CN103139008A - Self-adaption method and device capable of detecting message heartbeat period - Google Patents
Self-adaption method and device capable of detecting message heartbeat period Download PDFInfo
- Publication number
- CN103139008A CN103139008A CN201210037272XA CN201210037272A CN103139008A CN 103139008 A CN103139008 A CN 103139008A CN 201210037272X A CN201210037272X A CN 201210037272XA CN 201210037272 A CN201210037272 A CN 201210037272A CN 103139008 A CN103139008 A CN 103139008A
- Authority
- CN
- China
- Prior art keywords
- heart beat
- beat cycle
- neural network
- sample
- sample set
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/10—Active monitoring, e.g. heartbeat, ping or trace-route
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/142—Network analysis or design using statistical or mathematical methods
Landscapes
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Cardiology (AREA)
- Physics & Mathematics (AREA)
- Algebra (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Pure & Applied Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
Abstract
The invention relates to a self-adaption method and a device capable of detecting a message heartbeat period. The method includes that a neural network module is built. An input layer node of the neural network module is a heartbeat period parameter which affects a heartbeat period, and an output layer node is the heartbeat period. A sample set of the neutral network model is composed of a sample parameter of the heartbeat period and a mapping relation of a heartbeat period sample value; when a trigger condition which conforms to self-learning is judged, the neutral network module is triggered to conduct self-learning according the sample set; when a trigger condition is judged to conform to calculation of the heartbeat period, a heartbeat period calculation parameter is collected and inputted to the neural network module, and a heartbeat period calculation value is obtained; when a trigger condition is judged to conform to update of the heartbeat period, a current heartbeat period is updated according to heartbeat period calculation value, or the sample set is updated, and a self-learning step is returned. The self-adaption method and the device capable of detecting the message heartbeat period can dynamically adjust and detect the heartbeat period of message according to a current environment condition.
Description
Technical field
The present invention relates to communication technical field, relate in particular to a kind of adaptive approach and device of detection messages heart beat cycle.
Background technology
For network management system, it is a requisite function that network link detects.Poll check or the network element of network element device (abbreviation network element) by active response NM server (abbreviation webmaster) regularly guarantees that to the mode that webmaster sends the heart beat cycle of detection messages link is unobstructed.As shown in Figure 1, network management system is carried out periodic information interchange by the heart beat cycle of detection messages, network element sends the heart beat cycle of detection messages to webmaster, in the normal situation of link, webmaster receives the heart beat cycle of this detection messages and thinks that the link between webmaster and network element is normal; If abnormal or link causes webmaster not receive the heart beat cycle of the detection messages of network element extremely by network element, show that the communication link between network element and webmaster breaks down.
The following problem that mainly exists is surveyed in existing heartbeat.
1, heart beat cycle is difficult to set
Heart beat cycle generally is set as fixing length, but this cycle often is difficult to satisfy various application scenarios.
2, the heart beat cycle algorithm lacks self-learning function
Some network management system adopts the heart beat cycle Dynamic calculation method, and the fixing computing formula that also often adopts is difficult to adapt to network environment complicated and changeable.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of adaptive approach and device of detection messages heart beat cycle, to solve the problem of the computational methods bad adaptability that has the detection messages heart beat cycle now.
For solving above technical problem, the invention provides a kind of adaptive approach of detection messages heart beat cycle, the method comprises:
Construction step, build neural network model, the input layer of described neural network model is for affecting the heart beat cycle parameter of heart beat cycle, its output layer node is heart beat cycle, and the sample set of described neural network model comprises the sample parameter of heart beat cycle and the mapping relations of heart beat cycle sample value;
Self study triggers step, when judgement meets the trigger condition of self study, triggers described neural network model and carries out self study according to described sample set;
Calculate and trigger step, when judgement meets the trigger condition of calculating heart beat cycle, gather the described neural network model of heart beat cycle calculating parameter input, obtain the heart beat cycle calculated value;
When heart beat cycle step of updating, judgement meet the trigger condition of heart beat cycle renewal, upgrade current heart beat cycle according to described heart beat cycle calculated value, otherwise carry out the sample set step of updating;
The sample set step of updating is upgraded described sample set, returns to the self study step.
Further, described parameter comprises following one or more: network element number, network congestion coefficient, network traffics, capacity utilization, CPU usage, average heart beat cycle, current heart beat cycle, adjusted value.
