CN101018164A - A TCP/IP network performance evaluation prediction method - Google Patents
A TCP/IP network performance evaluation prediction method Download PDFInfo
- Publication number
- CN101018164A CN101018164A CN 200710048554 CN200710048554A CN101018164A CN 101018164 A CN101018164 A CN 101018164A CN 200710048554 CN200710048554 CN 200710048554 CN 200710048554 A CN200710048554 A CN 200710048554A CN 101018164 A CN101018164 A CN 101018164A
- Authority
- CN
- China
- Prior art keywords
- network
- network performance
- business
- evaluation
- taboo
- 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
Images
Abstract
The TCP/IP network performance evaluation prediction method comprises: analyzing bandwidth, time delay and missing package rate; using modified BPNN to decide network performance on condition with a certain load. This invention integrates several indexes comprehensively, can predict network load capacity and optimizes business for proper transmission control.
Description
1, technical field
The present invention relates to a kind of TCP/IP network performance evaluation prediction method, belong to the field tests of the Internet.This method is incorporated into intelligent algorithm in the assessment models, proposes based on the network evaluation method of improving the BP neural net, and can increase the traffic forecast method based on the network optimum of genetic algorithm.Compare with conventional method, the inventive method does not need to set up complicated mathematical model, can take all factors into consideration a plurality of network performance indexes, makes evaluation prediction more rationally effectively.
2, background technology
In Network Transmission, need to determine the current state of network, class of business that prediction can be transmitted and quantity, transmission arrive the expeced time of destination, the quality and the efficient of transmission, and extendible type of service.When transmission is broken down, or when transmission quality and inefficiency, need search and determine the key reason etc. of this phenomenon.When designing, safeguard, optimizing similar network simultaneously, need do performance prediction and assessment to objective network in advance.
Therefore, press for the science of foundation, quick, efficient, practical, believable evaluation prediction method, state and performance to network are assessed, for network information transfer provides optimizing decision (selecting suitable transmission opportunity, transmission place, delivering path and optimum combinations of services to transmit), and for network design, maintenance, optimization provides support and foundation.
When implementing network performance evaluation with prediction, need formulate model according to its key element, constitute the evaluation prediction system.Network performance evaluation prediction generally is divided into two kinds of qualitative evaluation and qualitative assessments.So-called qualitative evaluation prediction is meant that existing or network yet to be built carries out roughly performance estimation to one according to certain experience, judges that can network configuration satisfy user's needs etc.Qualitative assessment is exactly the influence to network performance when finding out numerical relation between the quantitative target of reflection network performance and certain or some indexs and changing of utilization mathematical tool or method of measurement.Compare qualitative mode, qualitative assessment prediction has reflected the actual conditions of network performance more accurately, for design and planning network provide more accurate, detailed foundation, makes the more science of making a strategic decision.
Quantitative analysis method commonly used has three kinds of mathematical methods, computer simulation method and actual measurement methods.
1. mathematical methods
Mathematical methods is mainly used the relation between mathematical measure reflection network performance index.By setting up the appropriate mathematical model, the network of a reality be abstracted into one theoretic simple relatively but can reflect the model of live network situation.Again according to each performance measure and other measure and influencing factor between mathematical relationship, realize accurate assurance to whole network performance.This method for network design, optimize and to have great importance, but the strength that the analysis of model need cost a lot of money goes to solve the complex mathematical problem, therefore often the scale of system is had certain restriction.
2. computer simulation method
For simulation, utilize computer on the model of real system, to carry out simulated experiment exactly, the restriction ratio of system scale and complexity is less, can reach the purpose that system is predicted quickly, and the cost of cost is less, only the time for manually working out the operation of simulation program and last machine and the result being analyzed.But between the analog result and actual value that the employing analogy method obtains, often there is error.
3. mensuration
Network measure is the basic means of awareness network operation performance.Mensuration is to use software or special hardware, and the network of setting up and normally run is carried out the collection and the statistical analysis of dynamic data.Popular most of network test softwares and equipment all belongs to this situation at present.The advantage of this method is that accuracy is higher, and the actual conditions of network system can both obtain from using field measurement such as response time, throughput etc.
More than cited method, can as the assessment network approach.And for a network that has existed and normally used, also lacking general evaluation prediction model and criterion at present, conventional method can not fine solution network evaluation forecasting problem.Therefore the present invention adopts intelligent method that network performance is carried out evaluation prediction, with the application extension of artificial intelligence in the field tests of the Internet.
