CN105610132A

CN105610132A - Line and bus protection method based on initial fault angle, transition resistance and machine learning

Info

Publication number: CN105610132A
Application number: CN201510768445.9A
Authority: CN
Inventors: 郭振威; 姚建刚
Original assignee: Individual
Current assignee: Individual
Priority date: 2014-11-13
Filing date: 2015-11-11
Publication date: 2016-05-25

Abstract

The invention relates to a line and bus protection method based on initial fault angle, transition resistance and machine learning. A lot of high-frequency transient signals are generated by line faults, most of the high-frequency transient signals are passed by ground distribution equivalent capacitance of a bus when passing the bus, and the higher the frequency is, the more the signals are passed by; high-frequency components generated under different initial fault angles and different transition resistances differ a lot; the fault feature, initial fault angle, transition resistance and fault area label of a typical fault sample are used to form an input vector sample set of machine learning to train machine learning; and machine learning after training is used for fault determination. A corresponding protection motion value is set for differential fault conditions, adverse influence of the initial fault angle and the transition resistance is eliminated, the protection reliability is greatly improved, and error caused by setting the protection motion value manually is prevented. Only the typical fault sample rather than fault samples under all fault conditions is needed, the training load of machine learning is reduced, and the method is feasible practically.

Description

Circuit bus bar protecting method based on fault initial angle transition resistance and machine learning

Technical field

The present invention relates to the resist technology of grid power transmission circuit and bus, refer more particularly to and utilize fault transient information, baseIn the transmission line of electricity of transition resistance, fault initial angle and machine learning and the intelligent protection method of bus, the method proposing is suitableWith the protection of circuit and the bus of high pressure, super-pressure and extra-high voltage grid.

Background technology

At present, the protection extensively adopting in power transmission network is mainly the protection taking power frequency amount variable quantity as fault signature, protectsProtecting action generally needs 20～40ms, and the protection simultaneously changing based on power frequency quality is easily subject to transmission line of electricity distribution capacity electricityThe impact of the factor such as saturated, the power system oscillation of impact, current transformer and transition resistance of stream. Along with constantly sending out of electrical networkExhibition, requires the time of relay protection excision fault shorter, simultaneously also higher to the requirement of protection reliability. Carry the eighties in 20th centuryThe traveling-wave protection of the fault high fdrequency component producing based on fault haveing has ultrahigh speed acting characteristic, and is not subject to current transformerThe impacts such as saturated, power system oscillation. Therefore,, since based on fault transient, protection is suggested, transient protection is subject to electrical networkRelay protection worker and scholar's extensive concern. Transient protection has tunneling traffic to protect without tunneling traffic from general principle pointProtection. Although transient protection is considered to the very promising protection of one, performance is not in actual applications for transient protectionStable, reliability is not high, and the fault transient characteristic quantity utilizing in the transient protection of tracing it to its cause is easily subject to fault initial angle and mistakeCross the impact of resistance, but traditional transient protection is not considered the impact of the two. Chinese invention patent prospectus CN2013100656942 (based on transition resistance and the single-ended transient protection of fault angle reduction transmission line of electricity self adaptation) and document " Anewmethodfornon-unitprotectionofpowertransmissionlinesbasedonfaultresistanceandfaultanglereduction”(InternationalJournalofElectricalPower&EnergySystems, Vol.55,2014) disclose a kind of based on by temporary after fault initial angle and transition resistance reductionThe single-ended transient protection method of state amount. With respect to traditional Transient method, this guard method has overcome fault initial angle and transition electricityThe negative effect of resistance to protection, is greatly improved reliability. This guard method is very novel, but still exist defect andDeficiency, such as it is when the failure judgement, what still adopt is the method that traditional fault actions value is adjusted, for different faultsFault under condition (different fault initial angle, transition resistances) all adopts identical setting valve, and fault signature is in different eventsValue under barrier initial angle and transition resistance condition after reduction differs larger, is difficult to the effect that reaches desirable, and its reduction precision needsFurther improve, to eliminate better fault initial angle and the impact of transition resistance on protection, protect to reach further raisingProtect reliability and the object of protecting sensitivity.

