CN104764979A - Virtual information fusion power grid alarming method based on probabilistic reasoning - Google Patents

Abstract

The invention discloses a virtual information fusion power grid alarming method based on probabilistic reasoning. Firstly, received alarms are classified and preprocessed, the voltage out-of-limit alarm, the protection action alarm and the breaker tripping alarm are extracted, and redundant information in the alarms is effectively screened out; secondly, the alarms are judged through probabilistic reasoning, possible causative events and event probability are obtained, and multi-angle and multi-level deep excavation and analysis on the alarm information is achieved through classification; finally, virtual information fusion is conducted on three judgment results based on the improved D-S evidence theory, and event reasoning results including an alarm source event, false alarms and missing alarms are obtained. By means of emulation proof, the method can effectively improve the timeliness and accurate rate of the alarms, the requirement for power grid real-time alarming can be met on the aspects of the processing speed and the accurate rate, feasibility is large, the method has good application prospects, and a new reference concept is provided for later alarm processing and studying.

Description

A kind of virtual information based on probability inference merges grid alarm method

Technical field

The invention belongs to the Fault Identification technical field of Operation of Electric Systems monitoring, be specifically related to a kind of virtual information based on probability inference and merge grid alarm method.

Background technology

When grid collapses or abnormal running; dispatching center's energy management system (energymanagement system; EMS) can a large amount of warning message be received continuously in a short period of time, comprise the various alarms such as voltage out-of-limit, trend are out-of-limit, protective relaying device action, switch trip, reclosing, garble.Along with expansion and the application of various intelligent electric equipment in electric system of electrical network scale, electric network composition is more complicated, and the warning message being submitted to dispatching center presents explosive growth, wherein also comprises various false alarm.How at short notice fast processing magnanimity warning message, for management and running personnel provide effective aid decision making, is one of Important Problems of electric system research.

Up to the present, proposed the multiple alarm processing method based on artificial intelligence technology both at home and abroad, be mainly divided into two classes: based on model class and rule-based class.Wherein, the technique study based on model is more late, mainly refers to the stratified flow model grid alarm that doctor Larsson in 2008 proposes with the cooperation of Utilities Electric Co. of Switzerland.In addition, " the electric network fault technical research based on Multi-information acquisition " (civilian Qingfeng County, master thesis) by studying the feature of Fault Recorder Information, propose the electric parameters fault model based on Hilbert-Huang transform, utilize quick intrinsic model analysis and Hilbert transform, change electric parameters failure message into quantitative fault measurement and carry out electric network failure diagnosis.For the different faults feature separately of different elements after fault; set up circuit and bus criterion model respectively; study the fault diagnosis algorithm measuring metrical information based on WAMS; make full use of WAMS information, make up conventional fault diagnosis method and diagnose inaccurate shortcoming in the situations such as protection information disappearance, exception.Consider the reliability of protection, breaker actuation simultaneously; in analytic model, introduce blur level, ask for resolve fault degree based on switching value; and characterize in conjunction with electric parameters probability of malfunction, carry out information fusion by D-S evidence theory, realize the fault diagnosis of Multi-source Information Fusion.

The technique study of rule-based class is comparatively ripe, comprise expert system, artificial neural network (artificialneural network, ANN), based on Fuzzy probabilistic inference, based on Petri network etc.The alarm processing method of rule-based class utilizes intelligent algorithm to set rule more, by the corresponding relation of alarm and event stored in rule base, infers electric network fault by finding legal event.These methods are all the unified process all alarms not being added differentiation when the problem of process, in fact, when electrical network generation disturbance or fault, produce 80% of warning message and caused by same event, that is, wherein there is the duplicate message of bulk redundancy, such as during certain line failure, the voltage out-of-limit alarm A of circuit can be produced, the protective relaying device meeting action generation alarm B that distance fault point is nearest, then corresponding isolating switch can trip and produce alarm C, A, B, inner link is had between C, the information redundancy problem caused like this can the serious speed restricting alert process.If but only utilized a kind of alarm wherein, as protection act alarm, the accuracy of judgement could be affected again because the problem such as false protection or tripping, communication disruption, garble may be there is in Operation of Electric Systems.

Therefore, how can utilize various warning message all sidedly, can carry out sieve again subtract the warning comprising same event information, infer the source event causing and report to the police more quickly and accurately, be an alert process problem urgently to be resolved hurrily.

Summary of the invention

The object of this invention is to provide a kind of virtual information based on probability inference and merge grid alarm method, to solve the Fault Identification problem under electric system alarm condition.

