing 6, i.e. ph1 and ph5 targets are assigned to calculate node 1,
Ph2 and ph4 targets are assigned to calculate node 2, and ph3 targets are assigned to calculate node 3;4. current Proctasknum tables
See attached drawing 7, the evaluation criterion on wherein one column of task amount is:
For ease of the understanding to the embodiment of the present invention, done further by taking several specific embodiments as an example below in conjunction with attached drawing
Explanation, and each embodiment does not form the restriction to the embodiment of the present invention.
Example one:Target data ph5 arrives (i.e. already present target data arrives, and does not disappear always)
Step 1:The Thread0 of Master nodes receives target data by network interface, is put into the buffering area opened up
In buffer, this receives the data of ph5 targets;
Step 2:Thread1 reads data ph5 from buffer, inquires map tables;There are target ph5's in map tables at this time
Data so supplemental data in the value being expert in map tables ph5, becomes the form of Fig. 5 (2), while record target data
The time of arrival;
Step 3:The length len of data packet in the value of ph5 is checked, if len is less than the sending threshold value N of setting, no
Transmission data terminates the processing procedure of ph5;If len is equal to sending threshold value N, data are ready for sending, are traversed first
Taskassign tables (Fig. 6) search the calculate node 1 where the target, and MPI_Send () function is called to send N bag datas
To calculate node 1;
Step 4:The sThread0 of Slave nodes receives task data ph5, traverses partial map tables first, if
There is no the data of the target, then are added in last column of table, if it is present by data supplementing to partial map tables
In the corresponding value values of middle ph5, meanwhile, semaphore signal adds 1, and notice sThread1 has data arrival;
Step 5:SThread1 has found that semaphore signal is more than 0, then reads data, and write from partial map tables
Enter the buffering area of sThread1, signal subtracts 1 and becomes 0 later;
Step 6:SThread1, which starts to calculate, carries out trajectory calculation, after having been calculated, calls MPI_Send () function that will tie
Fruit returns to Master nodes;
Step 7:The sThread1 that the Thread2 of Master nodes receives Slave nodes return as a result, Thread2 exists
MPI_Recv () function is called when receiving data, and is needed using MPI_ANY_SOURCE parameters, accordingly even when number is smaller
Calculate node is come to nothing return, will not influence the reception of other calculate node results below;
Step 8:So far, a wheel process flow of target ph5 data is terminated.
Example two:Target data ph6 arrives (i.e. fresh target data arrive)
Step 1:The Thread0 of Master nodes receives target data by network interface, is put into the buffering area opened up
In buffer, this receives the data of ph6 targets;
Step 2:Thread1 reads data ph6 from buffer, inquires map tables, and target ph6 is not present in map tables at this time
Data, so supplemental data in the value being expert in map tables ph6, while record the time of target data arrival;
Step 3:The length len of data packet is 1 in the value of ph6 at this time, less than the sending threshold value N of setting, is not then sent
Data terminate the first round processing procedure of ph6, and the data of ph6 continuously arrive later, when the length of len is equal to sending threshold value
N is then ready for sending data using task scheduling strategy;
Step 4:Use formulaThe task amount of each calculate node is assessed, therefrom selects minimum Pmin,
Ph6 tasks are distributed into the node, at this point, update Taskassign tables, are inserted into record ph6 in bottom line and are allocated to
Pmin, update Proctasknum tables (Fig. 7), the task amount of deposit pmin nodes.MPI_Send () function is called by N bag datas
It is sent to calculate node 1;
Step 5:Later step with example one step 4,5,6,7,8.
Example three:Target disappearance is handled, by taking ph4 as an example
Step 1:When the data ph4 of arrival is inserted into map tables by each Master nodes Thread1, current time can be calculated
And the difference of object time stamp record;
Step 2:If difference is more than N, then it is assumed that ph4 targets have disappeared before, the record before being deleted from map tables,
Update Taskassign tables and Proctasknum tables simultaneously;
Step 3:Target ph4 is thought for fresh target, is handled in a manner that fresh target arrives, such as example one;
Step 4:The sThread0 of Slave nodes safeguards partial map tables, but threshold value at this time using same method
At least N × T+a, wherein a are the Network Transmission Delays upper limit.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can
It is realized by the mode of software plus required general hardware platform.Based on such understanding, technical scheme of the present invention essence
On the part that the prior art contributes can be embodied in the form of software product in other words, the computer software product
It can be stored in storage medium, such as ROM/RAM, magnetic disc, CD, be used including some instructions so that a computer equipment
(can be personal computer, server or network equipment etc.) performs certain parts of the embodiment of the present invention or embodiment
The method.