Summary of the invention

The object of the invention is to overcome prior art deficiency, provide a kind of reliability higher for the protection of power transmission lineThe circuit bus bar protecting method based on fault initial angle transition resistance and machine learning of road and bus, general principle is: (1) is defeatedWhen electrical network is short-circuited fault, produce a large amount of high frequency transient electric current, voltage signal, high frequency transient signal is to each propagation pathPropagate, when by two ends bus, owing to there being very large distribution equivalent capacity over the ground on bus, transient high frequency is divided greatlyAmount is fallen by the distribution equivalent capacity over the ground of bus by-pass shunt, and the high frequency transient signal of same frequency range, by after bus, becomesSmall and weak a lot, and higher be bypassed more of frequency; (2) under the fault condition of different faults initial angle and transition resistance,The high fdrequency component difference that fault produces is very large; (3) vector that comprises fault signature, fault initial angle, transition resistance composition is doneFor the input vector of machine learning; (4) by protection zone with protection zone outside (typical fault position can in typical fault positionComprise bus, near the position of circuit two ends outlet, circuit centre position) and typical fault condition (typical fault condition, Ke YiweiTransition resistance is 1 Europe, 50 Europe, 100 Europe, 150 Europe, 300 Europe, with fault initial angle can be 0 °, 5 °, 45 °, 85 °, 90 °, bothBetween combination) under fault sample (information of each fault sample comprises fault signature, fault initial angle, transition resistanceThe input vector of machine learning of composition) composition typical fault sample set, the input vector of machine learning comprise fault signature,Fault initial angle, transition resistance, and apply these typical fault sample sets machine learning is trained; (5) application training is goodMachine learning carry out fault judgement, being equivalent to the corresponding protection action of having adjusted for different fault conditions and adjustingValue, can eliminate fault initial angle and transition resistance to protection adverse effect, significantly improves protection reliability. Corresponding therewith, thisThe bright problem that another will solve be to provide one can obtain a kind of reliability high, overcome transition resistance and fault initial angle shadowThe computational methods of the fault initial angle of the circuit bus bar protecting method ringing. For succinct and convenient for what narrate, definition the present inventionHigh fdrequency component treating capacity described in method, refers to high fdrequency component energy or the high fdrequency component wink of the high fdrequency component of fault-signalTime amplitude integration or high fdrequency component entropy or high fdrequency component complexity or high fdrequency component Singularity Degree or high fdrequency component modulus maximum or itDifference or their ratio or other high fdrequency component treating capacity in one. Engineering described in definition the inventive methodPractise, refer to SVMs or neutral net or genetic algorithm, K arest neighbors (k-NearestNeighbor, KNN) classificationAlgorithm or K-Means algorithm or C4.5 algorithm, Apriori algorithm, greatest hope (EM, Expectation –Maximization) algorithm, PageRank algorithm, Adaboost algorithm, NaiveBayes algorithm, classification and regression treeOne in the machine learning such as (CART, ClassificationandRegressionTrees) algorithm.

With regard to circuit bus bar protecting method of the present invention, implementation step comprises: (1) extracts (or the event of fault current signalBarrier voltage signal) high fdrequency component; (2) build the fault signature that comprises fault high-frequency information, fault high-frequency information is fault letterThe high fdrequency component treating capacity of number (fault-signal is fault current signal or failure voltage signal) or the ratio of high fdrequency component treating capacityValue or the difference of high fdrequency component treating capacity or be the capable ripple direction of fault-signal or fault-signal direction or fault-signal high frequencyComponent; High fdrequency component treating capacity is energy or high fdrequency component instantaneous amplitude integration or the high fdrequency component instantaneous amplitude of high fdrequency componentWith or high fdrequency component instantaneous amplitude, fault-signal entropy or fault-signal complexity, high fdrequency component entropy or high fdrequency component complexityDegree or high fdrequency component Singularity Degree or high fdrequency component modulus maximum; (3) discriminating fault types and Fault Phase Selection; (4) calculate transitionResistance and fault initial angle; (5) structure comprises the input vector of the machine learning of fault signature, fault initial angle, transition resistance;(6) training of machine learning, application typical fault sample set is trained machine learning; After machine learning trains, enteringWhen row fault judges, this step can be omitted, skip; (7) physical fault judgement, event is carried out in the machine learning that application training has been got wellBarrier judgement.

