CN105183624A - Data matching based simulation playback method - Google Patents
Data matching based simulation playback method Download PDFInfo
- Publication number
- CN105183624A CN105183624A CN201510543288.1A CN201510543288A CN105183624A CN 105183624 A CN105183624 A CN 105183624A CN 201510543288 A CN201510543288 A CN 201510543288A CN 105183624 A CN105183624 A CN 105183624A
- Authority
- CN
- China
- Prior art keywords
- data
- parallel data
- unit parallel
- critical event
- simulation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a data matching based simulation playback method for performing simulation data playback aimed at a high level architecture (HLA) based distributed simulation system. The method comprises the following steps of: performing integration on raw data of a simulation party plan in a database of the HLA distributed simulation system, and obtaining an N-element parallel data set about the simulation party plan, wherein each N-element parallel datum contains N elements; determining an outlier in the N-element parallel data set and removing the outlier, establishing a three-layer neural network structure of Kx(2K+1)xM, wherein K is the number of simulation time and object attribute values, and M is the sum of a critical event number and 1; determining the critical event number and sample data of the critical event number, and performing network training on the neural network; and sequentially inputting each N-element parallel datum into the trained neural network, obtaining an output result, screening out the N-element parallel data that belongs to the critical events, performing ranking according to the simulation time values, and obtaining a playback scheme for data playback.
Description
Technical field
The invention belongs to distributed simulation technology field, be specifically related to a kind of emulation back method based on Data Matching.
Background technology
In the actual development and application process of distributing emulation system, data record and playback is accurately the important foundation that engineering staff carries out quantitative test and feasibility assessment, especially in practical application simulation process, record the interactive information of each simulation object between the state and each simulation object in each moment and can provide accurate, fine-grained data for VV & A (Verification, ValidationandAccreditation).The data of record can not only produce significant role in the debugging of analogue system with correction, can also carry out complete reproduction to whole process, carry out Parameter analysis for engineering staff, and the correctness worked to each stage, validity are carried out comprehensively assessment and provided support.Therefore, data record and review occupies very high status in simulation process.
The domestic and international research for Data logger grid playback concentrates on the research of collecting method mostly, and improve the efficiency of image data, for playback, only to realize the playback of panorama overall process, the playback for special scenes and specific fragment is extremely limited to.Especially for the big-and-middle-sized analogue system of distributed emulation, its simulated members more and also emulation terminate after data volume huge, if the playback of carrying out of overall process is not a high efficiency method, many times intuitively can not reflect the actual conditions of emulation.Such as, HLA (HighLevelArchitecture) analogue system comprising eight federal members, about long two hours each simulation run time, but its actual effectively fragment and critical event only have some, the scheme that the overall process playback of 2 hours has not been, waste resource and time, playback efficiency is poor.In similar simulation process, the simulation run time is long and validity event is less, this just needs to redesign playback architecture, and also the data of not all record are necessary for playback and systematic analysis, from lengthy and jumbled data, valid data are got in pick, carry out playback for the critical event in whole replayed section and fragment, the optimizing research for analogue system provides aid decision making support.
Summary of the invention
In view of this, the invention provides a kind of emulation back method based on Data Matching, integrate the storage data after distributing emulation system end of run, recur for event, shorten the actual playback time, improve the efficiency of emulation playback.
In order to achieve the above object, the distributing emulation system that the present invention is directed to based on tall framed tube carries out emulated data playback, and technical scheme is:
Step 1: after the distributing emulation system simulation run of HLA terminates, select from the database of system and call emulation side's prediction scheme, reading the raw data about this emulation side's prediction scheme.
Step 2: the raw data that step 1 obtains is integrated, obtain the N unit parallel data collection about the raw data of this emulation side's prediction scheme, every bar N unit parallel data comprises N number of element and is respectively: simulation time value, simulation object class name, object instance name, object properties name, object properties type and object attribute values δ, object attribute values δ has k type.
Step 3: concentrate the object attribute values δ in each bar N unit parallel data to average to N unit parallel data
for each object attribute values δ, calculate from δ to
mahalanobis generalised distance
then outlier is detected by the maximum residual Grubb method of inspection; If
be confirmed as outlier, then the N unit parallel data belonging to this object attribute values concentrates rejecting as outlier from N unit parallel data;
Step 4: the three-layer neural network structure setting up a K × (2K+1) × M, K input node is respectively simulation time value and the object attribute values of N unit parallel data, K=k+1; Middle layer is (2K+1) individual node; The output state of this neural network is critical event belonging to this test input vector, then M is that critical event number adds 1, namely increases the output state that does not belong to any critical event.
