CN104459378B - A kind of intelligent substation method for diagnosing faults - Google Patents

Abstract

A kind of intelligent substation method for diagnosing faults, is carried out by following 10 steps：Step 1：Gather each variable of failure hypothesis vector；Step 2：Networking modeling analysis；Step 3：Sample matrix is set up to be estimated；Step 4：Calculate the corresponding target function value of failure hypothesis；Step 5：Judge whether to meet end condition；Step 6：Obtain correct failure hypothesis；Step 7：The alarm state received in practice；Step 8：Comparison information；Step 9：The alarm failing to report or report by mistake, the breaker of malfunction tripping；Step 10：Defect elimination is carried out to failure, recovers normal operation.Present invention reduces the management cost of power system, backward, outmoded artificial accident analysis and tupe are changed, the man power and material of intelligent substation operation, the production cost reduced is greatly reduced, operating efficiency is improved, is that intelligent substation construction and operation provide technical support.

Description

A kind of intelligent substation method for diagnosing faults

Technical field

The invention belongs to electricity field, more particularly to a kind of intelligent substation method for diagnosing faults.

Background technology

Intelligent substation is to use advanced, reliable, integrated and environmentally friendly smart machine, with whole station information digitalization, communication Platform network, information sharing are standardized as basic demand, are automatically performed information gathering, measurement, control, protection, metering and examine The basic function such as survey, meanwhile, possess support power network automatically control in real time, Intelligent adjustment, on-line analysis decision-making and collaboration interaction etc. The transformer station of Premium Features.Intelligent substation mainly includes intelligent high-voltage equipment and transformer station's integration information platform two parts.Intelligence Energy high-tension apparatus mainly includes intelligent transformer, Intelligent high voltage switch device, electronic mutual inductor etc..Intelligent transformer and control System is connected by telecommunication optical fiber, and transformer state parameter and service data can be grasped in time.When the method for operation changes, Voltage, power situation of the equipment according to system, decide whether to adjust tap；When equipment goes wrong, early warning can be sent simultaneously State parameter etc. is provided, operational management cost is reduced to a certain extent, hidden danger is reduced, transformer station high-voltage side bus reliability is improved.Intelligence Energy high-tension switch gear is switchgear and control device with superior performance, equipped with electronic equipment, sensor and actuator, With monitoring and diagnostic function.Electronic mutual inductor refers to pure optical fiber mutual inductor, magneto-optic glass transformer etc., can effectively overcome biography The shortcoming of system electromagnetic transformer.Transformer station's integration information platform function has two, and one is the horizontal information sharing of system, main table Now information acquisition is unitized for various upper layer applications in management system；Two be the standardization of the longitudinal information of system, main table The transparence now supported for each layer to its upper layer application.

Intelligent substation contacts hinge as control centre and transformer station, directly facing the O＆M of system operation and equipment Management, integrate commander, scheduling, monitoring, operate, its operation and intelligent decision-making level to raising power system self-healing ability And the reliability of power supply plays an important role.But the research of current alarm processing, fault diagnosis and accident treatment is mostly concentrated on Electric power system dispatching center or transformer station, the research of the power system alarm processing such as based on analytic modell analytical model, based on artificial god Research through metanetwork power system failure diagnostic, the research of Stackelberg Decision for Power System Restoration based on many agent.On State research and certain breakthrough is obtained in terms of theoretical research, some are even applied in systems in practice.And base Extremely lack in the method for diagnosing faults and the theory of accident treatment of intelligent substation and its application study.

On the other hand, when occurring accident, intelligent substation will often receive alarm up to hundreds of, person on duty in tens seconds Found by the Substation Bus Arrangement of monitoring system and determine accident impact scope, it is determined that and recording the breaker disconnected and protecting dynamic Make information, and be distributed to all departments, reporting scheduling, while carrying out crash analysis；Teams and groups operations staff, according on all messages Information is passed, tripping operation, protection act situation rule of thumb analyze cause of accident substantially, urgent referral patrols fibrillar center personnel on site Check.Because quantities of messages is excessive, accident treatment needs to complete within short a few minutes again, and the electric power that monitoring personnel relies on is known Know and experience is difficult timely and effective failure correctly to be judged.