3, summary of the invention
The present invention proposes a kind of performance estimating method of the TCP/IP of being used for network, comprise based on the network performance evaluation that improves the BP neural net with based on the network performance of genetic algorithm and predict two aspects.Its principal character comprises: set up the test and evaluation experiment porch; Performance index such as the bandwidth by phase-split network, time delay, packet loss are set up the evaluation index system of network performance; Proposition is based on the network evaluation method of improving the BP neural net, finish assessment to network state and performance, for network information transfer provides decision support (selecting suitable transmission opportunity, transmission place, delivering path and optimum combinations of services to transmit), and for network operation, optimization, design provides support and foundation; Proposition can increase the traffic forecast method based on the network optimum of genetic algorithm, comprises that network performance is estimated, network carrying ability is estimated and methods such as service optimization.
Based on the network performance evaluation that improves the BP neural net
Network performance comprises validity and two aspects of fairness.From user's angle, validity is meant service quality problems such as end-to-end time delay, delay variation, packet loss; And for fairness, different people has different understanding.
Therefore the present invention is directed to the validity of network, service quality (QoS) is estimated from user perspective.
(1) evaluation tasks determines
1. for the network of transmitting data service only, under the different bandwidth condition, according to the test result of correlated performance index, the service quality of assessment data business.
2. for the network of transmitting multimedia service, under different bandwidth and different business ratio,, assess QoS in Multimedia Service according to the test result of correlated performance index.
For main business-video in the multi-media network and audio frequency, with regard to quality, audio quality is divided into telephony quality and two grades of CD quality; Video quality is divided into high definition TV (HDTV), performs in a radio or TV programme quality, broadcasting-quality, VCR quality and five grades of low speed video conference quality.When assessing the audio frequency and video quality among the present invention, be not to audio frequency and video quality (comprising resolution, frame rate, compressed encoding form etc.) rating itself, but sensorial service quality such as appreciable picture of user and tonequality are assessed.
(2) evaluation index chooses
Different assessment contents, its relevant evaluation index is different.
1. if be single data service in the network, then associated index is mainly packet loss and available bandwidth.
2. if be multimedia service in the network, then, the evaluation index relevant with audio quality is chosen for time-delay, delay jitter and packet loss according to the characteristics of voice signal and vision signal itself based on Voice ﹠ Video; The index relevant with the Video service quality is chosen for delay jitter, packet loss and available bandwidth.
(3) formulation of assessment level
Assessment level of the present invention is: when available bandwidth big more, time-delay, delay jitter and packet loss more hour, then network performance is good more.
If the desired value of a certain Service Quality Metrics is Ψ, the actual value that obtains at moment t is ψ (t).
Definition 1: if the value of ψ (t) is big more, its service quality good more (such as available bandwidth) then.Definition normalization service quality function (t) is
Definition 2: if this service quality of the more little representative of value good more (such as time delay and packet loss) of ψ (t), then definition normalization service quality function (t) is
For n item Service Quality Metrics, have:
Wherein, w
kBe Service Quality Metrics function
kWeight.
When concrete enforcement was assessed, bandwidth was much actually, and how little time-delay, delay jitter and packet loss be actually just calculates, and also needs according to measurement data and contrast service quality to determine.
(4) use of appraisal procedure
In actual applications, the present invention uses adaptive learning speed and momentum term combination algorithm, and the Rprop algorithm improves neural net.In the neural network structure of setting up, as guiding, make the error in the whole learning process keep downward trend with error as far as possible, under the neural metwork training advantageous conditions, just accelerate pace of learning, under the disadvantageous condition, just slow down pace of learning.Method of adjustment is carried out according to following formula.
Wherein, α is a factor of momentum, and η is a learning rate, and m, n, k are system parameters.
Network performance prediction based on genetic algorithm
Under given finger target condition,,, but the network loaded service is predicted by genetic algorithm with the professional associated transport data such as data service, voice and video (comprising dynamic image and still image) that experiment obtains.
The genetic algorithm of the present invention's research is based on the genetic algorithm (Tabu-basedGenetic Algorithm is called for short TGA) of TABU search, promptly utilize genetic algorithm that the parallel search main frame is provided, embed the individual serial search mode of taboo algorithm, introduce greedy algorithm simultaneously, improve algorithm optimization performance index and search efficiency, had better problem-solving ability.