The circuit bus bar protecting method based on fault initial angle transition resistance and machine learning that the present invention proposes, has as followsBeneficial effect:

(1) without manually specifically protecting adjusting of working value, complete intelligently fault judgement, avoid manually adjustingThe issuable mistake of protection working value

(2) in the input vector due to machine learning, comprised fault initial angle and transition resistance, by machine learningMethod is carried out fault judgement intelligently, and it is equivalent to for the different fault conditions suitable working value of adjusting respectively; ByThe guard method of this visible this invention by the mode of machine learning eliminate intelligently fault initial angle with transition resistance to protectingAdverse effect, significantly improved protection reliability;

(3) machine learning of the inventive method is in training process, without the fault sample data under all fault conditions,And only need typical fault sample data, and not only alleviate the training burden of machine learning, the more important thing is and make side of the present inventionMethod has practical feasibility;

(4) the inventive method belongs to transient protection category, has advantages of traditional transient protection, and quick action, is not subject toCurrent transformer is saturated, system is shaken or the impact of system short circuit capacity;

(5) the inventive method has outside the advantage of traditional transient protection, also really accomplishes energy protection circuit total length;

(6) the inventive method to bus over the ground distribution capacity require little, can be suitable for well common high pressure, super-pressure,The protection of UHV transmission line and bus.

As the technical scheme 1 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) multiple high fdrequency component treating capacities, SVMs, fault initial angle and the method for a circuit of transition resistance protection, protectProtecting object is the circuit CB in Fig. 1, can comprise following step:

Step 1: extract above multiple high of two or three or three of fault current signal (or failure voltage signal)Frequency component:, at the C of protected circuit CB end, constantly gather fault current i₁(or voltage u₁), through analog-to-digital conversion by analog signalBe converted to data signal, the current signal transferring to after data signal is still used i₁Represent that (or voltage signal is still used u₁Represent), carryGet fault current i₁The high fdrequency component i of 3 frequency ranges_h1、i_h2、i_h3(or voltage u₁The high fdrequency component u of 3 frequency ranges_h1、u_h2、u_h3)；

Step 2: calculate the high fdrequency component treating capacity of the above multiple high fdrequency components of two or three or three, and with this twoIndividual or more than three or three multiple high fdrequency component treating capacities build fault signatures: by high fdrequency component i_h1、i_h2、i_h3(or u_h1、u_h2、u_h3) calculate their high fdrequency component treating capacity, be expressed as P₁、P₂、P₃; With the one of high component treating capacity, high frequency divisionThe energy of amount is example, according to formula Calculate fault high fdrequency component i_h1、i_h2、i_h3Energy P₁、P₂、P₃; And by high fdrequency component treating capacity P₁、P₂、P₃Together as fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: build machine learning---the support vector that comprises fault signature, fault initial angle, transition resistanceMachine---input vector;

The training of step 6: machine learning---SVMs---: by inner protected circuit CB and perimeter(information of each fault sample comprises that fault signature, fault are initial for fault sample under typical fault position, typical fault conditionThe input vector of angle, transition resistance composition) composition typical fault sample set; Application typical fault sample set is to machine learning---SVMs---train;

Step 7: fault judgement: comprise fault signature, fault initial angle, transition resistance in physical fault situation are formedVector be input in the machine learning---SVMs---having trained, machine learning---SVMs---Whether automatic decision fault is in protection zone;

The beneficial effect of technical scheme 1 is that the information of all high fdrequency components is fully used, and reliability improves.