Claims (7)
1. the method for parallel processing of aerial multi-target track prediction, which is characterized in that include the following steps:
It builds and includes under cluster environment:It is responsible for the host node of task scheduling and logic transaction management, predicts multi-target track
The system of calculate node, each calculate node is mutual indepedent, and the system is using MPI standard and Pthreads standards two-stage parallel
Mode realize communication between host node, calculate node, logic affairs it is parallel;
Host node by target data by map mapping tables are stored in after lot number classification, according to task scheduling strategy by sorted target
Data are sent to calculate node, and the number of calculate node that is assigned to of record every batch of target data is marked with Taskassign, with
Proctasknum token records are assigned to the task amount of each calculate node, and each calculate node is produced according to goal task disappearance probability
Raw task amount is estimated to obtain the task amount distributed to each calculate node;
Target data receiving thread and trajectory calculation thread parallel perform in calculate node, and record target is marked with partial map
Data receiver thread receives and by the target data that lot number is classified, and trajectory calculation thread is completed to send trajectory predictions knot after calculating
Fruit is to host node.
2. the method for parallel processing of aerial multi-target track prediction according to claim 1, which is characterized in that by expressing
Formula:Obtain being assigned to the task amount P of calculate node ii, j is the volume of calculate node i current goal tasks
Number, miFor the number of calculate node i current goal tasks,For target data disappear probability function, wherein,J-th goal task arrival time, initial position, current state in respectively i-th of calculate node, Δ t tables
Show that new task does not disappear within the time period.
3. the method for parallel processing of aerial multi-target track prediction according to claim 1 or 2, which is characterized in that described
The construction method of map mapping tables is:
According to the vector of the lot number generation record target data of initial target data, the vector for recording each batch of target data is formed
Map mapping tables,
When there is new target data to arrive, the fresh target data for having recorded lot number are appended in corresponding vector,
Vector corresponding with its lot number is generated for the fresh target data for not recording lot number,
After detecting that target leaves monitoring range, the record in relation to the target data in map mapping tables is deleted.
4. the method for parallel processing of aerial multi-target track prediction according to claim 3, which is characterized in that according to task
The method that sorted target data is sent to calculate node is by scheduling strategy:After new target data arrives, when new mesh
When marking the length of data more than target data packet sending threshold value, fresh target data are sent to the calculating of task amount minimum by host node
Node.
5. the method for parallel processing of aerial multi-target track prediction according to claim 3, which is characterized in that with N × T+a
Detect whether target leaves monitoring range for time threshold, N is target data packet sending threshold value, and T occurs for the first time for target
To the time interval between disappearance, a is the Network Transmission Delays upper limit.
6. the method for parallel processing of aerial multi-target track prediction according to claim 1, which is characterized in that the target
Data receiver thread and trajectory calculation thread are interacted using Semaphore Mechanism:The semaphore when calculate node receives data
Add 1, semaphore subtracts 1 after trajectory calculation thread copies data from partial map tables.