Before describing algorithm, establishing B is the bottleneck link bandwidth, B
iBe transmission occupied bandwidth, R
dBe the average transmission rate of every circuit-switched data, R
vBe the average transmission rate of every road dynamic image, R
pBe the average transmission rate of every road still image, R
sBe the average transmission rate of every road voice, f, are proportionality coefficient, and n is the kind of transport service, k
iTransmission way for a certain business.Just like giving a definition:
Definition 1: t in second with the UDP transfer files, if guarantee low packet loss ratio, the transmission bandwidth that then needs is
B
i=f×R
d
Definition 2: for guaranteeing the quality of UDP dynamic image, the transmission bandwidth that every road video needs is
B
i=R
v×
Definition 3: when transmission UDP still image or voice, needed bandwidth is respectively
B
i=R
s,B
i=R
p
Definition 4: under certain bottleneck bandwidth condition, the business that can transmit is
Algorithm design:
The network performance prediction can be interpreted as: guaranteeing to make the every business of network carrying as much as possible under certain transmission conditions.That is:
m=∑k
i
x
i∈{0,1}
Wherein, maxB is the maximum bottleneck bandwidth of network, and m is selective number of services, s
iBe the value of i kind business, maxs is a total value; x
iBe allele, x
i=1 expression the professional selected of this gene representative can transmit in network; x
iThe business of=0 this gene representative of expression does not have selected.
In network performance forecasting research: neighborhood function N (x) adopts 1-antiposition operator, successively with each gene antiposition, gets one group of neighborhood and separates; The taboo table is by being provided with a m[popsize] [Ichrom] two-dimensional array realizes that recently by the number of times of antiposition, wherein popsize is the population size in this storage of array path, Ichrom is a mrna length; Taboo length t rule of thumb gets t=(0.5~0.7) n with the big or small dynamic change of problem scale, (n is operation algebraically); Candidate solution is a subclass of the neighborhood disaggregation of current state, and promptly corresponding position is 0 the disaggregation that gene constituted among the taboo table m; Despise criterion and represent to avoid the fitness value that object produces if be better than maximum when former generation, still selecting it is next current state.Fitness function
Make Finess (individual (x)) that x ∈ X is a network performance forecasting problem, wherein x is a search condition, and x is finite state collection (being the search condition space).If being algorithm, k carries out algebraically, S
kAll set of separating that search for algorithm to the k generation.Its operating procedure comprises:
1. by greedy algorithm, press B
i/ s
iDescending all business are sorted, produce initial population at random, obtain one group of initial solution, preferentially select B
i/ s
iThan big and x
i=1 business is up to satisfying bottleneck bandwidth to greatest extent, for not transmitting and x
i=1 business just makes x
i=0.Simultaneously m[popsize] [Ichrom] be initialized as 0.
2. carry out the operation of evolving, revise the value of m according to each variation of separating.If certain value among the m is greater than taboo length t, then corresponding positions zero clearing.
3. judge that k whether greater than maximum taboo algebraically tsgen, carries out if satisfy then to turn to 4., otherwise each individuality is carried out TABU search, individual current separating is x
k, x is set
0It is an interim state.
Calculate x
kAll neighborhoods separate.As x
kAll neighborhoods separate all tested mistake, promptly the corresponding positions of m all is not 0, x
iBe tested separating the earliest, then make x
0=x
i, i=min{i; x
i∈ N (x
k), S
K+1=S
k-{ x
i}+{ x
K+1.
Do you if the last step is false, judges to satisfy and despise criterion not? as at x
kThe quilt neighborhood avoided separate, produced greater than the x that separates when the maximum adaptation degree value of former generation
Max, x then
0=x
Max, S
K+1=S
k+ { x
K+1; Otherwise the maximum x in separating according to the neighborhood that the taboo criterion is chosen not taboo
m, make x
0=x
m, S
K+1=S
k+ { x
K+1.
Revise taboo table m according to the principle of first in first out (FIFO).Upgrade current state, i.e. x with interim state
k=x
0, and make k=k+1.
4. produce new population, whether judge k, carry out if satisfy then to turn to 2. less than given maximum iteration time max gen, otherwise, termination routine operation, prediction of output end value.
The invention has the beneficial effects as follows, can provide the index that can quantize, judge for the user whether network adheres to specification moving or newly-built TCP/IP network carries out test, assessment on the performance.This will bring great convenience to network installation, operation, test, maintenance, optimization.Simultaneously, under the known situation of business load, the relevant performance index of network are predicted estimation, for the decision-making of user's transport service is provide advice and advised at particular network.The present invention can record the current performance index of network when business load is unknown, when needing the business load of change (increase or reduce) network, can predict estimation to the relevant performance index of network.