In order to realize above object, the technical solution adopted in the present invention is: a kind of virtual information based on probability inference merges grid alarm method, comprises the steps:

(1) all warning informations received when occurring according to grid disturbance or fault, determine all reason event sets Events, Events is element e _jset, then received all warning informations to be classified, set up three alert sequence set: voltage out-of-limit alarm aggregation VL{a _i, protection act alarm aggregation PR{a _i, breaker actuation alarm aggregation CB{a _i, wherein, e _jbe a grid event, a _iit is a warning information;

(2) theoretical according to probability inference, to element e _jcorresponding VL{a _i, PR{a _i, CB{a _ithree alert sequence set synchronously carry out event occurrence rate and calculate, and obtain the probability of happening P of corresponding reason event respectively _vL(e _j), P _pR(e _j), P _cB(e _j);

(3) corresponding to all elements in Events P _vL(e _j), P _pR(e _j), P _cB(e _j) be normalized and obtain respectively

(4) utilize improve D-S evidence theory, will with element e _jcorresponding probable value carry out virtual information fusion, draw fault distinguishing result.

In described step (1); actual conditions when occurring according to grid disturbance or fault; synchronously there is the trigger condition as alarm signal processing routine using protection act and corresponding circuit breaker trip, be triggered the moment as a time window TW (t using alarm signal processing routine _s, t _e) starting point t _s, the alarm of the last item circuit breaker trip is as TW (t _s, t _e) terminal t _e, the sequencing temporally gone up after alert category is incorporated to corresponding set respectively.

In described step (1), determine that the process of all reason event sets Events is as follows:

To warning information a _i, all reason event sets corresponding with it are:

Event{a _i}＝{e ₁,e ₂,e ₃,...}，i＝1,2,3...,m

Then $\mathrm{Events}=\underset{i=1}{\overset{m}{\∪}}\mathrm{Event}\left\{a_i\right\}=\{e_1,e_2,e_j,...e_n\}$

Wherein, e _jselection principle be: a _ie _jfirst element of caused alarm sequence set.

In described step (2), be directed to element e _jany one alert sequence, the computation process of its corresponding reason event occurrence rate is as follows: for element e _j, the alarm aggregation Alarm{e of its correspondence _j}={ a _i| i=1,2,3..., q}, so define a q dimensional vector X=[x ₁, x ₂, x ₃..., x _q], if reason event e _jalarm a in corresponding alarm aggregation _iat time window TW (t _s, t _e) in there occurs, so get x _i=1, otherwise get x _i=0, to VL{a _i, PR{a _i, CB{a _iany one in three alert sequence, e _jprobability of happening be P (e _j)=| X|/q.

To VL{a _i, PR{a _i, CB{a _iany one in three alert sequence, the P (e of its correspondence _j) the computing formula of normalized be:

In described step (4), will carry out information fusion as three separate evidences, form new evidence, specific formula for calculation is as follows:

In formula, m is the basic reliability distribution on identification framework B, λ _ifor the weight factor of each evidence.

Weight factor λ _icomputing formula as follows:

${\mathrm{\λ}}_i=1-\frac{d\left(B_i\right)}{\sqrt{\underset{\∩B_i=B}{\mathrm{\Σ}}{|m\left(B_i\right)-m_{\mathrm{center}}\left(B_i\right)|}^2}}$

$m_{\mathrm{center}}\left(B_i\right)=\frac{\underset{\∩B_i=B}{\mathrm{\Σ}}\underset{1\≤j\≤n}{\mathrm{\Σ}}{m}_j\left(B_i\right)}{n}$

d(B _i)＝|m(B _i)-m _center(B _i)|。

The virtual information that the present invention is based on probability inference merges grid alarm method, first the alarm received is classified and pre-service, extract voltage out-of-limit, protection act, the alarm of circuit breaker trip three class, and form respective alert sequence set, as three " virtual information sources ", effectively sieve the redundant information subtracted in alarm; Secondly all kinds of alarm is judged respectively by probability inference, draw possible reason event and probability of occurrence, achieve warning information multi-angle, multifaceted degree of depth mining analysis by classification; Finally with the D-S evidence theory information fusion method improved, virtual information fusion is carried out to three kinds of judged results, draw reasoning result, comprise warning source event, false alarm and alarming loss.The method is for IEEE14 node system, alarm processing result is verified, simulation results shows, the method effectively improves the ageing of warning and accuracy rate, processing speed and accuracy rate can meet the needs of electrical network Realtime Alerts, feasibility is comparatively large, has a good application prospect, and is that later alarm processing research provides new reference thinking.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the virtual information fusion grid alarm method based on probability inference;

Fig. 2 is for the emulation test system of IEEE-14 node system.