The content of the invention

In order to solve the above problems, the present invention provides a kind of intelligent substation method for diagnosing faults, the invention is characterised in that, Carried out by following 10 steps：

Step 1：Gather each variable of failure hypothesis vector；

Correlation logic between relay protection, breaker actuation, various warning messages is gathered, intelligent substation is mainly just gathered The breaker actuation alarm inside relay protection information and real time data Data2 inside internal stationary data Data1；

Step 2：Networking modeling analysis；

The information classification that step 1 is collected is 4 classes：Crucial warning information, relay protection information, reason event and outlet Breaker information；

Step 3：Sample matrix is set up to be estimated；

It is to screen and reject the weak information of correlation to set up the purpose that sample matrix is estimated, and improves the standard of collection information True property and importance；

Assuming that having n+m information in intelligent substation, n is relay protection information (B1~B6), and m is outlet breaker information (D1~D6), the business reasons hits within the period is k (C1~C5)；Set up sample matrix：

Wherein vector V_iRegard a sample group of SPACE V as, then sample group quantity is n in V, and dimension is k；

Assuming that sample group is divided into c class, i.e., a c group of planes.If clustering number of times is t (t >=1, t ∈ N), when initial T=1 is made, any selection c (1≤c≤n) individual sample groups are used as c class G from all sample groups_p(p=1,2 ..., c) initial Cluster centre

Then any one sample group V_iEuclidean distance to c cluster centre is

For any one sample group V_i, alwaysOne(p=1,2 ..., c) makeMinimum is reached, then By V_iIt is divided into class G_pIn；

All sample groups have been traveled through, 1 clustering is completed.Define the square error summation of cluster：

As the index for assessing similitude in class, by E^(t)＞ 1 data are all rejected.Following steps only calculate E^(t)≤1 Data；

Step 4：Calculate the corresponding target function value of failure hypothesis.

Initial signal for accident is acted with live tank circuit breakers separating brake；Whether powered analyzed by topological structure of breaker is carried out Once power supply interrupted district identification accident occurs, all possible reason event composition event of warning information searching after intelligently filters is utilized Hinder hypothesis collection, thus failure cause Solve problems are expressed as without constraint Zero-one integer programming problem；

IfTo actually receive the letter of the critical alert after intelligent alarm processing Breath collection, wherein a_iWithFor breaker actuation information and time of origin, b_mWithProtection act information and time of origin.To expect the critical alert information collection received.Failure hypothesis collection H is institute Possible reason event sets；

Object function Fobj (H_i) be used to weigh certain failure hypothesis H in failure hypothesis collection H_iReasonability, object function is：

Fobj(H_i)=min (w₁|▽M(H_i)|+w₂|T(H_i)|+w₃|H_i|)

Wherein, alarm coverage index ▽ M (H_i) it is K dimensional vectors (K is warning information sum)；|▽M(H_i) | reflect event Hinder hypothesis H_iCorrespondence expects that alarm can not cover the number of actual alarm information；I.e. failure assumes H_iThe live coverage with alarm, Its value is smaller, it is meant that a covering of the failure hypothesis closer to failure virtual condition；Failure hypothesis diversity factor index T (H_i) For K dimensional vectors, | T (H_i) | reflection actual alarm information and failure assume H_iCorresponding expectation alarm aggregation M^*(H) diversity factor, | T(H_i) | it is smaller, it is meant that faulty equipment, malfunction or the protection of tripping and fewer, the failure hypothesis H of breaker sum_iConfidence level It is higher；

Step 5：Judge whether to meet end condition；

Optimizing solution is carried out to the object function without constraint Zero-one integer programming problem of above-mentioned formation, finally given optimal Solve H^best, confidence level highest failure hypothesis, function Fobj (H can be obtained_i) there is optimal solution to meet end condition, meet condition Go to step 6；If function Fobj (H_i) 3 are then gone to step without optimal solution, proceed to set up fault sample matrix and be estimated, reject Gross error；

Step 6：Obtain correct failure hypothesis；

Correct failure hypothesis is now obtained, i.e., correct failure hypothesis correspondingly expects each alarm state received

Step 7：The alarm state received in practice；

The real protection actionable alarms provided inside intelligent substation inside real time data Data2 are provided.

Step 8：Comparison information；

The information of step 6 and step 7 is compared, diagnostic result is exported.