Further, the described neural network model process of carrying out self study comprises:
Certain sample parameter in the extraction sample set obtains the heart beat cycle sample calculation value of output layer node as the input layer of neural network model;
Calculate the error between heart beat cycle sample value corresponding in this heart beat cycle sample calculation value and described sample;
If this error amount is less than predetermined threshold value, flow process finishes, if this error amount is greater than predetermined threshold value, according to the connection weights of this this neural network model of error amount correction, until the error between the heart beat cycle sample calculation value that recomputates and corresponding heart beat cycle sample value is less than predetermined threshold value.
Further, the described connection weights of adjustment in direction that described neural network descends according to error gradient, and by the connection weights variable quantity that superposes than row last time.
Preferably, in described sample set step of updating, the mapping relations of described heart beat cycle calculating parameter and heart beat cycle calculated value are updated into former sample set, or, add former sample set after concentrating the mapping relations similar to the mapping relations that newly obtain to be weighted merging in former sample.
Alternatively, the trigger condition that described heart beat cycle upgrades is: the deviation between described heart beat cycle calculated value and n-1 the heart beat cycle calculated value that obtains before is all greater than first threshold, and the deviation of the mean value of described heart beat cycle calculated value and n-1 heart beat cycle calculated value obtaining before is greater than Second Threshold.
Further, the trigger condition of the trigger condition of the trigger condition of described calculating heart beat cycle, self study or described heart beat cycle renewal is clocked flip or Event triggered.
For solving above technical problem, the present invention also provides a kind of self-reacting device of detection messages heart beat cycle, and this device comprises:
The neural network model module, be used for calculating heart beat cycle, its input layer is for affecting the heart beat cycle parameter of heart beat cycle, its output layer node is heart beat cycle, the sample set of described neural network model comprises the sample parameter of heart beat cycle and the mapping relations of heart beat cycle sample value, and the method comprises:
The self study trigger module is used for triggering described neural network model and carrying out self study according to described sample set when judgement meets the self study trigger condition;
Calculate trigger module, be used for when judgement meets the trigger condition of calculating heart beat cycle, gather the described neural network model of heart beat cycle calculating parameter input, trigger described neural network model and calculate the heart beat cycle calculated value;
The heart beat cycle update module is used for when judgement meets the update condition of heart beat cycle, according to the current heart beat cycle of heart beat cycle calculated value renewal of described neural network model output;
The sample set update module is for the sample set that upgrades described neural network model.
The self-learning method of detection messages heart beat cycle of the present invention and device, can dynamically adjust the heart beat cycle of detection messages according to the current environment condition, avoid in the situation that the excessive webmaster of offered load or network element arranged the incorrect important service that causes and are affected due to heart beat cycle transmission cycle of detection messages.Simultaneously, avoid heart beat cycle that the incorrect bandwidth that causes and system resource waste are set, can also avoid the impact that network management performance is brought.
Description of drawings
Fig. 1 is heartbeat link detecting schematic diagram of mechanism;
Fig. 2 is that heart beat cycle calculates the BP Artificial Neural Network Structures;
Fig. 3 is the schematic flow sheet of the adaptive approach of detection messages heart beat cycle of the present invention;
Fig. 4 is the modular structure schematic diagram of the self-reacting device of detection messages heart beat cycle of the present invention.
Embodiment
The adaptive approach of this detection messages heart beat cycle as shown in Figure 3, comprises following concrete steps:
Step S101 builds neural network model;
Particularly, neural network model has multiple, and following the present invention describes the adaptive approach of detection messages heart beat cycle of the present invention to adopt BP (Back Propagation, backpropagation) neural network model be example.
Understandably, the inventive method is not limited to the BP neural network model.
Build neural network model when network management system starts, as shown in Figure 2, the input layer of neural network model (also referred to as input layer) is to affect each parameter (X that heart beat cycle calculates
1-X
n), mainly contain network element number n, network congestion coefficient δ, network traffics l, capacity utilization s, CPU usage m, average heart beat cycle v, current heart beat cycle t, adjusted value z etc.
Actual NE quantity in network element number n map network.
Network congestion coefficient δ can adopt following formula to calculate:
A, b represent respectively the numerical value of the load state of current webmaster and network element, the numerical value of the load state of webmaster and network element when a ', b ' represent respectively that the heart beat cycle of detection messages last time receives, 0≤a, b, a ', b '≤1, λ is the server weights, 1≤λ≤10.
Network traffics l is the network traffics that current network flow can be carried divided by maximum.
Capacity utilization s is that the network element device quantity of working in current network is divided by the network element device total quantity.
CPU usage m is the current server CPU usage.
Average heart beat cycle v is the mean value of a network element heart beat cycle after startup of server.
Current heart beat cycle t is the current heart beat cycle set point of corresponding network element to be calculated.