As the technical scheme 2 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) guard method of multiple high fdrequency component treating capacities, neutral net, fault initial angle and a circuit of transition resistance, it is right to protectResemble as the circuit CB in Fig. 1, can comprise following step:

Step 1: extract the above multiple height of fault current signal (or failure voltage signal) two or three or threeFrequency component:, at the C of protected circuit CB end, constantly gather fault current i₁(or false voltage u₁), will simulate through analog-to-digital conversionSignal is converted to data signal, and the current signal transferring to after data signal is still used i₁Represent that (or voltage signal is still used u₁TableShow), extract fault current i₁The high fdrequency component i of 3 frequency ranges_h1、i_h2、i_h3(or false voltage u₁The high fdrequency component of 3 frequency rangesu_h1、u_h2、u_h3)；

Step 2: calculate the high fdrequency component treating capacity of the above multiple high fdrequency components of two or three or three, and with this twoIndividual or more than three or three multiple high fdrequency component treating capacities build fault signatures: by high fdrequency component i_h1、i_h2、i_h3(or u_h1、u_h2、u_h3) calculate their high fdrequency component treating capacity, be expressed as P₁、P₂、P₃; With the one of high component treating capacity, high frequencyThe energy of component is example, according to formula Calculate fault high fdrequency component i_h1、i_h2、i_h3Energy P₁、P₂、P₃; And by high fdrequency component treating capacity P₁、P₂、P₃Together as fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: build the machine learning---neutral net---that comprises fault signature, fault initial angle, transition resistanceInput vector;

The training of step 6: machine learning---neutral net---: the allusion quotation by protected circuit CB inside with perimeter(information of each fault sample comprises that fault signature, fault are initial for fault sample under type abort situation, typical fault conditionThe input vector of angle, transition resistance composition) the input vector composition typical fault sample set of composition; Application typical fault sample setTo machine learning,---neutral net---trains;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning---neutral net---having trained, machine learning---neutral net---Whether automatic decision fault is in protection zone;

The beneficial effect of technical scheme 2 is that the information of all high fdrequency components is fully used, and reliability improves.

As the technical scheme 3 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) a high fdrequency component treating capacity, fault initial angle and the method for a circuit of transition resistance protection, object of protection is Fig. 1In circuit CB, can comprise following step:

Step 1: the high fdrequency component of extracting fault current signal (or failure voltage signal): at the C of protected circuit CBEnd, constantly gathers fault current i₁(or false voltage u₁), through analog-to-digital conversion, analog signal is converted to data signal, transfer number toCurrent signal after word signal is still used i₁Represent that (or the voltage signal transferring to after data signal is still used u₁Represent), extract eventBarrier current i₁High fdrequency component i_h(or false voltage u₁High fdrequency component u_h)；

Step 2: calculate the high fdrequency component treating capacity of high fdrequency component, and build fault spy with this high fdrequency component treating capacityLevy: by high fdrequency component i_h(or u_h) calculate its high fdrequency component treating capacity, be expressed as P₁; With the one of high component treating capacity, heightThe energy of frequency component is example, according to formulaCalculate fault high fdrequency component i_hEnergy; With fault high fdrequency component placeReason amount P₁As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: the input vector that builds the machine learning that comprises fault signature, fault initial angle, transition resistance;

Step 6: the training of machine learning: typical fault position, typical case by protected circuit CB inside with perimeter(information of each fault sample comprises that fault signature, fault initial angle, transition resistance form to fault sample under fault conditionInput vector) the input vector composition typical fault sample set of composition; Application typical fault sample set is instructed machine learningPractice;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, and whether machine learning automatic decision fault is in protection zone.