Step 9：The alarm failing to report or report by mistake, the breaker of malfunction tripping；

According to resulting diagnostic result, you can judge to fail to report and report by mistake, malfunction, tripping breaker；Its judgment principle is： Have in expecting and actually do not occur as failing to report or breaker tripping；And do not actually occurred in expecting as wrong report or breaker malfunction；

Step 10：Defect elimination is carried out to failure, recovers normal operation；

After accident diagnosis result comes out, according to the result of step 9, defect elimination is carried out to failure, intelligence rapidly carries out accident Processing, recovers normal operation.

The present invention compared with the existing technology, with advantages below and beneficial effect：The present invention proposes the event of intelligent substation Architecture feature of the barrier processing diagnostic method according to intelligent substation, two-dimentional classification is carried out emphatically to warning information, is set up Fault sample matrix is estimated, reject weak relevant information, intelligently filters alarm processing, the flow framework of fault diagnosis algorithm and Principle is described in detail, and demonstrates validity of the present invention and feasibility by embodiment.Present invention reduces power system Management cost, change fall behind, outmoded artificial accident analysis and tupe, greatly reduce intelligent substation operation Man power and material, the production cost reduced, improves operating efficiency, is national " 13 " intelligent substation construction and operation There is provided technical support.

Brief description of the drawings

Fig. 1 is a kind of step flow chart of intelligent substation method for diagnosing faults of the invention；

Fig. 2 is the topology diagram of step 3 networking modeling analysis of the present invention；

Fig. 3 is the electrical structure diagram of embodiment of the present invention.

Embodiment

Below in conjunction with accompanying drawing, technical scheme is clearly and completely described.

As shown in figure 1, a kind of intelligent substation method for diagnosing faults, the invention is characterised in that, enter by following 10 steps OK：

Step 1：Gather each variable of failure hypothesis vector；

Correlation logic between relay protection, breaker actuation, various warning messages is gathered, intelligent substation is mainly just gathered The breaker actuation alarm inside relay protection information and real time data Data2 inside internal stationary data Data1；

Step 2：Networking modeling analysis；

The information classification that step 1 is collected is 4 classes：Crucial warning information, relay protection information, reason event and outlet Breaker information；

Step 3：Sample matrix is set up to be estimated；

It is to screen and reject the weak information of correlation to set up the purpose that sample matrix is estimated, and improves the standard of collection information True property and importance；

Assuming that having n+m information in intelligent substation, n is relay protection information (B1~B6), and m is outlet breaker information (D1~D6), the business reasons hits within the period is k (C1~C5)；Set up sample matrix：

Wherein vector V_iRegard a sample group of SPACE V as, then sample group quantity is n in V, and dimension is k；

Assuming that sample group is divided into c class, i.e., a c group of planes.If clustering number of times is t (t >=1, t ∈ N), when initial T=1 is made, any selection c (1≤c≤n) individual sample groups are used as c class G from all sample groups_p(p=1,2 ..., c) initial Cluster centre

Then the Euclidean distance of any one sample group Vi to c cluster centres is

For any one sample group V_i, alwaysOne(p=1,2 ..., c) makeMinimum is reached, then By V_iIt is divided into class G_pIn；

All sample groups have been traveled through, 1 clustering is completed.Define the square error summation of cluster：

As the index for assessing similitude in class, by E^(t)＞ 1 data are all rejected.Following steps only calculate E^(t)≤1 Data；

Step 4：Calculate the corresponding target function value of failure hypothesis.

Initial signal for accident is acted with live tank circuit breakers separating brake；Whether powered analyzed by topological structure of breaker is carried out Once power supply interrupted district identification accident occurs, all possible reason event composition event of warning information searching after intelligently filters is utilized Hinder hypothesis collection, thus failure cause Solve problems are expressed as without constraint Zero-one integer programming problem；

IfTo actually receive the letter of the critical alert after intelligent alarm processing Breath collection, wherein a_iWithFor breaker actuation information and time of origin, b_mWithProtection act information and time of origin.To expect the critical alert information collection received；Failure hypothesis collection H is institute Possible reason event sets；

Object function Fobj (H_i) be used to weigh certain failure hypothesis H in failure hypothesis collection H_iReasonability, object function is：

Fobj(H_i)=min (w₁|▽M(H_i)|+w₂|T(H_i)|+w₃|H_i|)