4, description of drawings
Fig. 1 is the experiment porch topological diagram
Fig. 2 is the evaluating system illustraton of model
Fig. 3 obtains the interface for the intelligent evaluation system data
Fig. 4 is a neural net assessment algorithm block diagram
Fig. 5 is intelligent evaluation system synthesis evaluation result figure
Fig. 6 is a network performance Forecasting Methodology block diagram
Fig. 7 is the TGA algorithm flow chart
Fig. 8 can increase professional optimum programming figure as a result for UDP
5, embodiment
Make up the experiment porch shown in the accompanying drawing 1, this platform is formed looped network by SDH, and by the PCM limiting bandwidth.Professional transmission such as loading data, voice and video can record the key index of the current operation of network: bottleneck bandwidth, available bandwidth, packet loss retransmission rate, time delay etc. on this platform.Set up the assessment models shown in the accompanying drawing 2, every data that the collection network test draws are carried out feature extraction and data processing, and according to evaluation criteria the function and the performance of network are assessed, and generate form, document, curve chart etc. intuitively.According to this model can finish the final assessment result of system, network carrying ability is estimated and work such as service optimization.Network performance intelligent evaluation prognoses system as development platform, can realize the assessment and the prediction of off-line with VC++ on a main frame, its data are obtained the interface and seen accompanying drawing 3.
Below mainly from execution mode being carried out specific description based on the network performance evaluation that improves the BP neural net with based on two aspects of network performance prediction of genetic algorithm.
Based on the network performance evaluation that improves the BP neural net
(1) single QoS assessment
In network, have only single data service, then formulate evaluation tasks and be: for the network of transmitting data service only, under the different bandwidth condition, according to the test result of correlated performance index, the transmission quality of assessment data business.
The index of correlation of data traffic transmission quality is chosen for packet loss and available bandwidth.
1. assessment is implemented
At first need available bandwidth and packet loss are carried out normalized, provide target output, as the foundation (detailed step is seen assessment QoS in Multimedia Service part) of evaluation credit rating.
The input layer number of neural net is determined that by the evaluation index number output layer neuron number is determined that by the assessment result type class hidden neuron number adopts the experience trial and error procedure to determine; The hidden layer transfer function is selected nonlinear Sigmoid function for use, and it is the linear function of k that output layer is selected proportionality coefficient for use.The neural net assessment algorithm is seen accompanying drawing 3.
2. assessment result analysis
Data service service quality is divided into " excellent, good, in, poor " 4 grades.
Excellent: corresponding output average is 0.6<d≤1.0, and transfer of data does not have packet loss, and link also has more vacant, and the user can suitably increase traffic carrying capacity.
Very: corresponding output average is 0.4<d≤0.6, and link has necessarily vacant, and the user can increase traffic carrying capacity on a small quantity.
In: corresponding output average is 0.2<d≤0.4, and link is vacant less or almost taken, and a small amount of packet loss occurs, and the user can keep the current business amount.
Difference: corresponding output average is 0.0<d≤0.2, link congestion, and it is serious packet loss to occur, need reduce traffic carrying capacity immediately for guaranteeing transmission.
More than for how to come the given grade of service according to the output result.But still the output result of many group measured values in the short time need be averaged, and come the rating services grade with this average
(2) assessment QoS in Multimedia Service
Video, audio frequency are as the main business in the multi-media network, and its service quality has determined the service quality of whole multimedia business to a great extent.Therefore video is more consistent with the audio service mass ratio in addition, as the discussion of appraisal procedure, below is that representative describes with the assessment video traffic mainly.
1. evaluation tasks: for the network of transmitting multimedia service, under different bandwidth and different business ratio, according to the test result of correlated performance index, assessment video traffic service quality.
2. evaluation index is chosen
Link bandwidth is limited in 2Mbps, 1Mbps and 512kbps, is multimedia service in the network, selects for use available bandwidth, video packet loss, delay jitter as evaluation index.
3. the preliminary treatment of measured value
As the input variable of neural net, the measured value of performance index need be carried out preliminary treatment.
With the measurement data set number scale is n, and the index number that each group data comprises is designated as p, and β is a proportionality coefficient, and then all measurement data constitute matrix X:
X=[x
ij]
n×p
Delay jitter, available bandwidth all need to carry out normalization by above-mentioned formula to be handled, and result is respectively z
I1, z
I2Packet loss is originally as a percentage, and the normalized value of this estimation items only need be converted into percentage corresponding decimal and get final product, and is designated as z
I3
4. neural network structure design
At evaluation tasks, the input layer number of neural net determined by the evaluation index number, respectively with the index normalized as inlet, as z
I1, z
I2, z
I3The output layer neuron number is determined by the assessment result type class; The hidden neuron number adopts the experience trial and error procedure to determine; The hidden layer transfer function is selected nonlinear Sigmoid function for use, and it is the linear function of k that output layer is selected proportionality coefficient for use.
