CN113740666B - Method for positioning root fault of storm alarm in power system of data center - Google Patents

Abstract

The invention provides a method for positioning a fault of an alarm storm root cause of a power system of a data center, which comprises the steps of firstly determining the equipment connection mode of a power distribution system in a park for faults in a large environment of a power distribution network, sequencing and labeling each node equipment, then constructing an incidence matrix of the power distribution network according to the connection mode and the equipment label, then establishing a fault information vector according to alarm information sent by alarm equipment, finally performing matrix multiplication operation by using a topology matrix and the fault vector to obtain a fault judgment matrix, and finally positioning a fault root cause according to the characteristics of the fault judgment matrix.

Description

Method for positioning storm source fault of data center power system alarm

Technical Field

The invention relates to the field of power grid quality detection, in particular to a method for positioning a fault of an alarm storm root cause of a power system of a data center.

Background

With the continuous expansion of the scale and the continuous enhancement of the importance of the intelligent large-scale data center, in the management of a data park power distribution system, a higher requirement is also provided for the processing efficiency of fault alarm information of power distribution equipment in a park, how to efficiently and accurately position the fault equipment, and therefore, the reduction of the number of alarm information analyzed by operation and maintenance personnel also becomes a focus of research attention.

Because the quantity of distribution equipment in the data garden is many, in order to discover the trouble condition of equipment in time, all be equipped with trouble alarm device on every equipment. When a certain power distribution equipment in a park breaks down, a plurality of equipment connected with the power distribution equipment can be directly caused to generate abnormal operation states, so that multi-node fault coupling alarm is caused, an alarm storm is formed, and core data loss and calculation paralysis of a large-scale data center are directly caused. For the problems, the current main processing modes include two modes, namely, the existing fault alarm and the alarm root cause are matched in a machine learning mode, and when an accident occurs next time, judgment is carried out through an expert rule compiled in a model in advance; and secondly, searching the alarm position of the root cause all the time through establishing a causal relationship network and analyzing the causal association of the fault.

The two fault positioning modes have defects, and the problems are that a machine learning mode needs a large amount of data as training samples, and machine learning parameters influencing the judgment accuracy are difficult to determine; the second problem is that because the number of distribution equipment in the park is large and fault association caused by current, voltage, temperature rise and the like is complex, for example, under the condition that the park commercial power is disconnected, the park equipment is powered off, so that voltage measuring points of a large number of equipment send alarms; if the ventilation system fails, the temperature rise of the associated part of the cabinet and the tunnel will increase rapidly, resulting in a large number of associated temperature alarms. Due to the complexity and diversity of these faults, it is difficult to establish a wide range of causal network matrices.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides a method for positioning the root cause fault of an alarm storm of a power system of a data center.

The invention is realized by the following technical scheme:

a method for positioning a fault of a storm source of an alarm of a power system of a data center comprises the following steps:

step 1, sequencing and labeling each node device according to the connection mode of a power distribution system in a park;

step 2, constructing a topology association matrix according to the labels of the node devices and by combining a power distribution network;

step 3, an alarm storm occurs in the park alarm system, and when the quantity of alarm information in the alarm storm is larger than a set value, root cause searching of the alarm information is carried out;

step 4, according to the position of the alarm information, eliminating the alarm information with the distance greater than the set distance;

and 5, extracting parameters of the alarm information, constructing a fault information matrix by combining the state of the equipment, constructing a fault judgment matrix according to the fault information matrix and the association matrix, and positioning the position of the alarm information according to the fault judgment matrix.

Preferably, the nodes in step 2 include a mains supply input node, a power distribution cabinet node and an electric equipment node.

Preferably, the method for constructing the topology association matrix in step 2 is as follows:

the topological incidence matrix is an n multiplied by n square matrix, and a is obtained when current flows from the i node to the j node _ij Taken as 1, a when there is no current connection between the i and j nodes _ij 0 is taken and the elements defining the diagonal are all 1.