Embodiment

Below in conjunction with accompanying drawing and specific embodiment, the present invention is described further.

As shown in Figure 1, the invention provides a kind of virtual information based on probability inference and merge grid alarm method, comprise the steps:

(1) all warning informations received when occurring according to grid disturbance or fault, determine all reason event sets Events, Events is element e _jset, then received all warning informations to be classified, set up three alert sequence set: voltage out-of-limit alarm aggregation VL{a _i, protection act alarm aggregation PR{a _i, breaker actuation alarm aggregation CB{a _i, wherein, e _jbe a grid event, a _iit is a warning information.

In this step, synchronously there is the trigger condition as alarm signal processing routine using protection act and corresponding circuit breaker trip, be triggered the moment as a time window TW (t using alarm signal processing routine in actual conditions when occurring according to grid disturbance or fault _s, t _e) starting point t _s, set up three alert sequence set, voltage out-of-limit alarm aggregation VL{a _i, protection act alarm aggregation PR{a _j, breaker actuation alarm aggregation CB{a _k, the sequencing temporally gone up after alert category is incorporated to corresponding set respectively.In order to ensure as far as possible comprehensively effectively to cover warning information in an alarm processing, using the alarm of the last item circuit breaker trip as TW (t _s, t _e) terminal t _e.

For e _jand a _ibetween relation, first clearly relevant concept: e _ibe a grid event (equipment failure, system disturbance etc.), a _ibe a warning information (voltage out-of-limit, protective device action or circuit breaker trip etc.), different events can produce different alarms, and both relations can be described as:

e _i→A _i＝{a ₁,a ₂,a ₃,...}

In formula: A _ifor event e _iduring generation, the alarm aggregation that energy management system EMS receives.Meanwhile, same alarm may be caused by different events, and both relations can be described as:

a _i→E _i＝{e ₁,e ₂,e ₃,...}

In formula: E _ifor alarm a may be caused _ithe set of all Possible events.

At a time window TW (t _s, t _e) in, three class alarms are handled as follows respectively:

To each warning a _i, determine all possible reason event sets

Event{a _i}＝{e ₁,e ₂,e ₃,...e _j}，i＝1,2,3...,m

E _jselection principle be: a _ie _jfirst element of caused alarm sequence set, selecting like this is to reduce Event{a _iin the quantity of element, and corresponding first alarm of alarm aggregation of event is maximum with the relevance of event comparatively speaking, like this selection have logic according to.

By the reason event sets Event{a obtained _i(i=1,2,3..., m) get union, to obtain final product

$\mathrm{Events}=\underset{i=1}{\overset{m}{\∪}}\mathrm{Event}\left\{a_i\right\}=\{e_1,e_2,e_j,...e_n\}.$

(2) theoretical according to probability inference, to element e _jcorresponding VL{a _i, PR{a _i, CB{a _ithree alert sequence set synchronously carry out event occurrence rate and calculate, and obtain the probability of happening P of corresponding reason event respectively _vL(e _j), P _pR(e _j), P _cB(e _j).

For any one alert sequence, be directed to element e _j, the computation process of its corresponding reason event occurrence rate is as follows: reason event e _jcorresponding alarm aggregation is Alarm{e _j}={ a _i| i=1,2,3..., q}, so define a q dimensional vector X=[x ₁, x ₂, x ₃..., x _q], if e _jalarm a in corresponding alarm aggregation _iat time window TW (t _s, t _e) in there occurs, so get x _i=1, otherwise get x _i=0, to VL{a _i, PR{a _i, CB{a _iany one in three alert sequence, then e _jprobability of happening be P (e _j)=| X|/q.

(3) corresponding to all elements in Events P _vL(e _j), P _pR(e _j), P _cB(e _j) be normalized and obtain respectively

In order to obtain unified evident information, to VL{a _i, PR{a _i, CB{a _iany one in three alert sequence, the P (e of its correspondence _j) the computing formula of normalized be:

(4) for element e _j, utilize the D-S evidence theory improved, by the probable value corresponding with it carry out virtual information fusion, draw fault distinguishing result.

Will carry out information fusion as three separate evidences, the compositional rule according to the classical evidence theory improved merges it, and form new evidence, specific formula for calculation is as follows:

In formula, m is the basic reliability distribution on identification framework B, λ _ifor the weight factor of each evidence.

Subjective given λ _ivalue often lack rationale, therefore, the present invention, on the basis of previous work, in conjunction with the concept of evidence center and distance, proposes a kind of λ _iobtaining value method.