Adjusted value z is dynamically-adjusting parameter, by manually adjusting according to overall calculation deviation situation.
The output layer node of described neural network model (also referred to as the output layer neuron) is heart beat cycle, and the sample set of described neural network model comprises the sample parameter of heart beat cycle and the mapping relations of heart beat cycle sample value.
Hidden layer is responsible for self adaptation and is regulated heart beat cycle, and in general the number of hidden layer node is inputted, the number of output unit and the impact of complication degree of problem to be solved.If input layer performance parameter number is n, according to statistic analysis result, hidden layer node (also referred to as hidden layer neuron) number adopts following formula to calculate the value of gained, and its effect is better:
Wherein, n is the node number of input layer, and m is the node number of output layer, and z is the node number of hidden layer, and k is adjusted value, and the span of k is the integer between 1 to 5.
For hidden layer, j node is output as:
Wherein order
F is the transfer function of hidden layer, X
i, O
jBe respectively the output of i node of input layer and j node of hidden layer, i=1,2 ..., n; J=1,2 ..., z.θ
jBe the threshold value of j neural unit in hidden layer, W
ijThat i node of input layer is to the connection weights of j node of hidden layer.
The output layer node is the heart beat cycle after calculating.Output layer is output as:
Wherein order
G is the output layer transfer function, O
p, Y
qBe respectively the output of p node of hidden layer and q node of output layer, p=1,2 ..., z; Q=1,2 ..., m, θ
qBe q node threshold value of output layer.W
pqThat p node of hidden layer is to the connection weights of q node of output layer.
Step S102 judges whether and need to carry out self study to neural network model, comprises two kinds of situations: initialization self study, and regularly self study.During initialization study, initialization connection weights are the random number (i.e. 0<W<1) between 0-1, if satisfy execution in step S103, otherwise execution in step S104
Self study triggers and can be divided into clocked flip and Event triggered, such as initialization triggering, manual activation, perhaps during network parameter values generation acute variation etc.
Self study and calculating do not have inevitable precedence relationship, for example can carry out at last self study yet.
Step S103 triggers neural network model and carries out self study;
Network element to be learnt is put into formation to be learnt, carry out adaptive learning after other network element self study is complete.According to learning algorithm neural network training model hereinafter described.
Calculate to intert for the self study of each network element and heart beat cycle and carry out.
S101 to S107 can carry out self adaptation for the heart beat cycle of each network element under gateway, also can carry out self adaptation for the heart beat cycle of network element self.
All network elements should be followed same rule or algorithm arranges heart beat cycle in theory.
When netinit or self study cycle of setting when arriving, need to carry out self study to neural network model.
According to current network running status in early stage, utilize existing sample set to learn during initialization.That is to say to provide an experience sample set, each sample of this sample set is as certain specific input and output value of neural network model, the desired output heart beat cycle when namely each performance parameter is worth for certain.
When the self study cycle arrives, the sample set in the time of need to utilizing initialization, the input when calculating before with calculate output valve as new learning sample, utilize learning algorithm again to upgrade each weights of neural network model.That is, the mapping relations that subsequent calculations is obtained add the original sample collection, realize the renewal to sample set.
Exist manyly during to mapping relations in sample set, the mapping relations that are used for carrying out self study can be chosen from sample set at random or successively, can be also up-to-date mapping relations, or a few to or whole weighted averages of mapping relations.
Upgrade the connection weights of neural network model by learning dynamics, make in the situation that input parameter is identical, the heart beat cycle calculated value reaches unanimity with the heart beat cycle sample value as far as possible, thereby improves the accuracy that heart beat cycle calculates.
Concrete, the self study process is as follows:
At first the error function that needs to define reflection neural network model desired output and calculate error size between output is:
E=1/2×(T-O)
2
Wherein, T is the heart beat cycle of experience sample, and 0 is the heart beat cycle of the calculating of output layer node.
The neural network model transfer function can adopt various ways as required.The hidden layer transfer function can adopt tanh S type function Tan-sigmoid function:
f(u)=tansig(u)=2/(1+exp(-2u))-1
The output layer transfer function can adopt linear transfer function purelin (v):
g(v)=purelin(v)=v
Calculate the error of the heart beat cycle calculated value of heart beat cycle sample value and actual output, along the direction of error gradient decline, from output layer through each each parameter value of hidden layer layer-by-layer correction neural network model, until input layer.Right
Adjustment amount
Be proportional to squared error function to the negative derivative of this parameter, right
Adjustment amount
Available following formula represents:
Right
Adjustment amount
Available following formula represents:
This process hockets repeatedly, until the heart beat cycle error in length functional value that calculates less than given minimum ε (value of ε is chosen according to the demand of computational accuracy, 10
-3<ε<10
-7, the ε value is less, and precision is higher, computation complexity is higher, calculates consuming time also more of a specified duration), thereby determined the connection weights corresponding with minimal error, study can stop.Neural network model after study can to input message by oneself, calculate heart beat cycle.