Preferably, the expression of the alarm storm is as follows:

wherein, gamma is _s Starting a quantity value, gamma, for the alarm convergence procedure _e For the alarm convergence procedure stop count value, ζ (t) represents the number of alarms generated in a time period of 0 to t.1 and 0 represent whether an alarm storm ψ (t) occurs or not, and the initial value should be set to 0.

Preferably, the parameters of the alarm information include voltage, current, frequency, position, equipment number corresponding to the alarm information, and time when the alarm information is generated.

Preferably, the device state includes a short circuit state and an open circuit state.

Preferably, when a node has a short-circuit fault, the fault information matrix is as follows:

F＝[3 1 1 1 1 1 1 1 1 1 1]

when the node is short-circuited, the fault information matrix is as follows:

F＝[3 3 1 1 1 1 1 1 1 1 1]

preferably, the fault determination matrix is as follows:

Q＝F×A＝[q ₁ q ₂ ]

wherein A is a topological correlation matrix.

Preferably, when historical alarm information exists in the park, high-level information is divided into different alarm storm clusters according to different alarm root causes;

and (4) comparing the alarm storm in the step (3) with the historical alarm cluster, and directly positioning the position of the alarm storm if the historical alarm cluster has the same alarm storm.

Preferably, the method for locating the position of the alarm storm comprises the following steps:

and fitting the alarm storm with the historical alarm cluster, and positioning to the head end equipment of the alarm storm cluster according to the root cause of the alarm information with the same frequency when the alarm storm and the historical alarm cluster have the alarm information with the same frequency.

Compared with the prior art, the invention has the following beneficial technical effects:

the invention provides a method for positioning a fault of a root cause of an alarm storm of a data center power system, which is used for positioning a fault in a large environment of a power distribution network, and comprises the steps of firstly determining the equipment connection mode of the power distribution system in a park, carrying out sequencing labeling on each node equipment, then constructing an association matrix of the power distribution network according to the connection mode and the equipment label, then establishing a fault information vector according to alarm information sent by alarm equipment, finally carrying out matrix multiplication operation by using a topology matrix and the fault vector to obtain a fault judgment matrix, and finally positioning a fault root cause according to the characteristics of the fault judgment matrix. The invention combines the fault diagnosis method (correlation matrix method) of the transmission line of the power distribution network and improves the method on the basis, so that the method can process the alarm storm of the power system of the data center and make up the blank of the existing fault positioning technical scheme. Compared with a frequency pattern method, the method has low requirement on historical alarm data and wider application range; compared with a neural network method, the algorithm complexity is low, the calculation speed is quicker, and the occupied memory is small; compared with the original correlation matrix method, the method can be used for judging two different faults of short circuit and open circuit, and the accuracy degree is higher.

Drawings

FIG. 1 is a diagram of a power distribution network connection of the present invention;

fig. 2 is a flowchart of a method for locating a storm-source-alarming fault according to the present invention.

Detailed Description

The present invention will now be described in further detail with reference to the attached drawings, which are illustrative, but not limiting, of the present invention.

Referring to fig. 1 and 2, a method for positioning a storm source fault of a data center power system alarm comprises the following steps:

step 1: and dividing the historical alarm information acquired by the data center park into different alarm storm clusters according to different alarm root factors.

Because the alarm information caused by the same alarm root is relatively close in time, the alarm information is divided into K clusters by using a K-Means clustering algorithm according to the occurrence time of the alarm information, wherein K is the number of different clusters into which the alarm information is divided, and the initial value of a clustering center is selected by using a roulette algorithm so as to reduce the iteration time of the clustering algorithm.

And (3) when the park is a newly-built park or a park without the historical alarm data, the step is not carried out, and the step 2 is directly executed.

Step 2: and determining the connection mode of the power distribution system and the equipment in the park, and sequencing and labeling each node equipment.