Evidence center is such fresh evidence, and similarity degree summation is maximum on evidence for it and institute, asks for formula to be:

$m_{\mathrm{center}}\left(B_i\right)=\frac{\underset{\∩B_i=B}{\mathrm{\Σ}}\underset{1\≤j\≤n}{\mathrm{\Σ}}{m}_j\left(B_i\right)}{n};$

The distance at each evidence and evidence center:

d(B _i)＝|m(B _i)-m _center(B _i)|；

Because distance is nearer, then the support of evidence to result is larger, and weight factor therefore can be considered to be taken as:

${\mathrm{\λ}}_i=1-\frac{d\left(B_i\right)}{\sqrt{\underset{\∩B_i=B}{\mathrm{\Σ}}{|m\left(B_i\right)-m_{\mathrm{center}}\left(B_i\right)|}^2}}.$

After using the Weighted Fusion formula improved, if certain evidence and other evidences conflict are comparatively serious, namely far away with evidence centre distance, the weight of this evidence when merging can be less, less on the impact of result reasoning, meanwhile, also misdata can be searched intuitively.

Get IEEE-14 node system as emulation test system, as shown in Figure 2.

Suppose circuit L ₆break down, B ₃side protection PR ₉action, CB ₉tripping operation; B ₄side protection PR ₁₀tripping, causes circuit L ₉, L ₁₁the low-pressure side protection act of upper transformer, CB ₁₄, CB ₂₀tripping operation, meanwhile, as main protection PR ₁₀back-up protection, B ₂, B ₅side protection PR ₅, PR ₁₂action, CB ₅, CB ₁₂tripping operation.Alarm aggregation after fault occurs is as shown in table 1.

Table 1: alarm aggregation and classification

Alarm	Type	Numbering
			Bus B 3Voltage out-of-limit	VL	a 3
Bus B 4Voltage out-of-limit	VL	a 4
			Protection PR 9Action	PR	b 9
Isolating switch CB 9Tripping operation	CB	c 9
			Protection PR 14Action	PR	b 14
Isolating switch CB 14Tripping operation	CB	c 14

Bus B 7Voltage out-of-limit	VL	a 7
			Bus B 8Voltage out-of-limit	VL	a 8
Protection PR 20Action	PR	b 20
			Isolating switch CB 20Tripping operation	CB	c 20
Bus B 9Voltage out-of-limit	VL	a 9
			Bus B 10Voltage out-of-limit	VL	a 10
Bus B 14Voltage out-of-limit	VL	a 14
			Protection PR 5Action	PR	b 5
Isolating switch CB 5Tripping operation	CB	c 5
			Bus B 2Voltage out-of-limit	VL	a 2
Bus B 1Voltage out-of-limit	VL	a 1
			Protection PR 12Action	PR	b 12
Isolating switch CB 12Tripping operation	CB	c 12
			Bus B 5Voltage out-of-limit	VL	a 5
Bus B 6Voltage out-of-limit	VL	a 6

In order to simplify problem, when initialization rule base, the most complicated event only considers the situation of the tripping of this route protection and adjacent lines relay fail when certain line fault, as circuit L ₁when breaking down, corresponding protection act alarm aggregation only includes PR at most ₁, PR ₈, PR ₄, PR ₆, PR ₂, PR ₂₂the situation of tripping.

Utilize the sorted alarm of table 1 to carry out probability inference, and information fusion is carried out to three kinds of probability, can judged result be obtained, as shown in table 2, in order to contrast, also list alarm in table 2 and not classifying, directly to the unified judged result of carrying out probability inference of all alarms.

Table 2: merge and non-fused Comparative result

e j	P PR	P CB	P VL	P f	P nf
						L 3	0.1072	0.1072	0.1600	0.1572	0.1834
L 6	0.5341	0.5341	0.5600	0.7348	0.6725
						L 7	0.1283	0.1283	0.1200	0.1735	0.2011
L 9	0.1230	0.1230	0.0800	0.1603	0.1194
						L 11	0.1074	0.1074	0.0800	0.1413	0.1236

Wherein, e _jfor possible failure cause event, as L ₃represent circuit L ₃break down, P _pR, P _cBand P _vLbe respectively the value after the probability of happening normalization utilizing the alarm of protection act class, the alarm of circuit breaker trip class and voltage out-of-limit class alarm reasoning gained event, P _ffor the result after fusion, P _nffor carrying out the result of probability inference to all alarm unifications, ask for P _fprogram runtime be 0.087s, ask for P _nfprogram runtime be 0.531s.

Contrast can be found out:

1) result after fusion for classification is higher, more reliable than the result accuracy rate of directly carrying out Fuzzy probabilistic inference.