As the technical scheme 4 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) a high fdrequency component instantaneous amplitude integration (or instantaneous amplitude and or instantaneous amplitude), fault initial angle and transition resistance protectProtect the method for a circuit, object of protection is the circuit CB in Fig. 1, can comprise following step:

Step 1: the high fdrequency component of extracting fault current signal (or failure voltage signal): at the C of protected circuit CBEnd, constantly gathers fault current i₁(or false voltage u₁), through analog-to-digital conversion, analog signal is converted to data signal, transfer number toCurrent signal after word signal is still used i₁Represent that (or voltage signal is still used u₁Represent);

Step 2: the high fdrequency component treating capacity or instantaneous amplitude integration or instantaneous amplitude and or the instantaneous width that calculate high fdrequency componentValue, and with this high fdrequency component treating capacity or instantaneous amplitude integration or instantaneous amplitude and or instantaneous amplitude build fault signature: withOne in high component treating capacity, high fdrequency component energy instantaneous amplitude integration is example, application Hilbert-Huang transform calculates eventBarrier current i₁(or false voltage u₁) the instantaneous amplitude IA of high fdrequency component, by formulaBy IA one section timeBetween on carry out integration and obtain the instantaneous amplitude integration IOIA of high fdrequency component, using fault high fdrequency component instantaneous amplitude integration IOIA asFault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: typical fault position, typical case by protected circuit CB inside with perimeter(information of each fault sample comprises that fault signature, fault initial angle, transition resistance form to fault sample under fault conditionInput vector) input vector composition typical fault sample set; Application typical fault sample set is trained machine learning;

As the technical scheme 5 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) entropy, fault initial angle and the method for a circuit of transition resistance protection, object of protection is the circuit CB in Fig. 1, can wrapContain following step:

Step 1: extract fault current signal (or failure voltage signal):, at the C of protected circuit CB end, constantly gatherFault current i₁(or false voltage u₁), through analog-to-digital conversion, analog signal is converted to data signal, transfer the electricity after data signal toStream signal is still used i₁Represent that (or voltage signal is still used u₁Represent);

Step 2: calculate the entropy of fault-signal, and build fault signature with this entropy: calculate fault current signal i₁(faultVoltage signal u₁) entropy, such as according to formulaCalculate signal i₁Entropy E(s), s wherein_nRepresent signal i₁Projection coefficient in an Orthogonal Wavelets; And using the entropy E (s) of fault-signal as faultFeature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: typical fault position, typical case by protected circuit CB inside with perimeter(information of each fault sample comprises that fault signature, fault initial angle, transition resistance form to fault sample under fault conditionInput vector) composition typical fault sample set; Application typical fault sample set is trained machine learning;

As the technical scheme 6 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) the difference, fault initial angle of two high fdrequency component treating capacities and the method for a circuit of transition resistance protection, object of protectionFor the circuit CB in Fig. 1, can comprise following step:

Step 1: two high fdrequency components extracting fault current signal (or failure voltage signal): at protected circuit CBC end, constantly gather fault current i₁(or false voltage u₁), through analog-to-digital conversion, analog signal is converted to data signal, transfer toCurrent signal after data signal is still used i₁Represent that (or the voltage signal transferring to after data signal is still used u₁Represent), extractFault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2(or false voltage u₁The high fdrequency component u of 2 frequency ranges_h1、u_h2)；

Step 2: calculate the difference of high fdrequency component treating capacity of two high fdrequency components, and with the difference of high fdrequency component treating capacityBuild fault signature: by high fdrequency component i_h1、i_h2(or u_h1、u_h2) calculate their high fdrequency component treating capacity, be expressed as P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculatei_h1、i_h2Energy P₁、P₂, wherein n is positive integer, desirable 200; Further calculate the difference of two high fdrequency component treating capacitiesdiff₁₂＝P₁-P₂, and by high fdrequency component treating capacity difference diff₁₂As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: typical fault position, typical case by protected circuit CB inside with perimeter(it is defeated that the information of each fault sample comprises that fault signature, fault initial angle, transition resistance form to the fault sample of fault conditionIncoming vector) the input vector composition typical fault sample set of composition; Application typical fault sample set is trained machine learning;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, and whether machine learning automatic decision fault at protected circuitOn;

The beneficial effect of technical scheme 6 is that the information of multiple high fdrequency components is fully used, and reliability improves.