The nodes comprise a mains supply input node, a power distribution cabinet node and an electric equipment node.

And step 3: and constructing a topological incidence matrix according to the labels of the node devices and the power distribution network.

The distribution networks in the park are connected in a radial mode, and in order to represent the connection relation of the distribution networks, incidence matrixes are used for describing the distribution networks. The definition of the elements in the correlation matrix is as follows. The topological correlation matrix is an n x n square matrix, and a is a when the current flows from the i node to the j node _ij Take 1, when there is no current connection between the i and j nodes then a _ij Take 0 and the elements defining the diagonal are all 1.

The topological incidence matrix constructed according to the topological structure is as follows

The above steps are set for the environment of the alarm convergence procedure, and the configuration is required to be completed before the alarm convergence procedure is started. When equipment replacement or supplement occurs in the park, the association matrix A needs to be supplemented according to actual conditions.

And 4, step 4: and carrying out iterative processing on the park alarm information, and searching the root cause of the alarm information when the quantity of the alarm information is greater than a set value.

The current standard for processing the alarm information by the data center refers to the standard of alarm storm in the industry, namely when the number of alarms generated in ten minutes is more than 5 alarms/engineer number, the alarm storm generated by the fault is defined. The concrete formula is as follows:

wherein, gamma is _s =10 alarm convergence procedure start quantity value, Γ _e Where =5 is the alarm convergence procedure stop count value, ζ (t) represents the number of alarms generated in a time period of 0 to t.1 and 0 represent whether an alarm storm ψ (t) occurs or not, and the initial value should be set to 0. When the alarm storm is at t _s When present, the parameters should be:

Ψ(t _s )＝1，Ψ(t _s -1)＝0

when the alarm storm is at t _e At the end, there should be:

Ψ(t _e )＝0，Ψ(t _e -1)＝1

therefore, when the parameter reaches the set value, the alarm convergence procedure is started.

And 5, eliminating the alarm information with the distance greater than the set distance according to the position of the alarm information.

The alarm information is stored in an alarm database, and the false alarms in the database are removed firstly, wherein the false alarms are generally characterized by a small number of alarm storms which do not belong to the same topological range. Therefore, when some alarm information measuring points are far away from a large number of alarm information measuring points on the topological structure, the relation between the far-away alarm information and the alarm storm is almost 0, and the false alarm is separated into an alarm list of the alarm storm in the range according to the difference of the distances of the points on the topological structure.

And 6, extracting parameters of the alarm information, storing the extracted values into a database, and counting the frequency and the sequence of the alarm.

The parameters comprise voltage, current signals, positions, equipment numbers corresponding to the alarm information and time for generating the alarm information.

And when the alarm storm cluster in the step 1 is stored in the park, comparing the alarm storm with a historical alarm cluster, and if the known alarm storm exists, directly positioning an alarm root cause.

And (3) fitting the alarm storm with the historical alarm cluster in the step (1), and when finding that the alarm with the same frequency as the historical alarm cluster is generated, listing the alarm storm as a known alarm storm and directly positioning the root cause to the head end equipment of the alarm storm cluster.

And 7, performing fault positioning analysis on the short circuit fault type and the open circuit fault type according to the voltage and the current of the alarm information. Numbering the different states of the device as

If the voltage and current at the node are both normal values, the value of the node is set to 0.

If the current and voltage of the node are both 0, the value of the node is set to 1.

If the voltage at the node is a normal value but the current is 0, the value of the node is set to 2.

If the current at that node exceeds the threshold but the voltage is 0, then the value of that node is set to 3.