2) easily judge according to fusion results: alarm is by circuit L ₆break down caused, and can go out lack protection PR according to rule-based reasoning ₁₀actionable alarms, can PR be judged ₁₀there occurs tripping situation.

3) method that the present invention proposes has greater advantages in alarm processing speed.

Above embodiment only understands core concept of the present invention for helping; the present invention can not be limited with this; for those skilled in the art; every according to thought of the present invention; the present invention is modified or equivalent replacement; any change done in specific embodiments and applications, all should be included within protection scope of the present invention.

Claims (7)

1. the virtual information based on probability inference merges a grid alarm method, it is characterized in that, comprises the steps:

(1) all warning informations received when occurring according to grid disturbance or fault, determine all reason event sets Events, Events is element e _jset, then received all warning informations to be classified, set up three alert sequence set: voltage out-of-limit alarm aggregation VL{a _i, protection act alarm aggregation PR{a _i, breaker actuation alarm aggregation CB{a _i, wherein, e _jbe a grid event, a _iit is a warning information;

(2) theoretical according to probability inference, to element e _jcorresponding VL{a _i, PR{a _i, CB{a _ithree alert sequence set synchronously carry out event occurrence rate and calculate, and obtain the probability of happening P of corresponding reason event respectively _vL(e _j), P _pR(e _j), P _cB(e _j);

(3) corresponding to all elements in Events P _vL(e _j), P _pR(e _j), P _cB(e _j) be normalized and obtain respectively

(4) utilize improve D-S evidence theory, will with element e _jcorresponding probable value carry out virtual information fusion, draw fault distinguishing result.

2. the virtual information based on probability inference according to claim 1 merges grid alarm method; it is characterized in that: in described step (1); actual conditions when occurring according to grid disturbance or fault; synchronously there is the trigger condition as alarm signal processing routine using protection act and corresponding circuit breaker trip, be triggered the moment as a time window TW (t using alarm signal processing routine _s, t _e) starting point t _s, the alarm of the last item circuit breaker trip is as TW (t _s, t _e) terminal t _e, the sequencing temporally gone up after alert category is incorporated to corresponding set respectively.

3. the virtual information based on probability inference according to claim 1 merges grid alarm method, it is characterized in that: in described step (1), determines that the process of all reason event sets Events is as follows:

To warning information a _i, all reason event sets corresponding with it are:

Event{a _i}＝{e ₁,e ₂,e ₃,...}，i＝1,2,3...,m

Then $\mathrm{Events}=\underset{i=1}{\overset{m}{\∪}}\mathrm{Event}\left\{a_i\right\}=\{e_1,e_2,e_j,...e_n\}$

Wherein, e _jselection principle be: a _ie _jfirst element of caused alarm sequence set.

4. the virtual information based on probability inference according to claim 1 merges grid alarm method, it is characterized in that: in described step (2), be directed to element e _jany one alert sequence, the computation process of its corresponding reason event occurrence rate is as follows: for element e _j, the alarm aggregation Alarm{e of its correspondence _j}={ a _i| i=1,2,3..., q}, so define a q dimensional vector X=[x ₁, x ₂, x ₃..., x _q], if reason event e ^jalarm a in corresponding alarm aggregation _iat time window TW (t _s, t _e) in there occurs, so get x _i=1, otherwise get x _i=0, to VL{a _i, PR{a _i, CB{a _iany one in three alert sequence, e _jprobability of happening be P (e _j)=| X|/q.

5. the virtual information based on probability inference according to claim 4 merges grid alarm method, it is characterized in that: to VL{a _i, PR{a _i, CB{a _iany one in three alert sequence, the P (e of its correspondence _j) the computing formula of normalized be:

6. the virtual information based on probability inference according to claim 1 merges grid alarm method, it is characterized in that: in described step (4), will carry out information fusion as three separate evidences, form new evidence, specific formula for calculation is as follows:

In formula, m is the basic reliability distribution on identification framework B, λ _ifor the weight factor of each evidence.

7. the virtual information based on probability inference according to claim 6 merges grid alarm method, it is characterized in that, weight factor λ _icomputing formula as follows:

${\mathrm{\λ}}_i=1-\frac{d\left(B_i\right)}{\sqrt{\underset{\∩B_i=B}{\mathrm{\Σ}}{|m\left(B_i\right)-m_{\mathrm{center}}\left(B_i\right)|}^2}}$

$m_{\mathrm{center}}\left(B_i\right)=\frac{\underset{\∩B_i=B}{\mathrm{\Σ}}\underset{1\≤j\≤n}{\mathrm{\Σ}}{m}_j\left(B_i\right)}{n}$

d(B _i)＝|m(B _i)-m _center(B _i)|。