In advance according to actual task and the target determination critical event quantity of analogue system, construct the sample data of mating with critical event, and use sample data to carry out network training to neural network.
Step 5: each bar N unit parallel data N unit parallel data after rejecting through step 3 concentrated inputs the neural network after training successively, obtain Output rusults, go out N unit parallel data according to Output rusults identifiable design and belong to which critical event or do not belong to any critical event.
Step 6: filter out the N unit parallel data belonging to critical event, and sort according to simulation time value, draw playback scheme.
Step 7: carry out data readback according to playback scheme.
Further, the object attribute values δ in each bar N unit parallel data is concentrated to average to N unit parallel data in step 3
time, only adopt the object D coordinates value in object attribute values, ask the average of every one-dimensional coordinate in this D coordinates value to obtain average three-dimensional coordinate, and calculate the Mahalanobis generalised distance of this D coordinates value to average three-dimensional coordinate.
Further, critical event type comprises: target is found, target is blocked and target arrives destination three class, if there be n target, then concrete event number is 2n+1.
Further, in neural network structure, the neuronic transport function in middle layer is S type tan, and the transport function of output neuron is S type logarithmic function, and select Trainlm to be the training function of network, learning function gets Learngdm, and performance function gets Mse.
Beneficial effect:
1, the present invention proposes a kind of emulation back method based on Data Matching, solves the shortcoming that the data separate existed in prior art is poor, emulation playback duration is long; The pre-service of data, data dependence coupling can be realized and emulate generation and the reproduction of playback event; The distributed information system playback Software for Design based on C/S structure can be realized, for the optimization of analogue system provides decision support.The lengthy and jumbled degree of its data reduces.Can batch execution data, for the emulated data that data volume is huge, there is good application.
2, emulate playback efficiency to improve.Improve the situation that longer, the emphasis consuming time of overall playback is not in the past given prominence to, be that the playback strategy of guiding not only shortens playback duration with event, more decision optimization provides and helps intuitively.
3, good extensibility.Be not only applicable to specific network information analogue system, can also be applicable to based on distributed military affairs and civilian analogue system, versatility is stronger.
Accompanying drawing explanation
Fig. 1 is simulated events back method process flow diagram;
Fig. 2 is three layers of BP neural network topology structure figure;
Fig. 3 is simulated events playback scheme schematic diagram;
Fig. 4 is event replay display and operation chart.
Embodiment
To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.
Below with the distributing emulation system-network information analogue system based on tall framed tube for background, by reference to the accompanying drawings, describe the present invention, the process flow diagram of whole simulated events playback is as shown in Figure 1.
Step 1: after simulation run terminates, open the MySQL database storing data, connection data storehouse, because network information analogue system has the square prediction scheme of many group different background, the 3rd group of side's of emulation prediction scheme (2 is called in this example, target arrives destination), the related data in reading database.
Step 2: integrate primary data information (pdi) according to " N unit parallel data acquisition pattern ", N number of element that the information extracting simulation time value, simulation object class name, object instance name, object properties name, object properties type and object attribute values comprises as every bar N unit parallel data." the N unit parallel data acquisition pattern " of network information analogue system comprises the three-dimensional coordinate information for simulation time value, target designation, target.Wherein object attribute values δ has k type.
Step 3: outlier inspection is carried out to above data.Mahalanobis generalised distance is used to detect 534 multivariate data points, check through Grubb, can obtain having 6 data points is in outside z value, namely Mahalanobis generalised distance has been exceeded, be identified as outlier, be respectively the data point that time value is 92.6,99.8,122.4,129.4,144.0,192.6, compare through real data, time value be the data point of 92.6,99.8,129.4,144.0 really for outlier, another time value be 127.8,192.2 data point be data exception point but do not check out.Above outlier is rejected from data acquisition scheme.
The object D coordinates value in object attribute values is only adopted to carry out outlier detection in the present embodiment, ask the average of every one-dimensional coordinate in this D coordinates value to obtain average three-dimensional coordinate, and calculate the Mahalanobis generalised distance of this D coordinates value to average three-dimensional coordinate.