And 8: the fault information vector is made according to the actual short circuit and open circuit conditions as follows:

(1) When a short-circuit fault occurs after the node 1, the fault information matrix is as follows:

F＝[3 1 1 1 1 1 1 1 1 1 1]

the fault determination matrix is as follows:

Q＝F×A＝[q ₁ q ₂ ]

q ₁ ＝[3 3+1 3+1 3+1+1 3+1+1 3+1+1]

q ₂ ＝[3+1+1 3+1+1 3+1+1 3+1+1+1 3+1+1+1]

(2) When a short circuit occurs after the node 2, the fault information matrix is as follows:

F＝[3 3 1 1 1 1 1 1 1 1 1]

the fault determination matrix is as follows:

Q＝F×A＝[q ₁ q ₂ ]

q ₁ ＝[3 3+3 3+1 3+3+1 3+3+1 3+3+1]

q ₂ ＝[3+1+1 3+1+1 3+1+1 3+3+1+1 3+3+1+1]

(3) When a short circuit occurs behind the node 4, the fault information matrix is as follows:

F＝[3 3 1 3 1 1 1 1 1 1 1]

the fault judgment matrix is as follows:

Q＝F×A＝[q ₁ q ₂ ]

q ₁ ＝[3 3+3 3+1 3+3+3 3+3+1 3+3+1]

q ₂ ＝[3+1+1 3+1+1 3+1+1 3+3+3+1 3+3+1+1]

as can be seen from the above calculation, when a short circuit occurs at a node in the power distribution network, a high short-circuit current occurs at the node and the nodes connected in the path from the power supply to the node, and the node closer to the end of the power supply network is farther from the power supply, the more the number of nodes through which the short-circuit current flows, the larger the sum of the factors 3 representing the value of the ith position in the fault determination vector Q, and the number of the added factors 3 representing the position of the node at which the short-circuit occurs. And the number of factors 3 is the largest.

(4) When the node 1 is disconnected, the fault information matrix is as follows:

F＝[2 1 1 1 1 1 1 1 1 1 1]

the fault determination matrix is as follows:

Q＝F×A＝[q ₁ q ₂ ]

q ₁ ＝[2 2+1 2+1 2+1+1 2+1+1 2+1+1]

q ₂ ＝[2+1+1 2+1+1 2+1+1 2+1+1+1 2+1+1+1]

(5) When node 2 is disconnected, the fault information matrix is

F＝[0 2 0 1 1 1 0 0 0 1 1]

The fault determination matrix is as follows:

Q＝F×A＝[0 2 0 2+1 2+1 2+1 0 0 0 2+1+1 2+1+1]

(6) When the node 4 is disconnected, the fault information matrix is as follows:

F＝[0 0 0 2 0 0 0 0 0 1 0]

the fault judgment matrix is as follows:

Q＝F×A＝[0 0 0 2 0 0 0 0 0 2+1 0]

(7) When an open circuit fault occurs behind nodes 10 and 11, then the fault information matrix is as follows:

F＝[0 0 0 2 2 0 0 0 0 2 2]

the fault determination matrix is as follows:

Q＝F×A＝[0 0 0 2 2 0 0 0 0 2+2 2+2]

as can be seen from the above calculation, when an open circuit occurs after a certain node i in the power distribution network, the node and the nodes connected in the path from the node to the end consumer lose electric energy and display a state of no voltage and no current, the closer the node to the head end of the power supply network is to the power supply, the greater the number of nodes affected at the time of the open circuit, the greater the number of nodes represented in the fault judgment vector Q at that time is the result of adding 1 or more numbers 2 to the value of the ith position, and the node with the largest 2 in the fault judgment matrix Q is the faulty node, and the method can have a certain fault tolerance, and if the fault information of the node 2 becomes 0 due to the report omission in (5), that is, the node 2 becomes a faulty node

F＝[0 0 0 1 1 1 0 0 0 1 1]

The fault judgment matrix is as follows:

Q＝F×A＝[0 0 0 0+1 0+1 0+1 0 0 0 0+1+1 0+1+1]

the node value of 4,5 and 6 is displayed and is 0+1, so that the node supplying power for the node 4,5 and 6 is proved to have alarm failure, the node 4,5 and 6 simultaneously loses electric quantity, but the alarm information is not displayed, and the purpose of identifying the failure of alarm is achieved.