Preferably, neural network is according to the described connection weights of adjustment in direction of error gradient decline, and by the connection weights variable quantity that superposes than row last time, the impact that is about to the weights variation is transmitted by a factor of momentum.When the value of factor of momentum was zero, the variation of weights only produced according to gradient descent method.When the value of factor of momentum was 1, new weights variable quantity was the variable quantity of front weights, and other situations are to carry out timing, the correcting value when once learning before adding by a certain percentage to connecting weights at every turn.Concrete formula is as follows:
X(n+1)=m*(X(n)-X(n-1))-(1-m)*ηΔF(X(n))
Here Δ F (X (n)) is the gradient of target function, and n is iterations, and η is learning rate, and m is factor of momentum, span 0 to 1.Increasing momentum term is to take out a part in a weights adjustment in the past to be superimposed in this weights adjustment amount, with convergence speedup speed.
Step S104 when judgement meets the trigger condition of calculating heart beat cycle, gathers the described neural network model of heart beat cycle calculating parameter input, obtains the heart beat cycle calculated value;
After the initialization self study, this neural network model can be used for the dynamic calculation heart beat cycle.
Gather each parameter, mainly contain the input layer of the input neural network models such as network element number, network congestion coefficient, network traffics, capacity utilization, CPU usage, average heart beat cycle, current heart beat cycle, adjusted value, calculate reference cardiac cycle by neural network model.
Further, in the described time setting, according to network load condition computing reference heart beat cycle, specifically comprise:
Within the time of setting, calculate heart beat cycle according to input parameter by neural network model;
Within the time of setting, network element loading condition (obtaining as heart beat cycle, network element message or other approach by detection messages) is learnt the neural network model input value in conjunction with current webmaster load, network element number, network congestion coefficient, network traffics, capacity utilization, CPU usage, average heart beat cycle, current heart beat cycle, adjusted value etc. as initialization calculate a reference cardiac cycle T.
Step S105 judges whether to meet the trigger condition that heart beat cycle upgrades, and to keep current heart beat cycle constant if do not meet, execution in step S107; Otherwise the heart beat cycle that shows current calculating is accurately, execution in step S106;
The trigger condition that heart beat cycle upgrades can be set according to concrete applied environment, below only provides a kind of example of trigger condition judgement, and is specific as follows:
After calculating n heart beat cycle calculated value, if the deviation between this n heart beat cycle calculated value and n-1 heart beat cycle calculated value obtaining before is all greater than first threshold (span is 1-10 second), and the deviation of described heart beat cycle calculated value and the mean value of n-1 the heart beat cycle calculated value that obtains before is greater than Second Threshold (span is 1-10 second).If meet above-mentioned trigger condition, changing heart beat cycle is n heart beat cycle calculated value.
Can certainly be set as when meeting trigger condition, heart beat cycle is updated to mean value with this n heart beat cycle calculated value.
Step S106 enables current heart beat cycle calculated value, and namely network element sends the heart beat cycle of detection messages, execution in step S107 to webmaster according to current heart beat cycle calculated value;
Step S107 upgrades sample set, turns execution in step S102;
Upgrading the method for sample set also can determine according to concrete algorithm, such as the mapping relations with described heart beat cycle calculating parameter and heart beat cycle calculated value are updated into former sample set, or, add former sample set after concentrating the mapping relations similar to the mapping relations that newly obtain to be weighted merging in former sample.
Understandably, step S102 to step S107 be the circular treatment flow process, do not have strict sequencing.
The self-learning method of heart beat cycle of the present invention, can dynamically adjust the heart beat cycle of detection messages according to the current environment condition, avoid in the situation that the excessive webmaster of offered load or network element arranged the incorrect important service that causes and are affected due to heart beat cycle transmission cycle of detection messages.Simultaneously, avoid heart beat cycle that the incorrect bandwidth that causes and system resource waste are set, can also avoid the impact that network management performance is brought.