5. assessment is implemented
The test data that experiment is obtained is through after the preliminary treatment, as the training data of neural net and the test data of neural net.With reference to the pertinent regulations of China's radio and television film industry standard " digital TV image quality subjective evaluation method ", provided the target output of video quality, simultaneously with foundation as the rating services grade.
In the process of assessing, the network initial weight is the random number between (1,1); Expected error value chooses 0.001; Frequency of training is 500 times.The neural net assessment algorithm is seen accompanying drawing 4.
6. assessment result analysis
The service quality of video traffic is divided into " fine, better, general, relatively poor, very poor " 5 grades.
Fine (correspondence is output as 0.8<e≤1.0): picture quality is splendid, and is very satisfied.Link also has more vacant, and the user can suitably increase the video traffic amount.
Better (correspondence is output as 0.6<d≤0.8): picture quality is good, and is satisfied.Link has necessarily vacant, and the user can suitably increase data business volume.
Generally (correspondence is output as 0.4<c≤0.6): picture quality is general. still can accept.The user can keep the current business amount.
Relatively poor (correspondence is output as 0.2<b≤0.4): poor image quality, can see reluctantly.Occurred the part stagnation behavior, the user can suitably reduce the current business amount.
Very poor (correspondence is output as 0.0<a≤0.2): picture quality is inferior, can not watch.Video is stagnated, and the user need reduce video way or other traffic carrying capacity immediately.
Because network has uncertainty, can not come the evaluation services quality according to a certain group of measured value separately, the output result of many group measured values in the short time need be averaged, and come the rating services grade with this average.Comprehensive evaluation result is seen accompanying drawing 5, and the user imports 4 groups of evaluating datas, obtains 4 output results, these values is averaged be final appraisal results, and obtained the corresponding grade of service according to evaluation result.
Network performance prediction based on genetic algorithm
By the forecasting research of network transmission performance, for the business transmission is offered suggestions.Network performance Forecasting Methodology block diagram is seen accompanying drawing 6.
(1) determines the value of every kind of UDP business
According to transmission requirement, give different value to every kind of UDP business.If the transport service kind comprises dynamic image, still image, voice and data, wherein need at first to guarantee the transmission of data service, therefore, the value maximum of data service.
(2) determine the bandwidth of every kind of UDP business
With the bottleneck link bandwidth is that the situation of 2000kbps is an example, the network measure data are analyzed, as: R
dBe about 25kbps, R
vBe about 400~650kbps, R
pBe about 180kbps, R
sBe about 25kbps.Determine the bandwidth of the every business of UDP again according to corresponding computational methods, the parameter f value is about 2.5~30, and (the fluctuation situation on background traffic is decided, if almost ripple disable value is 2.5, if fluctuation more then value is 30), the value is about 200kbps, and the n value is 4, as B one timing, k
iValue determine by Bi.
(3) use genetic algorithm to carry out the network performance prediction
By greedy algorithm every kind of UDP business is sorted, produce initial population, re-use taboo-genetic manipulation and evolve, TABU search memory level line maxm=5.TGA algorithm main process is seen accompanying drawing 7 (a), and individual TABU search flow process is seen accompanying drawing 7 (b).UDP can increase professional optimum programming and the results are shown in accompanying drawing 8.
In the experiment, each professional transmission configuration under typical case's bottleneck bandwidth, as: the bottleneck link bandwidth is 2048kbps, when guaranteeing transmission 1 circuit-switched data, 2 road dynamic images, 1 road still image, 3 road voice be can transmit, 9 road still images and 10 road voice perhaps transmitted.When the bottleneck link bandwidth is 1024kbps, when guaranteeing transmission 1 circuit-switched data, can transmit 1 road dynamic image or 5 road still images, 3 road voice.When the bottleneck link bandwidth is 512kbps,, can transmit 2 road still images and 3 road voice for guaranteeing the transmission of data.
Claims (3)
1, a kind of TCP/IP network performance evaluation prediction method, it is characterized in that: by analyzing network key performance index such as bandwidth, time delay, packet loss, utilize and improve the BP neural net method, under given input load, network performance is judged, and adopt network optimum can increase the traffic forecast method based on genetic algorithm, network carried out bearing capacity is estimated and service optimization, the suggestion of reasonable Control Network transmission is provided for the user.