And step 9: and (4) processing the incidence matrix and the fault information vector to obtain a fault location, and then shielding non-root alarm. And releasing the determined alarm root cause and the alarm information stored in the step 6 and reporting the alarm root cause and the alarm information to operation and maintenance personnel.

The invention provides a method for positioning a fault of an alarm storm root cause of a power system of a data center, which is used for establishing a topological matrix according to the connection relation of each node in a power distribution network for the fault in the large environment of the power distribution network, then establishing a fault vector by using the voltage and current values measured on each node, then performing matrix multiplication operation by using the topological matrix and the fault vector to obtain a fault judgment matrix, and finally positioning the fault root cause according to the characteristics of the fault judgment matrix.

The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims (6)

1. A method for positioning a storm source fault of a data center power system alarm is characterized by comprising the following steps:

step 1, sequencing and labeling each node device according to the connection mode of a power distribution system in a park;

step 2, constructing a topology incidence matrix according to the labels of the node devices and by combining a power distribution network;

the topological correlation matrix is as follows:

the topological incidence matrix is an n multiplied by n square matrix, and a is obtained when current flows from the i node to the j node _ij Take 1, when there is no current connection between the i and j nodes then a _ij Taking the value as 0, wherein elements of the specified diagonal are all 1;

step 3, an alarm storm occurs in the park alarm system, and when the quantity of alarm information in the alarm storm is larger than a set value, root cause searching of the alarm information is carried out;

the expression of the alarm storm is as follows:

wherein, gamma is _s Starting a quantity value, gamma, for the alarm convergence procedure _e In order to stop the quantity value of the alarm convergence procedure, zeta (t) represents the quantity of alarms generated in the time period from 0 to t, 1 and 0 represent whether an alarm storm occurs, and psi (t) should be set to 0 as an initial value;

step 4, eliminating the alarm information with the distance greater than the set distance according to the position of the alarm information;

step 5, extracting parameters of the alarm information, constructing a fault information matrix by combining the state of the equipment, constructing a fault judgment matrix according to the fault information matrix and the correlation matrix, and positioning the position of the alarm information according to the fault judgment matrix;

when the equipment node has short-circuit fault, the fault information matrix is as follows:

F＝[3 1 1 1 1 1 1 1 1 1 1]

the fault determination matrix is as follows:

Q＝F×A＝[q ₁ q ₂ ]

a is a topological incidence matrix;

when the equipment node is disconnected, the fault information matrix is

F＝[0 2 0 1 1 1 0 0 0 1 1]

The fault determination matrix is as follows:

Q＝F×A＝[0 2 0 2+1 2+1 2+1 0 0 0 2+1+1 2+1+1]。

2. the method according to claim 1, wherein the nodes in step 2 include a utility power input node, a power distribution cabinet node and a power-consuming equipment node.

3. The method for locating the storm source fault in the data center power system according to claim 1, wherein the parameters of the alarm information include voltage, current, frequency, location, equipment number corresponding to the alarm information, and time when the alarm information is generated.

4. The method of claim 3, wherein the equipment status comprises a short circuit status and an open circuit status.

5. The method for locating the fault of the alarm storm source of the data center power system according to claim 1, wherein when historical alarm information exists in a park, high-level information is divided into different alarm storm clusters according to different alarm root causes;

and (3) comparing the alarm storm in the step (3) with the historical alarm cluster, and directly positioning the position of the alarm storm if the historical alarm cluster has the same alarm storm.

6. The method for locating the fault of the source of the storm alarm in the power system of the data center according to claim 5, wherein the method for locating the position of the storm alarm comprises the following steps:

and fitting the alarm storm with the historical alarm cluster, and positioning to the head end equipment of the alarm storm cluster according to the root cause of the alarm information with the same frequency when the alarm storm and the historical alarm cluster have the alarm information with the same frequency.

Images