Wherein, alarm coverage index ▽ M (H_i) it is K dimensional vectors (K is warning information sum)；|▽M(H_i) | reflect event Barrier hypothesis Hi correspondences expect that alarm can not cover the number of actual alarm information；I.e. failure assumes H_iThe live covering with alarm Degree, its value is smaller, it is meant that a covering of the failure hypothesis closer to failure virtual condition；Failure hypothesis diversity factor index T (H_i) it is K dimensional vectors, | T (H_i) | reflection actual alarm information and failure assume H_iCorresponding expectation alarm aggregation M^*(H) difference Degree, | T (H_i) | it is smaller, it is meant that faulty equipment, malfunction or the protection of tripping and fewer, the failure hypothesis H of breaker sum_iCan Reliability is higher；

Step 5：Judge whether to meet end condition；

Optimizing solution is carried out to the object function without constraint Zero-one integer programming problem of above-mentioned formation, finally given optimal Solve H^best, confidence level highest failure hypothesis, function Fobj (H can be obtained_i) there is optimal solution to meet end condition, meet condition Go to step 6；If function Fobj (H_i) 3 are then gone to step without optimal solution, proceed to set up fault sample matrix and be estimated, reject Gross error；

Step 6：Obtain correct failure hypothesis；

Correct failure hypothesis is now obtained, i.e., correct failure hypothesis correspondingly expects each alarm state received

Step 7：The alarm state received in practice；

The real protection actionable alarms provided inside intelligent substation inside real time data Data2 are provided.

Step 8：Comparison information；

The information of step 6 and step 7 is compared, diagnostic result is exported.

Step 9：The alarm failing to report or report by mistake, the breaker of malfunction tripping；

According to resulting diagnostic result, you can judge to fail to report and report by mistake, malfunction, tripping breaker；Its judgment principle is： Have in expecting and actually do not occur as failing to report or breaker tripping；And do not actually occurred in expecting as wrong report or breaker malfunction；

Step 10：Defect elimination is carried out to failure, recovers normal operation；

After accident diagnosis result comes out, according to the result of step 9, defect elimination is carried out to failure, intelligence rapidly carries out accident Processing, recovers normal operation.

For the clearly explanation above method, analyzed with the following Examples.

Embodiment：Certain 220kV transformer stations 1M buses physical fault is case, verifies fault diagnosis side proposed by the invention Method.The running status of power network is as shown in Figure 3 before failure.

Original message 110 was received within 60 seconds, using method proposed by the present invention to warning information compress filtering and After merging treatment, nearly ten are reduced to, obtained critical alarms information is as shown in table 1, the corresponding networking modeling analysis such as institute of table 2 Show, as shown in table 3, related breaker and numbering are as shown in table 4 for event code the reason for related, analysis institute's relative breaker and Message coding is as shown in table 5.

Table 1：The critical alarms information received

Table 2：Corresponding networking modeling analysis

Sequence number	Protect name	Reason event	Outlet breaker
				A1	B1	C11	D1~D5
A2	B1	C11	D1~D5
				A3	B2	C1~C14	D6
A3	B3	C1~C14	D6
				A4	B2	C1~C14	D5
A4	B3	C1~C14	D5
				A5	B2	C1~C6, C11~C14	D4
A5	B3	C1~C6, C11~C14	D4
				A1	B1	C12	D5~D7, D8~D10
A2	B2	C12	D5~D7, D8~D10

Wherein each symbol is so representative and its implication are as follows：

B1 zero-sequencedirectional protection B2 overcurrent protection B3 bus differential protections

Table 3：Related the reason for event code

Table 4：Related breaker and numbering

Numbering	Breaker name	Numbering	Breaker name
				D1	4736 breakers	D2	4738 breakers
D3	4740 breakers	D4	2012 breakers
				D5	2015 breakers	D6	2026 breakers

The related crucial alarm of analysis and its outlet breaker

Table 5：Analyze institute's relative breaker and message coding

Failure hypothesis vector：H=[c1, c2, c3, c4, c5, c6, c7, c8, c9, c10, c11, c12, c13, c14] is represented Whether each reason event really occurs, and there occurs then ci=1, does not occur to be then 0.The alarm aggregation actually received：Relay protection Action message A1~A5 and breaker D1~D6 action message：A6~A11.