Step 4: use BP neural network algorithm coupling emulated data, " N unit parallel data acquisition pattern " is after step 2 pre-service and step 3 outlier detection, as data can be utilized to carry out algorithmic match, choose 400 groups of data and train, as input and the target sample of network.
Determine concrete neural network structure.Adopt three layers of BP neural network structure of K × (2K+1) × M, K input node is respectively simulation time value and the object attribute values of N unit parallel data, K=k+1; Middle layer is (2K+1) individual node; The output state of this neural network is critical event belonging to this test input vector, then M is that critical event number adds 1, namely increases the output state that does not belong to any critical event.
From the actual conditions of network information analogue system, test input variable comprises: simulation time U
t, have in units of s, in object attribute values: object number U
nthe X-direction position coordinate value U of (0 represents destination, and 1 represents first aim, and 2 represent the 2nd target), object
x, object Y-direction position coordinate value U
y, object Z-direction position coordinate value U
z; Suitably expand for former drainage pattern in the present embodiment, add customized event attribute and 0-1 rule, then also have in object attribute values: target be detected situation U
d(0 represent target undiscovered, 1 represents target is found), this object damaged condition U
e(0 represents this object is not damaged, and 1 represents this object damages completely, and the decimal between 0 to 1 represents different damage degree).After above-mentioned design, the value of input layer N is 7, and the value of middle layer (2K+1) is 15.Namely this BP network structure is: input layer has 7 neurons, and there are 15 neurons in middle layer.
Determine critical event number and title and determine output layer M.In network information analogue system, its event mainly comprises three classes: target is found, target is blocked, target arrives destination, and concrete event number is relevant to target number.Such as there is n target, then be associated with 2n+1 kind state respectively, critical event corresponding when this 2n+1 kind state is also emulation playback, that is: target 1 is found, target 2 is found ..., target N is found, target 1 is blocked, target 2 is blocked ..., target N is blocked, target arrives destination, respectively corresponding classification 1,2 ..., 2n+1.Because what call in step 1 is the 3rd group of side's of emulation prediction scheme, namely have two targets, critical event number is 5, respectively: Y
1(target 1 is found) Y
2(target 2 is found) Y
3(target 1 is blocked) Y
4(target 2 is blocked) Y
5(target arrival destination).Consider in the analogue system operational process of reality, the change having a lot of moment object properties to occur is not associated with above-mentioned event, therefore " N unit parallel data acquisition pattern " has 6 kinds through the classification of BP neural network coupling, so that can (0 be adopted, 1) coding form represents, with (0,0,0) represent that classification 0 (without directly related data, also can be expressed as Y
0), with (0,0,1) classification 1 (target 1 is found), (0 is represented, 1,0) classification 2 (target 2 is found), (1 is represented, 0,0) classification 3 (target 1 is blocked), (1 is represented, 1,0) classification 4 (target 2 is blocked), (1,1 is represented, 1) classification 5 (target arrival destination) is represented, therefore the value of output layer M is 6, namely whole BP network hierarchical structure is 7 × 15 × 6, and topological structure as shown in Figure 2.
In advance according to actual task and the target determination critical event quantity of analogue system, construct the sample data of mating with critical event, and use sample data to carry out network training to neural network.
Matlab Neural Network Toolbox is utilized to calculate, the neuronic transport function in middle layer is designed to S type tan, the transport function of output neuron is S type logarithmic function, the input amendment vector of network is represented with P, T represents the object vector of network, and select Trainlm to be the training function of network, learning function gets Learngdm, performance function gets Mse, and wherein function Minmax sets the threshold range of input vector element.After 19 network trainings, network error meets the requirements.
Step 5: each bar N unit parallel data N unit parallel data after rejecting through step 3 concentrated inputs the neural network after training successively, obtain Output rusults, go out N unit parallel data according to Output rusults identifiable design and belong to which critical event or do not belong to any critical event;
Step 6: filter out the N unit parallel data belonging to critical event, and sort according to simulation time value, draw playback scheme; Drawn the playback critical event list of network information analogue system by step 5, comprise 5 kinds of critical events, target 1 is found, target 2 is found, target 1 is blocked, target 2 is blocked, target arrive destination.Each critical event its playback scheme corresponding, comprise its critical event title, simulation time and with event correlation attribute value, for " target arrival destination " event, its playback list is as shown in Figure 3.