7. the method for parallel processing of aerial multi-target track prediction according to claim 3, which is characterized in that the record
The vector of each batch of target data includes:Target location coordinate, warp-wise speed, broadwise speed, observation moment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510357525.5A CN105005505B (en) | 2015-06-25 | 2015-06-25 | The method for parallel processing of aerial multi-target track prediction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510357525.5A CN105005505B (en) | 2015-06-25 | 2015-06-25 | The method for parallel processing of aerial multi-target track prediction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105005505A CN105005505A (en) | 2015-10-28 |
CN105005505B true CN105005505B (en) | 2018-06-26 |
Family
ID=54378185
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510357525.5A Active CN105005505B (en) | 2015-06-25 | 2015-06-25 | The method for parallel processing of aerial multi-target track prediction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105005505B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105426256B (en) * | 2015-11-03 | 2019-05-03 | 中电莱斯信息系统有限公司 | A kind of high-volume real-time target method for parallel processing based on multi-process collaboration |
CN109144941A (en) * | 2018-10-12 | 2019-01-04 | 北京环境特性研究所 | Ballistic data processing method, device, computer equipment and readable storage medium storing program for executing |
CN109597680B (en) * | 2018-10-22 | 2023-07-07 | 创新先进技术有限公司 | Task queuing response parameter estimation method and device |
CN110398985B (en) * | 2019-08-14 | 2022-11-11 | 北京信成未来科技有限公司 | Distributed self-adaptive unmanned aerial vehicle measurement and control system and method |
CN115208954B (en) * | 2022-06-07 | 2024-04-26 | 北京一流科技有限公司 | Parallel policy preset system for distributed data processing system and method thereof |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6393362B1 (en) * | 2000-03-07 | 2002-05-21 | Modular Mining Systems, Inc. | Dynamic safety envelope for autonomous-vehicle collision avoidance system |
CN102110079B (en) * | 2011-03-07 | 2012-09-05 | 杭州电子科技大学 | Tuning calculation method of distributed conjugate gradient method based on MPI |
CN103645952B (en) * | 2013-08-08 | 2017-06-06 | 中国人民解放军国防科学技术大学 | A kind of non-precision tasks in parallel processing method based on MapReduce |
CN103716867B (en) * | 2013-10-25 | 2017-10-27 | 华南理工大学 | Based on event driven wireless sensor network multi-target real-time tracking system |
-
2015
- 2015-06-25 CN CN201510357525.5A patent/CN105005505B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN105005505A (en) | 2015-10-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105005505B (en) | The method for parallel processing of aerial multi-target track prediction | |
CN105190543B (en) | For the coordination based on getatability of looping traffic | |
Zhu et al. | Effective and efficient trajectory outlier detection based on time-dependent popular route | |
US8291005B2 (en) | Providing consistency in processing data streams | |
JP5847956B2 (en) | Apparatus, method, and computer program for moving an event detector process | |
CN104601562B (en) | The exchange method and system of game server and database | |
CN105139035A (en) | Mixed attribute data flow clustering method for automatically determining clustering center based on density | |
CN101132270B (en) | Multi-node coordinated time consistency management method | |
CN111813858B (en) | Distributed neural network hybrid synchronous training method based on self-organizing grouping of computing nodes | |
CN107423539A (en) | A kind of label distribution method of GM PHD wave filters | |
CN105897811A (en) | data synchronization method and device | |
Qi et al. | A task-driven sequential overlapping coalition formation game for resource allocation in heterogeneous UAV networks | |
CN108564028A (en) | A kind of multithreading face identification system based on embedded system | |
CN102289491A (en) | Parallel application performance vulnerability analyzing method and system based on fuzzy rule reasoning | |
CN108875035A (en) | The date storage method and relevant device of distributed file system | |
CN110264497A (en) | Track determination method and device, the storage medium, electronic device of duration | |
CN106257447A (en) | The video storage of cloud storage server and search method, video cloud storage system | |
CN111190711B (en) | BDD combined heuristic A search multi-robot task allocation method | |
Forestiero et al. | Flockstream: a bio-inspired algorithm for clustering evolving data streams | |
Berger et al. | A hybrid genetic approach for airborne sensor vehicle routing in real-time reconnaissance missions | |
CN112765766B (en) | Dynamic interactive fine-grained discrete event system time sequence advancing method | |
CN108363865A (en) | The asynchronous transport simulation method and system that PARTICLE TRANSPORT FROM domain decomposition parallel calculates | |
Coles et al. | A temporal relaxed planning Graph heuristic for planning with envelopes | |
CN102932199B (en) | A kind of method and system of multiple nucleus system detection P2P streams | |
Elboher et al. | A formal metareasoning model of concurrent planning and execution |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |