CN106452877B - The method of fault location of power information network - Google Patents

Abstract

The invention discloses a fault location method for an electric power information network. The Bayesian network model is used to describe the relationship between the detections in the candidate detection set and the nodes in the electric power information network, and the information gain of a single detection is combined with the important nodes in the single detection path. The number is used as the detection value to measure the diagnostic ability of each detection in the candidate detection set. Select the detection with the largest detection value from the candidate detection set to form a fault location set, and obtain the most likely state information of the power information network from the returned detection results. To locate the faulty node. The method of the invention is simple in theory, the directed edge between the information node and the detection makes the fault location process clearer, defines a new detection value as the criterion for selecting the fault location set, and improves the accuracy of the fault location; the conditions of the Bayesian network are used. Independence, it is divided into several subnets, and only the detections of the same subnet are updated when the detection value is updated, which reduces the detection selection time and improves the timeliness of fault location.

Description

Fault location method of power information network

technical field

The invention relates to the technical field of electric power information, in particular to a fault location method for an electric power information network.

Background technique

In the smart grid, the power information network carries various power production services and is an important channel to ensure power communication services. However, the huge network structure causes faults to occur frequently, and it is difficult to locate the fault after the fault occurs. If the fault cannot be quickly and accurately located, it will cause a power accident. Traditional network fault management mostly uses northbound interfaces to passively collect alarm information from devices, and establish an alarm-fault correlation model to analyze the root cause of the fault. However, passively waiting for alarms will lead to poor fault location timeliness, and once the alarm information is false, lost, or redundant This phenomenon cannot guarantee the accuracy of fault location. In order to make up for the shortcomings of traditional network management, active measurement is gradually applied to the operation and maintenance of power information network, which has important application significance for improving the management and operation level of power information network.

At home and abroad, there are relatively many research results on using network detection in public networks to improve network operation and maintenance level. Most of them are actively sending detection packets to obtain network and service quality data, and find network hidden dangers to locate faults. Typical application scenarios include Fault detection, location and identification, etc. Active detection aims to select the best detection set, generally including pre-selection and interactive selection. The pre-selection method has a simple calculation method, but the execution process is inefficient, and the accuracy is relatively low due to poor real-time performance. The interactive probe set selection method uses the calculation-and-send method to reduce the number of executed probes and make the network load lower.

The patent number CN103840967A is entitled "A method for fault location in electric power communication network", which provides a method for fault location in electric power communication network. Modeling, introducing fault impact weights, using full probability and Bayesian thinking under the bipartite graph model, converting the prior fault probability into conditional probability, calculating the fault impact degree, and adding credible parameters to control the impact of suspected faults, combining the coverage degree and Contribution to select a completely plausible explanation for the symptoms that occurred. The theory involved in this method is too cumbersome and complicated to implement. It needs to analyze the topology of the entire network, and the collection of symptoms is too passive, and the positioning timeliness is low.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is that the existing method is complicated to implement, and cannot satisfy high fault location timeliness while ensuring the fault location accuracy.

The present invention provides a method for locating faults in a power information network, comprising the following steps:

S1. Randomly select network nodes in the power information network to deploy probes, obtain all probes from all probes to all network nodes, and form an available probe set T _all ;

S2. Select probes that can cover all nodes in the network from the available probe set T _all , and periodically send probe packets to the information network. These probes constitute a fault detection set T _obs , and a candidate probe set T _sta =T _all -T is obtained at the same time _obs , the fault location set T _dia consisting of probes capable of locating faults is empty at this time;

S3. Establish a Bayesian network model between the detection t _j in the candidate detection set T _sta and the network node x _i , where j=1,2,...,m, i=1,2,..., n, m is the number of probes in the candidate probe set T _sta , n is the number of network nodes;

S4. Divide the Bayesian network model established in S3 into several sub-networks according to the returned results of all detections in the fault detection set T _obs ;

P(X,T)是X和T的联合概率，P(X|T)是已知T情况下X的条件概率，X＝(S(x₁),S(x₂),...,S(x_n))为整体网络节点的状态向量，T＝(S(t'₁),S(t'₂),...,S(t'_h))为当前已发送探测包的探测的状态向量，t'₁,t'₂,...,t'_h为当前已发送探测包的探测，h为当前已发送探测包的探测的数目，α和β为预设权值；S5. Calculate the information gain G(t _j )=H(X|T)-H(X|T∪{S(t _j )}) of each detection in the candidate detection set T _sta and the number of important nodes of each detection Number N _Im (t _j ), obtain the detection value V(t _j )=αG(t _j )+βN _Im (t _j ) of each detection, and classify the detections in the candidate detection set T _sta according to the detection value from large to Small sorting, take t _max as the detection with the largest detection value in the current candidate detection set T _sta ; where S(t _j ) is the state of detection t _j ,

P(X,T) is the joint probability of X and T, P(X|T) is the conditional probability of X given T, X=(S(x ₁ ),S(x ₂ ),..., S(x _n )) is the state vector of the overall network node, T=(S(t' ₁ ), S(t' ₂ ),..., S(t' _h )) is the detection of the currently sent detection packet The state vector of , t' ₁ , t' ₂ ,...,t' _h is the detection of the currently sent detection packet, h is the number of detections of the currently sent detection packet, α and β are preset weights;

S6. If the cost of the fault location set C(T _dia )≥B, go to S9, otherwise the detection t _max sends a detection packet to the information network to obtain the current state of the detection S(t _max ) of the information network, and at the same time the detection t _max is added to the fault location set T _dia and deleted from the candidate detection set T _sta . If the candidate detection set T _sta is empty at this time, go to S9; where B is the preset detection cost threshold;

S7, take t' _max as the detection with the largest detection value in the current candidate detection set T _sta , if t' _max and t _max are in the same subnet, then recalculate the detection value V(t' _max ) of the detection t' _max , Otherwise go directly to S8;

为候选探测集T_sta中排在探测t'_max的后一位探测，若

则令t_max为t'_max，转到S6，否则判断

与t_max是否在同一子网内，若不在同一子网，则比较t'_max与

的探测价值，若在同一子网，则重新计算

的探测价值

再比较t'_max与

的探测价值；如果

则令t'_max为

转到S8，否则令t_max为t'_max，转到S6；S8, take

is the next probe in the candidate probe set T _sta in the probe t' _max , if

Then let t _max be t' _max , go to S6, otherwise judge

Whether it is in the same subnet as t _max , if not, compare t' _max with

The detection value of , if in the same subnet, recalculate

detection value

Then compare t' _max with

The detection value of ; if

Then let _t'max be

Go to S8, otherwise let _tmax be _t'max , go to S6;

S9. Calculate an instance x ^* =argmaxP(X|ST) of the overall network node state X, and obtain the most probable state vector x ^* of the overall network node, thereby locating the faulty node; where ST=(S(T ₁ ), S(T ₂ ),...) is the state vector of all probes in the current fault detection set T _obs and fault location set T _dia .

Further, step S4 specifically includes:

和覆盖该网络节点x_i的所有探测返回结果一致的比例

若同时满足a_i＞λ和b_i＞θ，则定义该网络节点x_i为近似可观测节点，利用D-分离方法将S3建立的贝叶斯网络模型划分成若干个相互独立的小规模贝叶斯子网络；其中c'为可用探测集T_all中覆盖网络节点x_i的探测，c为故障检测集T_obs中覆盖网络节点x_i的探测，

e为故障检测集T_obs中覆盖网络节点x_i且状态为k的探测，|·|表示求集合包含元素的个数，λ和θ为预设系数。Calculate the number of times the network node x _i should be probed according to the return results of all probes in the fault detection set T _obs

The proportion that is consistent with the return results of all probes covering the network node x _i

If both a _i > λ and b _i > θ are satisfied, the network node _xi is defined as an approximately observable node, and the Bayesian network model established by S3 is divided into several independent small-scale Bayesian network models using the D-separation method. Yeasian sub-network; where c' is the detection of the overlay network node _xi in the available detection set T _all , c is the detection of the overlay network node _xi in the fault detection set T _obs ,

e is the detection that covers the network node x _i and the state is k in the fault detection set T _obs , |·| represents the number of elements contained in the set, and λ and θ are preset coefficients.

Further, the cost C(T _dia ) of the fault location set T _dia in step S6 is the number of probes in the fault location set T _dia .

The beneficial effects of the present invention: (1) The method of the present invention uses a Bayesian network model to describe the relationship between the probes in the candidate probe set and the nodes in the power information network, which effectively simplifies the problem, and the directed relationship between the information nodes and the probes The edge makes the fault location process clearer; (2) Combine the information gain of a single detection and the number of important nodes in a single detection path as the detection value to measure the diagnostic ability of each detection in the candidate detection set, and select the detection value from the candidate detection set. The largest detection constitutes a fault location set, which improves the accuracy of fault location; (3) Using the conditional independence of the Bayesian network, it is divided into several subnets, and only the detections of the same subnet are updated when the detection value is updated, reducing the number of detections. The selection time improves the timeliness of fault location; (4) the sub-model is represented in the process of detection value update, which reduces the number of update detections, speeds up the detection selection time, and improves the timeliness of fault location.

Description of drawings

Figure 1 is a flow chart of the method of the present invention.

FIG. 2 is a comparison diagram of the fault location accuracy of the method of the present invention and the BPEA method.

Figure 3 is a time-sensitive comparison diagram of the method of the present invention and the BPEA method.

Detailed ways

The concepts of several probe sets used in the present invention are as follows:

1) Available probe set T _all : randomly select several nodes in the information network to deploy probes, each probe can send active probe packets to any network node in the information network, which is called a probe, according to the results returned by the probe It is possible to infer whether the state of the network nodes is good or bad. A set consisting of all probes from all probes to all network nodes is an available probe set T _all , and the probe sets described below are all subsets of this set.

2) Fault detection set T _obs : To ensure the reliable operation of the power information network, we cannot passively wait for users to issue fault reports, and we need to monitor the network operation in real time. Therefore, we need to select some probes from the available probe set T _all to periodically send probes. When the packet is sent to the network, the selected and sent probes must be able to cover all nodes in the network. The set composed of these probes is the fault detection set T _obs .

3) Candidate detection set T _sta : a set consisting of the detections in the available detection set T _all that do not belong to the fault detection set T _obs , that is, T _sta =T _all -T _obs .

4) Fault locating set T _dia : The fault locating set is a set consisting of probes selected from the candidate probe set T _sta to achieve the purpose of locating faults when the return results of each probe in the fault detection set T _obs are known.

A method for locating faults in a power information network provided by the present invention includes the following steps:

S1. Randomly select network nodes in the power information network to deploy probes to obtain an available probe set T _all ;

S2. Select a probe that can cover all nodes of the network from the available probe set T _all to form a fault detection set T _obs , and obtain a candidate probe set T _sta at the same time;

S3. Establish a Bayesian network model between the detection t _j in the candidate detection set T _sta and the network node x _i , where j=1,2,...,m, i=1,2,..., n, m is the number of probes in the candidate probe set T _sta , n is the number of network nodes;

S4. Divide the Bayesian network model established in S3 into several sub-networks according to the returned results of all detections in the fault detection set T _obs ;

P(X,T)是X和T的联合概率，P(X|T)是已知T情况下X的条件概率，X＝(S(x₁),S(x₂),...,S(x_n))为整体网络节点的状态向量，T＝(S(t'₁),S(t'₂),...,S(t'_h))为当前已发送探测包的探测的状态向量，t'₁,t'₂,...,t'_h为当前已发送探测包的探测，h为当前已发送探测包的探测的数目，α和β为预设权值；S5. Calculate the information gain G(t _j )=H(X|T)-H(X|T∪{S(t _j )}) of each detection in the candidate detection set T _sta and the number of important nodes of each detection Number N _Im (t _j ), obtain the detection value V(t _j )=αG(t _j )+βN _Im (t _j ) of each detection, and classify the detections in the candidate detection set T _sta according to the detection value from large to Small sorting, take t _max as the detection with the largest detection value in the current candidate detection set T _sta ; where S(t _j ) is the state of detection t _j ,