Step 7: be loaded into simulating scenes in interface, calls corresponding flag library, selects Initial situation, interface is presented at the position of " 0s " moment two objects and destination, as shown in Fig. 4 (a).Select the event needing playback in the event column of right side, as " target arrival base ", after this event is chosen, " beginning " is selected in menu bar, target according to event simulation playback counter-plan requirement, can bring into operation according to moment and three-dimensional position, according to the time scale adjustment playback duration of 1:5, according to path locus, reproduce simulated events.Select " time-out " in operational process, in moment " 88.4s " situation as shown in Fig. 4 (b), the moment " 206.4s " terminates for this event, shows as shown in Fig. 4 (c), and target arrives destination, completes playback demonstration.
To sum up, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (4)
1., based on an emulation back method for Data Matching, carry out emulated data playback for the distributing emulation system based on tall framed tube, the method is characterized in that, comprise the steps:
Step 1: after the distributing emulation system simulation run of described HLA terminates, select from the database of system and call emulation side's prediction scheme, reading the raw data about this emulation side's prediction scheme;
Step 2: the raw data that step 1 obtains is integrated, obtain the N unit parallel data collection about the raw data of this emulation side's prediction scheme, every bar N unit parallel data comprises N number of element and is respectively: simulation time value, simulation object class name, object instance name, object properties name, object properties type and object attribute values δ, and object attribute values δ has k type;
Step 3: concentrate the object attribute values δ in each bar N unit parallel data to average to described N unit parallel data
for each object attribute values δ, calculate from δ to
mahalanobis generalised distance
then outlier is detected by the maximum residual Grubb method of inspection; If
be confirmed as outlier, then the N unit parallel data belonging to this object attribute values concentrates rejecting as outlier from N unit parallel data;
Step 4: the three-layer neural network structure setting up a K × (2K+1) × M, K input node is respectively simulation time value and the object attribute values of N unit parallel data, K=k+1; Middle layer is (2K+1) individual node; The output state of this neural network is critical event belonging to this test input vector, then M is that critical event number adds 1, namely increases the output state that does not belong to any critical event;
In advance according to actual task and the target determination critical event quantity of analogue system, construct the sample data of mating with described critical event, and use described sample data to carry out network training to described neural network;
Step 5: each bar N unit parallel data N unit parallel data after rejecting through step 3 concentrated inputs the neural network after training successively, obtain Output rusults, go out N unit parallel data according to Output rusults identifiable design and belong to which critical event or do not belong to any critical event;
Step 6: filter out the N unit parallel data belonging to critical event, and sort according to simulation time value, draw playback scheme;
Step 7: carry out data readback according to described playback scheme.
2. a kind of emulation back method based on Data Matching as claimed in claim 1, is characterized in that, concentrates the object attribute values δ in each bar N unit parallel data to average in described step 3 to described N unit parallel data
time, only adopt the object D coordinates value in object attribute values, ask the average of every one-dimensional coordinate in this D coordinates value to obtain average three-dimensional coordinate, and calculate the Mahalanobis generalised distance of this D coordinates value to average three-dimensional coordinate.
3. a kind of emulation back method based on Data Matching as claimed in claim 1, it is characterized in that, described critical event type comprises: target is found, target is blocked and target arrives destination three class, if there be n target, then concrete event number is 2n+1.