In order to realize above method, the present invention also provides a kind of self-reacting device of detection messages heart beat cycle, and as shown in Figure 4, this device comprises:
The neural network model module, be used for calculating heart beat cycle, its input layer is for affecting the heart beat cycle parameter of heart beat cycle, its output layer node is heart beat cycle, the sample set of described neural network model comprises the sample input parameter of heart beat cycle and the mapping relations of heart beat cycle sample value, and the method comprises:
The self study trigger module is used for triggering described neural network model and carrying out self study according to described sample set when judgement meets the self study trigger condition;
Calculate trigger module, be used for when judgement meets the trigger condition of calculating heart beat cycle, gather the described neural network model of heart beat cycle calculating parameter input, trigger described neural network model and calculate the heart beat cycle calculated value;
The heart beat cycle update module is used for when judgement meets the update condition of heart beat cycle, according to the current heart beat cycle of heart beat cycle calculated value renewal of described neural network model output;
The sample set update module is for the sample set that upgrades described neural network model.
The said parameter of the present invention comprises following one or more: network element number, network congestion coefficient, network traffics, capacity utilization, CPU usage, average heart beat cycle, current heart beat cycle, adjusted value.
Particularly, described sample set update module is updated into former sample set with the mapping relations of described heart beat cycle calculating parameter and heart beat cycle calculated value.
Below just provide the mode that a kind of sample set upgrades, added former sample set after can also concentrating the mapping relations similar to the mapping relations that newly obtain be weighted merging in former sample.
The trigger condition that described heart beat cycle upgrades is: the deviation between described heart beat cycle calculated value and n-1 the heart beat cycle calculated value that obtains before is all greater than first threshold, and the deviation of the mean value of described heart beat cycle calculated value and n heart beat cycle calculated value obtaining before is greater than Second Threshold.
The trigger condition that the trigger condition of described calculating heart beat cycle, the trigger condition of self study or described heart beat cycle upgrade is clocked flip or Event triggered.
The adaptive approach of detection messages heart beat cycle of the present invention and device, can dynamically adjust the heart beat cycle of detection messages according to the current environment condition, avoid in the situation that the excessive webmaster of offered load or network element arranged the incorrect important service that causes and are affected due to heart beat cycle transmission cycle of detection messages.Simultaneously, avoid heart beat cycle that the incorrect bandwidth that causes and system resource waste are set, can also avoid the impact that network management performance is brought.
By the explanation of embodiment, should be to reach technological means and the effect that predetermined purpose takes to be able to more deeply and concrete understanding to the present invention, yet appended diagram only be to provide with reference to the use of explanation, the present invention is limited.
One of ordinary skill in the art will appreciate that all or part of step in said method can come the instruction related hardware to complete by program, described program can be stored in computer-readable recording medium, as read-only memory, disk or CD etc.Alternatively, all or part of step of above-described embodiment also can realize with one or more integrated circuits.Correspondingly, each module in above-described embodiment can adopt the form of hardware to realize, also can adopt the form of software function module to realize.The present invention is not restricted to the combination of the hardware and software of any particular form.
Claims (12)
1. the adaptive approach of a detection messages heart beat cycle, is characterized in that, the method comprises:
Construction step, build neural network model, the input layer of described neural network model is for affecting the heart beat cycle parameter of heart beat cycle, its output layer node is heart beat cycle, and the sample set of described neural network model comprises the sample parameter of heart beat cycle and the mapping relations of heart beat cycle sample value;
Self study triggers step, when judgement meets the trigger condition of self study, triggers described neural network model and carries out self study according to described sample set;
Calculate and trigger step, when judgement meets the trigger condition of calculating heart beat cycle, gather the described neural network model of heart beat cycle calculating parameter input, obtain the heart beat cycle calculated value;
When heart beat cycle step of updating, judgement meet the trigger condition of heart beat cycle renewal, upgrade current heart beat cycle according to described heart beat cycle calculated value, otherwise carry out the sample set step of updating;
The sample set step of updating is upgraded described sample set, returns to the self study step.
2. the method for claim 1, it is characterized in that: described parameter comprises following one or more: network element number, network congestion coefficient, network traffics, capacity utilization, CPU usage, average heart beat cycle, current heart beat cycle, adjusted value.
3. the method for claim 1, it is characterized in that: the process that described neural network model carries out self study comprises:
Certain sample parameter in the extraction sample set obtains the heart beat cycle sample calculation value of output layer node as the input layer of neural network model;
Calculate the error between heart beat cycle sample value corresponding in this heart beat cycle sample calculation value and described sample;
If this error amount is less than predetermined threshold value, flow process finishes, if this error amount is greater than predetermined threshold value, according to the connection weights of this this neural network model of error amount correction, until the error between the heart beat cycle sample calculation value that recomputates and corresponding heart beat cycle sample value is less than predetermined threshold value.