P(X,T) is the joint probability of X and T, P(X|T) is the conditional probability of X given T, X=(S(x ₁ ),S(x ₂ ),..., S(x _n )) is the state vector of the overall network node, T=(S(t' ₁ ), S(t' ₂ ),..., S(t' _h )) is the detection of the currently sent detection packet The state vector of , t' ₁ , t' ₂ ,...,t' _h is the detection of the currently sent detection packet, h is the number of detections of the currently sent detection packet, α and β are preset weights;

S6. If the cost of the fault location set C(T _dia )≥B, go to S9, otherwise the detection t _max sends a detection packet to the information network to obtain the current state of the detection S(t _max ) of the information network, and at the same time the detection t _max is added to the fault location set T _dia and deleted from the candidate detection set T _sta . If the candidate detection set T _sta is empty at this time, go to S9; where B is the preset detection cost threshold;

S7, take t' _max as the detection with the largest detection value in the current candidate detection set T _sta , if t' _max and t _max are in the same subnet, then recalculate the detection value V(t' _max ) of the detection t' _max , Otherwise go directly to S8;

为候选探测集T_sta中排在探测t'_max的后一位探测，若

则令t_max为t'_max，转到S6，否则判断

与t_max是否在同一子网内，若不在同一子网，则比较t'_max与

的探测价值，若在同一子网，则重新计算

的探测价值

再比较t'_max与

的探测价值；如果

则令t'_max为

转到S8，否则令t_max为t'_max，转到S6；S8, take

is the next probe in the candidate probe set T _sta in the probe t' _max , if

Then let t _max be t' _max , go to S6, otherwise judge

Whether it is in the same subnet as t _max , if not, compare t' _max with

The detection value of , if in the same subnet, recalculate

detection value

Then compare t' _max with

The detection value of ; if

Then let _t'max be

Go to S8, otherwise let _tmax be _t'max , go to S6;

S9. Calculate an instance x ^* =argmaxP(X|ST) of the overall network node state X, and obtain the most probable state vector x ^* of the overall network node, thereby locating the faulty node; where ST=(S(T ₁ ), S(T ₂ ),...) is the state vector of all probes in the current fault detection set T _obs and fault location set T _dia .

In the Bayesian network model established in step S3, the nodes x _i (i=1, 2,...,n) represent the nodes in the information network, and the nodes t _j (j=1, 2,...,m) represent the For a probe in the candidate probe set T _sta , the directed edge between the network node and the probe indicates that the probe's probe path includes this node. Node x _i has two states, S(x _i )=0 means that the node is not faulty, S(x _i )=1 means that the node is faulty, probe the state S(t _j ) of t _j and all the networks that point to it The state of the node is related. As long as the state of a network node it points to is 0, the state of the probe is S(t _j ), and any node failure on the probe path will cause the probe to fail, that is, as long as the probe t _j points to the network One of the nodes has state S(x _i )=0, and this probe has state S(t _j )=0. Abstracting probes and network nodes into nodes in the Bayesian network model effectively simplifies the problem, and the directed edges between nodes and probes make the fault location process clearer.

Step S4 divides the sub-network. Specifically, according to the results returned by all detections in the fault detection set T _obs , the “approximately observable node” is defined by the following formula, indicating that the state of the network node x _i has been approximately determined:

in,

|·| represents the number of elements contained in the set, λ and θ are preset values, λ represents the minimum number of times that approximately observable nodes should be detected, θ represents the at least proportion of consistent results in the detection results, 0≤θ≤ 1. If the network node x _i satisfies both equations (1) and (2), it is considered to be approximately observable, and the detections it points to satisfy the conditional independence. Using the D-separation principle, the large-scale Bayesian The network is converted into several independent sub-networks, and only the remaining probes in the same sub-network need to be updated when updating the detection value of candidate probes.