Expect alarm aggregation：Corresponding to the different values of each failure hypothesis, different expectation alarms are had, are occurred with c12 Exemplified by, illustrate the calculating process of the target function value of failure hypothesis.Each coefficient, w in object function calculation formula₁, w₂, w₃Respectively For 10000,10,1.If c12=1, remaining ci=0 (i ≠ 12) is 0, calculates its corresponding expectation.C12 is in mixing rule net Obtain the protection correspondence of all crucial warning information, outlet breaker, its triggering of its corresponding " 220kV 2M bus-bar faults " Message, collectively constitute C11 it is corresponding expectation, including：A1, A2, A3, A4, A5, A9, A10, A11, A12, A13, A14, A15. Expectation state and virtual condition are substituted into formula and calculate ▽ M (H)=30000, T (H)=70, | H |=1, Fobj (H_i)= 30071。

The result of other failure hypothesis also can class this calculating, program analysis after, the result for obtaining optimization is：C11= 1, ci=0 (i ≠ 11), i.e. 220kV 1M bus-bar faults, during the object function corresponding to it is asked for, expect alarm and reality Alarm is completely superposed, and the number that the failure hypothesis vector reason event after coincidence occurs is 1, is optimal solution.

Analysis process in the case of information is imperfect：Assuming that having leaked certain two message in the alarm received：A2220kV1M- The BP-2B of 2M bus differential protections two；The breaker actuation of A6220kV Yunnan Shandong III lines 4736.Its analysis method is consistent with above procedure, by Table 2 understands that element contained by its failure hypothesis H is constant；Object function is calculated, in upper section case consistent with the method normally calculated Optimal solution c11=1, remaining ci=0 (i ≠ 11) is 0, due to not receiving A2, and A6, and the object function that correspondence is obtained can become Greatly, its desired value can be tried to achieve：Coverage is 0, and diversity factor is 2 × 10=20, obtains Fobj (H_i)=21, and pass through sequential operation knot Fruit understands that this is minimum value, and fault diagnosis result is still 220kV 1M bus-bar faults, meets the end condition of the step 5, turns Step 6, that is, correct failure hypothesis is obtained.

Draw after result above, according to analysis principle program may recognize that A2 is misses alarm, and A6 is to breaker just For the breaker of tripping.

According to fault diagnosis result：The eastern regions of the Yunnan Province intelligent station 220kV 1M bus-bar faults, search the eastern regions of the Yunnan Province intelligent station bus automatically The accident treatment prediction scheme of failure, technical staff starts out at once carry out troubleshooting.

Present invention employs advanced reliable, safe diagnostic method, be conducive to strengthen marketing, transformer station operation and Aggregate level is lifted, is that power supply enterprise and company explore and save a large amount of human and material resources, makes intelligent substation operating cost saving More than 60%.

Claims (1)

1. a kind of intelligent substation method for diagnosing faults, it is characterised in that carried out by following 10 steps：

Step 1：Gather each variable of failure hypothesis vector；

Correlation logic between relay protection, breaker actuation, various warning messages is gathered, intelligent substation internal stationary number is gathered According to the breaker actuation warning information inside the relay protection information inside Data1 and real time data Data2；

Step 2：Networking modeling analysis；

The information classification that step 1 is collected is 4 classes：Crucial warning information, relay protection classification information, reason event and outlet Breaker information；

Step 3：Sample matrix is set up to be estimated；

It is to screen and reject the weak information of correlation to set up the purpose that sample matrix is estimated, and improves the accuracy of collection information And importance；

Assuming that having jbn+jbm information in intelligent substation, jbn is relay protection classification information (B1~B6), and jbm is disconnected for outlet Road device information (D1~D6), in the information and real time data Data2 inside collection intelligent substation internal stationary data Data1 The reason for information time section in face, event hits was n × k (C1~C5)；Set up sample matrix：

Wherein vector V_iRegard a sample group of SPACE V as, then sample group quantity is n in V, and dimension is k；

Assuming that sample group is divided into c class, i.e., a c group of planes；If clustering number of times is t, t >=1, t ∈ N, initial season t= 1, arbitrarily c sample group of selection is used as c class G from all sample groups_p, in 1≤c≤n, p=1,2 ..., c initial clustering The heart