4. method as claimed in claim 3 is characterized in that: the described connection weights of adjustment in direction that described neural network descends according to error gradient, and by than row stack connection weights variable quantity last time.
5. the method for claim 1, it is characterized in that: in described sample set step of updating, the mapping relations of described heart beat cycle calculating parameter and heart beat cycle calculated value are updated into former sample set, or, add former sample set after concentrating the mapping relations similar to the mapping relations that newly obtain to be weighted merging in former sample.
6. the method for claim 1, it is characterized in that: the trigger condition that described heart beat cycle upgrades is: the deviation between described heart beat cycle calculated value and n-1 the heart beat cycle calculated value that obtains before is all greater than first threshold, and the deviation of the mean value of described heart beat cycle calculated value and n-1 heart beat cycle calculated value obtaining before is greater than Second Threshold.
7. the method for claim 1 is characterized in that: the trigger condition that the trigger condition of described calculating heart beat cycle, the trigger condition of self study or described heart beat cycle upgrade is clocked flip or Event triggered.
8. the self-reacting device of a detection messages heart beat cycle, is characterized in that, this device comprises:
The neural network model module, be used for calculating heart beat cycle, its input layer is for affecting the heart beat cycle parameter of heart beat cycle, its output layer node is heart beat cycle, the sample set of described neural network model comprises the sample parameter of heart beat cycle and the mapping relations of heart beat cycle sample value, and the method comprises:
The self study trigger module is used for triggering described neural network model and carrying out self study according to described sample set when judgement meets the self study trigger condition;
Calculate trigger module, be used for when judgement meets the trigger condition of calculating heart beat cycle, gather the described neural network model of heart beat cycle calculating parameter input, trigger described neural network model and calculate the heart beat cycle calculated value;
The heart beat cycle update module is used for when judgement meets the update condition of heart beat cycle, according to the current heart beat cycle of heart beat cycle calculated value renewal of described neural network model output;
The sample set update module is for the sample set that upgrades described neural network model.
9. device as claimed in claim 8, it is characterized in that: described parameter comprises following one or more: network element number, network congestion coefficient, network traffics, capacity utilization, CPU usage, average heart beat cycle, current heart beat cycle, adjusted value.
10. device as claimed in claim 8, it is characterized in that: described sample set update module is updated into former sample set with the mapping relations of described heart beat cycle calculating parameter and heart beat cycle calculated value, or, add former sample set after concentrating the mapping relations similar to the mapping relations that newly obtain to be weighted merging in former sample.
11. device as claimed in claim 8, it is characterized in that: the trigger condition that described heart beat cycle upgrades is: the deviation between described heart beat cycle calculated value and n-1 the heart beat cycle calculated value that obtains before is all greater than first threshold, and the deviation of the mean value of described heart beat cycle calculated value and n-1 heart beat cycle calculated value obtaining before is greater than Second Threshold.
12. device as claimed in claim 8 is characterized in that: the trigger condition that the trigger condition of described calculating heart beat cycle, the trigger condition of self study or described heart beat cycle upgrade is clocked flip or Event triggered.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210037272XA CN103139008A (en) | 2011-11-23 | 2012-02-17 | Self-adaption method and device capable of detecting message heartbeat period |
PCT/CN2012/077723 WO2013075489A1 (en) | 2011-11-23 | 2012-06-28 | Adaptive method and device for detecting heartbeat period of message |
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110376348.7 | 2011-11-23 | ||
CN201110376348 | 2011-11-23 | ||
CN201210037272XA CN103139008A (en) | 2011-11-23 | 2012-02-17 | Self-adaption method and device capable of detecting message heartbeat period |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103139008A true CN103139008A (en) | 2013-06-05 |