Steps S5 to S8 are the steps of selecting the fault location set T _dia from the candidate detection set T _sta .

Specifically, in order to measure the pros and cons of the detection in the candidate detection set T _sta , the information entropy is introduced to quantify the uncertainty of the state of the power information network, and the vector X=(S(x ₁ ), S(x ₂ ),..., S(x _n )) represents the state of the entire network node, where n is the number of network nodes, and the definition vector T=(S(t' ₁ ), S(t' ₂ ),...,S(t' _h ) ) represents the state of the probe that has sent the probe packet, where h is the number of probes that have sent the probe packet at present, then after obtaining the probe return result, the uncertainty of the information network state is expressed as:

In formula (4), P(X, T) is the joint probability of X and T, and P(X|T) is the conditional probability of X when T is known. Use the information gain G(t) to quantify the pros and cons of a single probe t:

G(t)=H(X|T)-H(X|T∪{S(t)}) (5)

In order to ensure that more important network nodes are preferentially detected, the importance of network nodes needs to be considered when comparing the pros and cons of two probes. The method of the present invention defines V(t) to represent the detection value of a single probe:

V(t)=αG(t)+βN _Im (t) (6)

In the formula (6), N _Im (t) represents the number of important nodes in the detection path of detection t. In this embodiment, the important nodes are the core layer network nodes in the center of the network; α and β are weights, adjusting α, The value of β can balance the information gain and the weight of important nodes in the detection value to meet the needs of different information network structures.

Using equations (4) to (6), the detection value of each detection in the candidate detection set T _sta is calculated. From the candidate detection set T _sta , the detection with the greatest detection value is continuously selected and added to the set T ^* . When C(T ^* )=B, the selection ends, and the fault location set T _dia =T ^* is obtained. Among them, C(T ^* ) represents the cost of the set T ^* , which is generally the price of the probe or the time required to complete the detection, etc. In this embodiment, C(T ^* ) is the number of probes in the set T ^* , and B is based on the actual The detection cost threshold set by the situation.

In the process of selecting the detection with the largest detection value, when a new detection result is observed, the detection value of all detections in the remaining candidate detection set becomes smaller after the update, so only the detection with the current maximum detection value needs to be updated. After the update, the detection value of this detection is greater than that of other detections before the update, so it can be determined that this detection is the best choice for the next round without updating the other detections, which greatly saves time.

Specifically, in step S9, after the fault location set T _dia is obtained, the state of the nodes in the power information network is analyzed according to the fault detection set T _obs and the results returned by detection in the fault location set T _dia , that is, to find:

x ^* =argmaxP(X|ST) (8)

where ST=(S(T ₁ ), S(T ₂ ), . . . ) is the set of states returned by sent probes, and x ^* is an instance of network state X. If the status of any network node in x ^* is 0, it indicates that the network node is faulty. Using the fault detection set to comprehensively judge the status of the information network nodes can reduce the burden of the fault location set, and can obtain better fault location results when the detection cost threshold is constant.

In Fig. 2 and Fig. 3, IPCA is the method of the present invention, BPEA is the traditional multi-fault network detection and selection algorithm, Fig. 2 shows the comparison diagram of the fault location accuracy between the method of the present invention and the BPEA method, and Fig. 3 shows the method of the present invention and the Timeliness comparison chart of BPEA method. As can be seen from FIG. 2 , as the number of network nodes increases, the accuracy of the method of the present invention is approximately the same as that of the BPEA method, indicating that the method of the present invention ensures the accuracy of fault location. As can be seen from FIG. 3 , with the increase of the number of nodes, the method of the present invention has obvious improvement effect on time, and improves the time efficiency of fault diagnosis.