2, according to the described method of claim 1, it is characterized in that: described utilization improves the BP neural net method, and the step of under given input load network performance being judged comprises:
A, determine evaluation tasks;
B, select corresponding evaluation index: if in the network based on the multimedia service of Voice ﹠ Video, then, the evaluation index relevant with audio quality is chosen for time-delay, delay jitter and packet loss according to the characteristics of voice signal and vision signal itself; The index relevant with the Video service quality is chosen for delay jitter, packet loss and available bandwidth.;
C, formulate assessment level: assessment level is: when available bandwidth big more, time-delay, delay jitter and packet loss are more hour, think that network performance is good more, but bandwidth is much actually, how little time-delay, delay jitter and packet loss be actually just calculates, and also needs according to measurement data and contrast service effectiveness to determine;
D, use the intelligent evaluation method: with adaptive learning speed combine with momentum term, the Rprop algorithm improves the BP neural net method, and it applied in the network performance evaluation.
3, according to the described method of claim 1, it is characterized in that: it is as follows that described network optimum based on genetic algorithm can increase the design of traffic forecast method:
The network performance prediction can be interpreted as: guaranteeing to make the every business of network carrying as much as possible, that is: under certain transmission conditions
m=∑k
i
x
i∈{0,1}
Wherein, maxB is the maximum bottleneck bandwidth of network, and m is selective number of services, s
iBe the value of i kind business, maxs is a total value, x
iBe allele, x
i=1 expression the professional selected of this gene representative can transmit x in network
iThe business of=0 this gene representative of expression does not have selected;
In network performance forecasting research: neighborhood function N (x) adopts 1-antiposition operator, successively with each gene antiposition, gets one group of neighborhood and separates; The taboo table is by being provided with a m[popsizeIIchrom] two-dimensional array realizes that recently by the number of times of antiposition, wherein popsize is the population size in this storage of array path, Ichrom is a mrna length; Taboo length t rule of thumb gets t=(0.5~0.7) n with the big or small dynamic change of problem scale, (n is operation algebraically); Candidate solution is a subclass of the neighborhood disaggregation of current state, and promptly corresponding position is 0 the disaggregation that gene constituted among the taboo table m; Despise criterion and represent to avoid the fitness value that object produces if be better than maximum when former generation, still selecting it is next current state, fitness function
Make Finess (individual (x)) that x ∈ X is a network performance forecasting problem, wherein x is a search condition, and x is finite state collection (being the search condition space).If being algorithm, k carries out algebraically, S
kBe all set of separating that algorithm to the k generation searches, its operating procedure comprises:
1. by greedy algorithm, press B
i/ s
iDescending all business are sorted, produce initial population at random, obtain one group of initial solution, preferentially select B
i/ s
iThan big and x
i=1 business is up to satisfying bottleneck bandwidth to greatest extent, for not transmitting and x
i=1 business just makes x
i=0, simultaneously m[popsizeIIchrom] be initialized as 0;
2. the operation of carry out evolving is revised the value of m according to each variation of separating, if certain value among the m is greater than taboo length t, then corresponding positions zero clearing;
3. judge that k whether greater than maximum taboo algebraically tsgen, carries out if satisfy then to turn to 4., otherwise each individuality is carried out TABU search, individual current separating is x
k, x is set
0It is an interim state;
Calculate x
kAll neighborhoods separate, as x
kAll neighborhoods separate all tested mistake, promptly the corresponding positions of m all is not 0, x
iBe tested separating the earliest, then make x
0=x
i, i=min{i; x
i∈ N (x
k), S
K+1=S
k-{ x
i}+{ x
K+1;
If the last step is false, judges whether to satisfy and despise criterion, as at x
kThe quilt neighborhood avoided separate, produced greater than the x that separates when the maximum adaptation degree value of former generation
Max, x then
0=x
Max, S
I+1=S
k+ { x