Family
ID=48469070
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210037272XA Pending CN103139008A (en) | 2011-11-23 | 2012-02-17 | Self-adaption method and device capable of detecting message heartbeat period |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN103139008A (en) |
WO (1) | WO2013075489A1 (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103532805A (en) * | 2013-11-06 | 2014-01-22 | 中国联合网络通信集团有限公司 | Sending method and device for periodic messages of home terminal |
CN107078925A (en) * | 2014-10-29 | 2017-08-18 | 华为技术有限公司 | The method to set up and terminal of a kind of heart beat cycle |
CN109548066A (en) * | 2018-12-27 | 2019-03-29 | 郑州天迈科技股份有限公司 | A kind of wireless communication link condition detection method |
CN110049532A (en) * | 2019-03-12 | 2019-07-23 | 上海感悟通信科技有限公司 | A kind of heartbeat method, controlled node and control node for wireless self-networking |
CN110417565A (en) * | 2018-04-27 | 2019-11-05 | 华为技术有限公司 | A kind of model update method, apparatus and system |
CN111953569A (en) * | 2020-08-27 | 2020-11-17 | 浪潮电子信息产业股份有限公司 | State information reporting method, device, equipment and medium |
CN112671633A (en) * | 2020-12-01 | 2021-04-16 | 重庆邮电大学 | Binary detection heartbeat interval system and method based on BP neural network prediction |
CN113162813A (en) * | 2020-01-22 | 2021-07-23 | 烽火通信科技股份有限公司 | PTP message loss detection analysis method and device |
WO2023174043A1 (en) * | 2022-03-14 | 2023-09-21 | 华为技术有限公司 | Power line communication system, and in-network state detection method and device |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111698132B (en) * | 2020-06-12 | 2022-03-01 | 北京字节跳动网络技术有限公司 | Method, apparatus, device and medium for controlling heartbeat events in a cluster |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6370656B1 (en) * | 1998-11-19 | 2002-04-09 | Compaq Information Technologies, Group L. P. | Computer system with adaptive heartbeat |
CN1584433A (en) * | 2004-06-04 | 2005-02-23 | 广东科龙电器股份有限公司 | Noise source identifying method for air-conditioner based on nervous network |
CN101115313A (en) * | 2007-08-23 | 2008-01-30 | 中兴通讯股份有限公司 | Soft heartbeat detecting method and system |
CN101408933A (en) * | 2008-05-21 | 2009-04-15 | 浙江师范大学 | Method for recognizing license plate character based on wide gridding characteristic extraction and BP neural network |
CN101504790A (en) * | 2008-11-07 | 2009-08-12 | 清华大学 | Infrared beam type fire disaster smoke detector and detecting method thereof |
US20110071378A1 (en) * | 2009-09-24 | 2011-03-24 | Nellcor Puritan Bennett Llc | Signal Processing Warping Technique |
CN102201937A (en) * | 2011-06-13 | 2011-09-28 | 刘胜利 | Method for detecting Trojan quickly based on heartbeat behavior analysis |
CN102223257A (en) * | 2011-07-25 | 2011-10-19 | 莫雅静 | Heartbeat interval adjusting method, server, client terminal and communication system |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101594268B (en) * | 2009-06-23 | 2011-11-23 | 中兴通讯股份有限公司 | Link detection method |
-
2012
- 2012-02-17 CN CN201210037272XA patent/CN103139008A/en active Pending
- 2012-06-28 WO PCT/CN2012/077723 patent/WO2013075489A1/en active Application Filing
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6370656B1 (en) * | 1998-11-19 | 2002-04-09 | Compaq Information Technologies, Group L. P. | Computer system with adaptive heartbeat |
CN1584433A (en) * | 2004-06-04 | 2005-02-23 | 广东科龙电器股份有限公司 | Noise source identifying method for air-conditioner based on nervous network |
CN101115313A (en) * | 2007-08-23 | 2008-01-30 | 中兴通讯股份有限公司 | Soft heartbeat detecting method and system |
CN101408933A (en) * | 2008-05-21 | 2009-04-15 | 浙江师范大学 | Method for recognizing license plate character based on wide gridding characteristic extraction and BP neural network |
CN101504790A (en) * | 2008-11-07 | 2009-08-12 | 清华大学 | Infrared beam type fire disaster smoke detector and detecting method thereof |
US20110071378A1 (en) * | 2009-09-24 | 2011-03-24 | Nellcor Puritan Bennett Llc | Signal Processing Warping Technique |
CN102201937A (en) * | 2011-06-13 | 2011-09-28 | 刘胜利 | Method for detecting Trojan quickly based on heartbeat behavior analysis |
CN102223257A (en) * | 2011-07-25 | 2011-10-19 | 莫雅静 | Heartbeat interval adjusting method, server, client terminal and communication system |
Non-Patent Citations (3)
Title |
---|
李季等: "基于动量因子技术的BP神经网络改进算法及应用", 《电气应用》 * |
赵海军: "新一代互联网服务及故障检测若干关键技术的研究", 《中国博士学位论文全文数据库》 * |
郭巧惠: "基于小波神经网络的心电数据压缩方法研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103532805A (en) * | 2013-11-06 | 2014-01-22 | 中国联合网络通信集团有限公司 | Sending method and device for periodic messages of home terminal |
CN107078925A (en) * | 2014-10-29 | 2017-08-18 | 华为技术有限公司 | The method to set up and terminal of a kind of heart beat cycle |