Claims (3)

1. A power information network fault positioning method is characterized by comprising the following steps:

s1, randomly selecting network nodes in the power information network to deploy probes, obtaining all probes from all the probes to all the network nodes, and forming an available probe set T_all；

S2, from available detection set T_allSelects probes covering all nodes of the network, periodically sends probe packets to the information network, these probes form a fault detection set T_obsWhile obtaining a candidate probe set T_sta＝T_all-T_obsNow fault location set T consisting of detections capable of locating faults_diaIs empty;

s3, establishing a candidate detection set T_staDetection of (1)_jAnd network node x_iWhere j 1,2, is 1,2, n, m is a candidate probe set T_staThe number of probes, n is the number of network nodes;

s4, detecting set T according to faults_obsDividing the Bayesian network model established in S3 into a plurality of subnets according to the return results of all the probes;

s5, calculating candidate detection set T_staGain of information per probe G (t)_j)＝H(X|T)-H(X|T∪{S(t_j) }) and the number of important nodes N detected each_Im(t_j) To obtain the detection value V (t) of each detection_j)＝αG(t_j)+βN_Im(t_j) Candidate probe set T_staThe middle detection is ranked from big to small according to the detection value, and t is taken_maxFor the current candidate probe set T_staDetecting with the maximum detection value; wherein S (t)_j) To detect t_jIn the state of (a) to (b),

p (X, T) is the joint probability of X and T, P(X | T) is the conditional probability of X given T, X ═ S (X)₁),S(x₂),...,S(x_n) Is the state vector of the entire network node, T ═ S (T'₁),S(t'₂),...,S(t'_h) Is a state vector of probes, t 'for currently transmitted probe packets'₁,t'₂,...,t'_hH is the number of the detection of the currently sent detection packet, and alpha and beta are preset weights;

s6 cost C (T) of fault location set_dia) If B is equal to or more than B, go to S9, otherwise detect t_maxSending the detection packet to the information network to obtain the current state S (t) of the detection in the information network_max) While t will be detected_maxAdd fault location set T_diaAnd from the candidate probe set T_staIn the case of deletion, if the candidate probe set T is in the process_staIf empty, go to S9; wherein B is a preset detection cost threshold;

s7, taking t'_maxFor the current candidate probe set T_staDetection of medium detection value, if t'_maxAnd t_maxWithin the same subnet, then probe t 'is recalculated'_maxIs detected as a value V (t'_max) Otherwise go directly to S8;

s8, getting

As candidate probe set T_staMiddle row at probe t'_maxThe latter bit is detected if

Let t_maxIs t'_maxGo to S6, otherwise, judge

And t_maxIf the two sub-networks are not in the same subnet, t 'is compared'_maxAnd

if the same is detectedSubnet, then recalculate

Detection value of

Then t 'is compared'_maxAnd

the detection value of (2); if it is not

Then t 'is'_maxIs composed of

Go to S8, otherwise let t_maxIs t'_maxGo to S6;

s9, calculating an example X of the overall network node state X^*Get the most likely state vector X of the whole network node (X | ST) ═ argmaxP (X | ST)^*Thereby locating the faulty node; wherein ST ═ S (T)₁),S(T₂) ,..) is the current fault detection set T_obsAnd fault location set T_diaAll detected state vectors.

2. The method according to claim 1, wherein the step S4 specifically includes:

according to fault detection set T_obsIn the return results of all probes, compute network node x_iNumber of times to be detected

And overlay the network node x_iAll probes of (2) return a consistent ratio of results

If a is satisfied simultaneously_iλ and b_i＞θ，Then the network node x is defined_iIn order to approximate observable nodes, a Bayesian network model established in S3 is divided into a plurality of mutually independent small-scale Bayesian sub-networks by using a D-separation method; wherein c' is the available probe set T_allMedium coverage network node x_iC is a fault detection set T_obsMedium coverage network node x_iThe detection of (a) is performed,

k is 0,1, i is 1,2_obsMedium coverage network node x_iAnd the state is the detection of k, | · | represents the number of elements included in the solution set, and λ and θ are preset coefficients.

3. Method for locating faults in an electrical power information network according to claim 1 or 2, characterized in that in step S6, a fault location set T is set_diaCost C (T)_dia) Set T for fault location_diaThe number of probes in.

Images