K+1; Otherwise the maximum x in separating according to the neighborhood that the taboo criterion is chosen not taboo
m, make x
0=x
m, S
K+1=S
k+ { x
K+1;
Revise taboo table m according to the principle of first in first out (FIFO), upgrade current state, i.e. x with interim state
k=x
0, and make k=k+1;
4. produce new population, whether judge k, carry out if satisfy then to turn to 2. less than given maximum iteration time max gen, otherwise, termination routine operation, prediction of output end value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200710048554 CN101018164A (en) | 2007-02-28 | 2007-02-28 | A TCP/IP network performance evaluation prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200710048554 CN101018164A (en) | 2007-02-28 | 2007-02-28 | A TCP/IP network performance evaluation prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN101018164A true CN101018164A (en) | 2007-08-15 |
Family
ID=38726931
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 200710048554 Pending CN101018164A (en) | 2007-02-28 | 2007-02-28 | A TCP/IP network performance evaluation prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101018164A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102055531A (en) * | 2009-11-11 | 2011-05-11 | 上海摩波彼克半导体有限公司 | Method for improving transmission performance of TCP in wireless sensor network |
CN102055613A (en) * | 2010-12-13 | 2011-05-11 | 宁波大学 | Network quality evaluation method |
CN101394324B (en) * | 2008-09-12 | 2011-05-18 | 武汉理工大学 | Hybrid unidirectional time delay estimation method suitable for network control system |
CN101634995B (en) * | 2009-08-13 | 2011-09-21 | 浙江大学 | Network connection speed predicting method based on machine learning |
CN101527895B (en) * | 2009-04-08 | 2012-05-23 | 华为技术有限公司 | Method and device for obtaining service state information |
CN102724543A (en) * | 2012-07-06 | 2012-10-10 | 赛特斯网络科技(南京)有限责任公司 | HMM-based method for implementing media quality analysis evaluation control in IP network |
CN106533745A (en) * | 2016-10-26 | 2017-03-22 | 广西瀚特信息产业股份有限公司 | Network early warning method and device based on TCP |
CN106845627A (en) * | 2017-01-25 | 2017-06-13 | 中国银联股份有限公司 | A kind of data predication method and device |
CN107493519A (en) * | 2016-06-13 | 2017-12-19 | 中兴通讯股份有限公司 | A kind of network quality appraisal procedure and device based on user video experience |
CN108011763A (en) * | 2017-12-07 | 2018-05-08 | 国家电网公司 | Communications data network investment construction appraisal procedure |
CN108039980A (en) * | 2018-02-07 | 2018-05-15 | 电子科技大学 | The more base SAR Topology Structure Design methods of GEO stars-machine |
CN108055147A (en) * | 2017-12-07 | 2018-05-18 | 国家电网公司 | Communications data network service feature analysis method |
CN108900416A (en) * | 2018-06-19 | 2018-11-27 | 浙江水利水电学院 | A kind of network communication optimal path acquisition methods based on mathematical model |
CN109347668A (en) * | 2018-10-17 | 2019-02-15 | 网宿科技股份有限公司 | A kind of training method and device of service quality assessment model |
CN110417664A (en) * | 2019-07-31 | 2019-11-05 | 国家电网有限公司信息通信分公司 | Business route distribution method and device based on power telecom network |
CN113315679A (en) * | 2021-05-26 | 2021-08-27 | 广东电网有限责任公司 | Link packet loss rate inference method and system based on network resource characteristics |
-
2007
- 2007-02-28 CN CN 200710048554 patent/CN101018164A/en active Pending
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101394324B (en) * | 2008-09-12 | 2011-05-18 | 武汉理工大学 | Hybrid unidirectional time delay estimation method suitable for network control system |
CN101527895B (en) * | 2009-04-08 | 2012-05-23 | 华为技术有限公司 | Method and device for obtaining service state information |
CN101634995B (en) * | 2009-08-13 | 2011-09-21 | 浙江大学 | Network connection speed predicting method based on machine learning |
CN102055531B (en) * | 2009-11-11 | 2013-09-25 | 上海摩波彼克半导体有限公司 | Method for improving transmission performance of TCP in wireless sensor network |
CN102055531A (en) * | 2009-11-11 | 2011-05-11 | 上海摩波彼克半导体有限公司 | Method for improving transmission performance of TCP in wireless sensor network |
CN102055613B (en) * | 2010-12-13 | 2012-12-26 | 宁波大学 | Network quality evaluation method |
CN102055613A (en) * | 2010-12-13 | 2011-05-11 | 宁波大学 | Network quality evaluation method |
CN102724543A (en) * | 2012-07-06 | 2012-10-10 | 赛特斯网络科技(南京)有限责任公司 | HMM-based method for implementing media quality analysis evaluation control in IP network |
CN102724543B (en) * | 2012-07-06 | 2014-07-30 | 赛特斯信息科技股份有限公司 | HMM-based method for implementing media quality analysis evaluation control in IP network |
CN107493519A (en) * | 2016-06-13 | 2017-12-19 | 中兴通讯股份有限公司 | A kind of network quality appraisal procedure and device based on user video experience |