4. a kind of emulation back method based on Data Matching as claimed in claim 1, it is characterized in that, in described neural network structure, the neuronic transport function in middle layer is S type tan, the transport function of output neuron is S type logarithmic function, select Trainlm to be the training function of network, learning function gets Learngdm, and performance function gets Mse.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510543288.1A CN105183624B (en) | 2015-08-28 | 2015-08-28 | A kind of emulation back method based on Data Matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510543288.1A CN105183624B (en) | 2015-08-28 | 2015-08-28 | A kind of emulation back method based on Data Matching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105183624A true CN105183624A (en) | 2015-12-23 |
| CN105183624B CN105183624B (en) | 2017-05-31 |
Family
ID=54905716
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510543288.1A Active CN105183624B (en) | 2015-08-28 | 2015-08-28 | A kind of emulation back method based on Data Matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105183624B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108885571A (en) * | 2016-04-05 | 2018-11-23 | 谷歌有限责任公司 | The input of batch machines learning model |
| CN111539178A (en) * | 2020-04-26 | 2020-08-14 | 成都市深思创芯科技有限公司 | Chip layout design method and system based on neural network and manufacturing method |
| WO2021168710A1 (en) * | 2020-02-26 | 2021-09-02 | 深圳市大疆创新科技有限公司 | Information processing method, information processing apparatus and storage medium |
| CN115422397A (en) * | 2022-06-13 | 2022-12-02 | 无人智境(北京)技术有限公司 | Container terminal simulation deduction recording and playback method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005250954A (en) * | 2004-03-05 | 2005-09-15 | Toshiba Corp | Distributed simulation system |
| CN101055530A (en) * | 2006-12-31 | 2007-10-17 | 中国人民解放军63791部队 | Embedded type distributed simulation platform based on HLA facing to space flight measurement and control and its realization method |
| CN103514321A (en) * | 2013-08-12 | 2014-01-15 | 北京理工大学 | General federal member applied to HLA distributed simulation method |
-
2015
- 2015-08-28 CN CN201510543288.1A patent/CN105183624B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005250954A (en) * | 2004-03-05 | 2005-09-15 | Toshiba Corp | Distributed simulation system |
| CN101055530A (en) * | 2006-12-31 | 2007-10-17 | 中国人民解放军63791部队 | Embedded type distributed simulation platform based on HLA facing to space flight measurement and control and its realization method |
| CN103514321A (en) * | 2013-08-12 | 2014-01-15 | 北京理工大学 | General federal member applied to HLA distributed simulation method |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108885571A (en) * | 2016-04-05 | 2018-11-23 | 谷歌有限责任公司 | The input of batch machines learning model |
| CN108885571B (en) * | 2016-04-05 | 2022-04-15 | 谷歌有限责任公司 | Input of batch processing machine learning model |
| WO2021168710A1 (en) * | 2020-02-26 | 2021-09-02 | 深圳市大疆创新科技有限公司 | Information processing method, information processing apparatus and storage medium |
| CN111539178A (en) * | 2020-04-26 | 2020-08-14 | 成都市深思创芯科技有限公司 | Chip layout design method and system based on neural network and manufacturing method |
| CN111539178B (en) * | 2020-04-26 | 2023-05-05 | 成都市深思创芯科技有限公司 | Chip layout design method and system based on neural network and manufacturing method |
| CN115422397A (en) * | 2022-06-13 | 2022-12-02 | 无人智境(北京)技术有限公司 | Container terminal simulation deduction recording and playback method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105183624B (en) | 2017-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110781765B (en) | Human body posture recognition method, device, equipment and storage medium | |
| CN110703057B (en) | Power equipment partial discharge diagnosis method based on data enhancement and neural network | |
| CN110443969A (en) | A kind of fire point detecting method, device, electronic equipment and storage medium | |
| CN113297174B (en) | Land utilization change simulation method based on deep learning | |
| CN105183624A (en) | Data matching based simulation playback method | |
| CN108304328B (en) | Text description generation method, system and device for crowdsourcing test report | |
| CN113821674B (en) | Intelligent cargo supervision method and system based on twin neural network | |
| CN109919084B (en) | Pedestrian re-identification method based on depth multi-index hash | |
| CN110148117A (en) | Power equipment defect identification method and device based on power image and storage medium | |
| CN107239468A (en) | Task node management method and device | |
| CN117114250B (en) | Intelligent decision-making system based on large model | |
| CN118228061A (en) | Device design diagram generation method and computing device | |
| CN112070227B (en) | Neural network training method, device and equipment | |
| CN116089504B (en) | Relational form data generation method and system | |
| CN111353412A (en) | End-to-end 3D-CapsNet flame detection method and device | |
| CN114882242A (en) | Violation image identification method and system based on computer vision | |
| CN112825143A (en) | Deep convolutional neural network compression method, device, storage medium and equipment | |
| CN118378103B (en) | Geographic information system data matching management method based on artificial intelligence | |
| CN112749670B (en) | Pixel-by-pixel classification method, medium and equipment for multi-source remote sensing image | |
| CN114065640B (en) | Data processing method, device, equipment and storage medium of federal tree model | |
| CN112883988B (en) | Training and feature extraction method of feature extraction network based on multiple data sets | |
| CN117744576A (en) | Graph learning-based method and system for estimating delay of signature-level interconnection line | |
| CN117993965A (en) | Evaluation method for site selection of transformer substation based on topographic data | |
| CN117036807A (en) | Distribution line external damage detection method, device and equipment | |
| CN115641535A (en) | Pedestrian detection method and system based on deep learning |
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 |