US10355963B2 (en) | 2014-10-29 | 2019-07-16 | Huawei Technologies Co., Ltd. | Heartbeat period setting method, and terminal |
CN110417565A (en) * | 2018-04-27 | 2019-11-05 | 华为技术有限公司 | A kind of model update method, apparatus and system |
US11451452B2 (en) | 2018-04-27 | 2022-09-20 | Huawei Technologies Co., Ltd. | Model update method and apparatus, and system |
CN110417565B (en) * | 2018-04-27 | 2021-01-29 | 华为技术有限公司 | Model updating method, device and system |
CN109548066A (en) * | 2018-12-27 | 2019-03-29 | 郑州天迈科技股份有限公司 | A kind of wireless communication link condition detection method |
CN109548066B (en) * | 2018-12-27 | 2022-04-22 | 郑州天迈科技股份有限公司 | Wireless communication link state detection method |
CN110049532A (en) * | 2019-03-12 | 2019-07-23 | 上海感悟通信科技有限公司 | A kind of heartbeat method, controlled node and control node for wireless self-networking |
CN110049532B (en) * | 2019-03-12 | 2021-04-30 | 上海感悟通信科技有限公司 | Heartbeat method, controlled node and control node for wireless ad hoc network |
CN113162813B (en) * | 2020-01-22 | 2022-06-03 | 烽火通信科技股份有限公司 | PTP message loss detection analysis method and device |
CN113162813A (en) * | 2020-01-22 | 2021-07-23 | 烽火通信科技股份有限公司 | PTP message loss detection analysis method and device |
CN111953569B (en) * | 2020-08-27 | 2022-04-29 | 浪潮电子信息产业股份有限公司 | State information reporting method, device, equipment and medium |
CN111953569A (en) * | 2020-08-27 | 2020-11-17 | 浪潮电子信息产业股份有限公司 | State information reporting method, device, equipment and medium |
CN112671633A (en) * | 2020-12-01 | 2021-04-16 | 重庆邮电大学 | Binary detection heartbeat interval system and method based on BP neural network prediction |
CN112671633B (en) * | 2020-12-01 | 2022-08-23 | 重庆邮电大学 | Binary detection heartbeat interval system and method based on BP neural network prediction |
WO2023174043A1 (en) * | 2022-03-14 | 2023-09-21 | 华为技术有限公司 | Power line communication system, and in-network state detection method and device |
Also Published As
Publication number | Publication date |
---|---|
WO2013075489A1 (en) | 2013-05-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103139008A (en) | Self-adaption method and device capable of detecting message heartbeat period | |
CN107231436B (en) | Method and device for scheduling service | |
Ngoduy et al. | Langevin method for a continuous stochastic car-following model and its stability conditions | |
Moreira-Matias et al. | An online learning approach to eliminate Bus Bunching in real-time | |
JP6275868B2 (en) | Neural watchdog | |
CN111444009A (en) | Resource allocation method and device based on deep reinforcement learning | |
CN109815332A (en) | Loss function optimization method, device, computer equipment and storage medium | |
US9064215B2 (en) | Learning spike timing precision | |
CN105989441A (en) | Model parameter adjustment method and device | |
CN108520325A (en) | A kind of integral life prediction technique based on acceleration degraded data under changeable environment | |
CN103489034B (en) | Prediction and the method and apparatus diagnosing online ocean current Monitoring Data | |
EP4131094A1 (en) | Prediction method and apparatus, readable medium, and electronic device | |
CN111198799A (en) | Machine room power consumption early warning method, system, terminal and storage medium based on LSTM | |
CN116055406B (en) | Training method and device for congestion window prediction model | |
CN115686846B (en) | Container cluster online deployment method integrating graph neural network and reinforcement learning in edge calculation | |
CN115085202A (en) | Power grid multi-region intelligent power collaborative optimization method, device, equipment and medium | |
CN114372581A (en) | Block chain-based federal learning method and device and computer equipment | |
CN115829297A (en) | Method and device for generating work package of assembly type building, terminal and storage medium | |
CN103106331A (en) | Photo-etching line width intelligence forecasting method based on dimension-reduction and quantity-increment-type extreme learning machine | |
CN104731709A (en) | Software defect predicting method based on JCUDASA_BP algorithm | |
CN104537224B (en) | Multi-state System Reliability analysis method and system based on adaptive learning algorithm | |
Ma et al. | Comparing paradigms for strategy learning of route choice with traffic information under uncertainty | |
CN117149410A (en) | AI intelligent model based training, scheduling, commanding and monitoring system | |
CN102801581B (en) | Method for predicting WEB service connection success rate | |
CN111092755B (en) | Edge service migration simulation method based on resource occupation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130605 |