$m_p^{\left(1\right)}=(m_{p,1}^{\left(1\right)},m_{p,2}^{\left(1\right)},...,m_{p,k}^{\left(1\right)})=(V_{i,1},V_{i,2},...,V_{i,k}),(p=1,2,...,c)$

Then any one sample group V_iEuclidean distance to c cluster centre is

$d(V_i,m_p^{\left(1\right)})=\sqrt{\underset{j=1}{\overset{k}{\Σ}}{(V_{i,j}-m_{p,j}^{\left(1\right)})}^2},(p=1,2,...,c)$

For any one sample group V_i, alwaysP=1,2 ..., c, makesMinimum is reached, then by V_i It is divided into class G_pIn；

All sample groups have been traveled through, 1 clustering is completed；Define the square error summation of cluster：

$E^{\left(t\right)}=\underset{p=1}{\overset{c}{\Σ}}\underset{V_i\∈G_p}{\Σ}|V_i-m_p^{\left(t\right)}{|}^2$

As the index for assessing similitude in class, by E^(t)＞ 1 data are all rejected；Following steps only calculate E^(t)≤ 1 number According to；

Step 4：Calculate the corresponding target function value of failure hypothesis；

Initial signal for accident is acted with live tank circuit breakers separating brake；Once accident occurs, check whether breaker is powered, and by Analysis of Topological Structure recognizes power supply interrupted district；Utilize all possible reason event composition event of warning information searching after intelligently filters Hinder hypothesis collection, thus failure cause Solve problems are expressed as without constraint Zero-one integer programming problem；

IfTo actually receive the critical alert information collection after intelligent alarm processing, Wherein a_prWithFor breaker actuation information and time of origin, b_qWithProtection act information and time of origin；To expect the critical alert information collection received；Failure hypothesis collection H is institute Possible reason event sets；

Object function Fobj (H_if) be used to weigh certain failure hypothesis H in failure hypothesis collection H_ifReasonability, object function is：

$Fobj\left(H_{if}\right)=min\left(w_1\right|\▿M\left(H_{if}\right)|+w_2|T\left(H_{if}\right)|+w_3|H_{if}\left|\right)$

Wherein, alarm coverage indexFor K dimensional vectors, K is warning information sum；Reflect failure hypothesis H_ifIt is right It should it is expected that alarm can not cover the number of actual alarm information, i.e. failure hypothesis H_ifThe live coverage with alarm, its value is smaller, Mean a covering of the failure hypothesis closer to failure virtual condition；Failure hypothesis diversity factor index T (H_if) for K tie up to Amount, | T (H_if) | reflection actual alarm information and failure hypothesis H_ifCorresponding expectation alarm aggregation M^*(H) diversity factor, | T (H_if) | it is smaller, it is meant that faulty equipment, malfunction or the protection of tripping and fewer, the failure hypothesis H of breaker sum_ifConfidence level it is higher；

Step 5：Judge whether to meet end condition；

Optimizing solution is carried out to the object function without constraint Zero-one integer programming problem of above-mentioned formation, optimal solution is finally given H^best, confidence level highest failure hypothesis, function Fobj (H can be obtained_if) there is optimal solution to meet end condition, meet condition and turn Step 6；If function Fobj (H_if) 3 are then gone to step without optimal solution, continue to set up fault sample matrix being estimated, reject thick mistake Difference；

Step 6：Obtain correct failure hypothesis；

Correct failure hypothesis is now obtained, i.e., correct failure hypothesis correspondingly expects each alarm state received；

Step 7：The alarm state received in practice；

The real protection breaker actuation warning information provided inside intelligent substation inside real time data Data2 is provided；

Step 8：Comparison information；

The information of step 6 and step 7 is compared, diagnostic result is exported；

Step 9：The alarm failing to report or report by mistake, the breaker of malfunction tripping；

According to resulting diagnostic result, you can judge to fail to report and report by mistake, malfunction, tripping breaker；Its judgment principle is：Expect In have, it is and actual when not occurring, be judged to fail to report or breaker tripping；Do not have in expecting, and actually occur, be judged to report by mistake or break Road device malfunction；

Step 10：Defect elimination is carried out to failure, recovers normal operation；

After accident diagnosis result comes out, according to the result of step 9, defect elimination is carried out to failure, intelligence rapidly carries out accident treatment, Recover normal operation.