CN106533745A (en) * | 2016-10-26 | 2017-03-22 | 广西瀚特信息产业股份有限公司 | Network early warning method and device based on TCP |
WO2018137442A1 (en) * | 2017-01-25 | 2018-08-02 | 中国银联股份有限公司 | Data prediction method and device |
CN106845627A (en) * | 2017-01-25 | 2017-06-13 | 中国银联股份有限公司 | A kind of data predication method and device |
CN108011763A (en) * | 2017-12-07 | 2018-05-08 | 国家电网公司 | Communications data network investment construction appraisal procedure |
CN108055147A (en) * | 2017-12-07 | 2018-05-18 | 国家电网公司 | Communications data network service feature analysis method |
CN108011763B (en) * | 2017-12-07 | 2020-08-11 | 国家电网公司 | Communication data network investment construction evaluation method |
CN108055147B (en) * | 2017-12-07 | 2020-11-03 | 国家电网公司 | Method for analyzing service performance of communication data network |
CN108039980A (en) * | 2018-02-07 | 2018-05-15 | 电子科技大学 | The more base SAR Topology Structure Design methods of GEO stars-machine |
CN108039980B (en) * | 2018-02-07 | 2019-10-25 | 电子科技大学 | The more base SAR Topology Structure Design methods of GEO star-machine |
CN108900416A (en) * | 2018-06-19 | 2018-11-27 | 浙江水利水电学院 | A kind of network communication optimal path acquisition methods based on mathematical model |
CN109347668A (en) * | 2018-10-17 | 2019-02-15 | 网宿科技股份有限公司 | A kind of training method and device of service quality assessment model |
CN109347668B (en) * | 2018-10-17 | 2020-11-06 | 网宿科技股份有限公司 | Training method and device for service quality assessment model |
CN110417664A (en) * | 2019-07-31 | 2019-11-05 | 国家电网有限公司信息通信分公司 | Business route distribution method and device based on power telecom network |
CN110417664B (en) * | 2019-07-31 | 2022-02-25 | 国家电网有限公司信息通信分公司 | Service route distribution method and device based on power communication network |
CN113315679A (en) * | 2021-05-26 | 2021-08-27 | 广东电网有限责任公司 | Link packet loss rate inference method and system based on network resource characteristics |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101018164A (en) | A TCP/IP network performance evaluation prediction method | |
CN110311987A (en) | Node scheduling method, apparatus, equipment and the storage medium of microserver | |
CN113114722B (en) | Virtual network function migration method based on edge network | |
CN111435463A (en) | Data processing method and related equipment and system | |
CN106454437A (en) | Streaming media service rate prediction method and device | |
Regnell et al. | An analytical model for requirements selection quality evaluation in product software development | |
CN113064700A (en) | Container deployment method and device | |
CN110210684A (en) | Grain processing scheme optimization method, apparatus, equipment and storage medium | |
Xiujuan et al. | Machine Learning Techniques for Multi-media Communications in Business Marketing. | |
CN113869521A (en) | Method, device, computing equipment and storage medium for constructing prediction model | |
CN110222379A (en) | Manufacture the optimization method and system of network service quality | |
CN116776006B (en) | Customer portrait construction method and system for enterprise financing | |
CN110852808A (en) | Asynchronous adaptive value evaluation method of electronic product based on deep neural network | |
Chen et al. | Fuzzy-neural approaches with example post-classification for estimating job cycle time in a wafer fab | |
CN114493641A (en) | Information display method and device, electronic equipment and computer readable medium | |
CN115002031B (en) | Federal learning network flow classification model training method, model and classification method based on unbalanced data distribution | |
CN106681803A (en) | Task scheduling method and server | |
CN116455820A (en) | Multi-transmission path adjustment system and method based on congestion avoidance | |
CN114418701A (en) | Method and device for generating recommendation list, electronic equipment and storage medium | |
CN113570114A (en) | Intelligent matching method and system for resource service and computer equipment | |
Kudou et al. | A Performance Evaluation of Tandem Queueing Systems by Machine Learning | |
CN115001580B (en) | Customized generation and distribution method for simulated service flow of multidimensional space optical cable network | |
CN112712160B (en) | AHP technology-based group behavior prediction method and device | |
CN117076131B (en) | Task allocation method and device, electronic equipment and storage medium | |
Korytkowski | Optimization of production capacity in intangible flow production systems |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Open date: 20070815 |