As the technical scheme 7 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) two high fdrequency component treating capacity ratios, fault initial angle and the method for a circuit of transition resistance protection, object of protection isCircuit CB in Fig. 1, can comprise following step:

Step 1: two high fdrequency components extracting fault current signal (or failure voltage signal): at protected circuit CBC end, constantly gather fault current i₁(or false voltage u₁), through analog-to-digital conversion, analog signal is converted to data signal, transfer toCurrent signal after data signal is still used i₁Represent that (or the voltage signal transferring to after data signal is still used u₁Represent), extractFault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2(or extract false voltage u₁The high fdrequency component u of 2 frequency ranges_h1、u_h2)；

Step 2: calculate the ratio of high fdrequency component treating capacity of two high fdrequency components, and with the ratio of high fdrequency component treating capacityBuild fault signature: by high fdrequency component i_h1、i_h2(or u_h1、u_h2) calculate their high fdrequency component treating capacity, be expressed as P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculatei_h1、i_h2Energy P₁、P₂; Further calculate the ratio R atio of two high fdrequency component treating capacities₁₂＝P₁/P₂, and by high fdrequency componentTreating capacity ratio R atio₁₂As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, and whether machine learning automatic decision fault is in protection zone;

The beneficial effect of technical scheme 7 is that the information of multiple high fdrequency components is fully used, and reliability improves.

As the technical scheme 8 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) two high fdrequency component treating capacities and the ratio of two high fdrequency component treating capacities, fault initial angle and transition resistance protection, a method for circuit, object of protection is the circuit CB in Fig. 1, can comprise following step:

Step 1: two high fdrequency components extracting fault current signal (or failure voltage signal) from protected circuit:

At the C of protected circuit CB end, constantly gather fault current i₁(or electric fault is pressed u₁), through analog-to-digital conversion by mouldAnalog signal is converted to data signal, and the current signal transferring to after data signal is still used i₁Represent (or to transfer to after data signalVoltage signal is still used u₁Represent), extract fault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2(or false voltage u₁2The high fdrequency component u of individual frequency range_h1、u_h2)；

Step 2: calculate the treating capacity of two high fdrequency components and the ratio of two high fdrequency component treating capacities, and with these twoThe ratio of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: calculate high fdrequency component i_h1、i_h2(or u_h1、u_h2) high fdrequency component treating capacity, be expressed as P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example,According to formulaCalculate i_h1、i_h2Energy P₁、P₂; Further calculate two high fdrequency component processingThe ratio R atio of amount₁₂＝P₁/P₂, and by P₁、P₂、Ratio₁₂As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

The beneficial effect of technical scheme 8 is that, when fault high fdrequency component is passed through bus, a large amount of high fdrequency components are divided over the ground by busCloth equivalent capacity by-pass shunt falls, and more these principles of the higher shunting of frequency have obtained complete embodiment simultaneously, and reliability is higher.

As the technical scheme 9 of circuit bus bar protecting method of the present invention, based on two fault current signal (or false voltagesSignal) high fdrequency component treating capacity and difference, fault initial angle and the transition electricity of the high fdrequency component treating capacity of two fault-signalsBus of resistance protection and the method that is connected in two circuits on this bus, object of protection is circuit CB, CD and the mother in Fig. 1Line C, can comprise following step:

Step 1: fault current signal (or the false voltage that extracts protected two circuits that are connected in same busSignal) high fdrequency component: the top that returns back out line CB and CD at two of bus C respectively constantly gathers fault current i₁、i₂(or thereforeBarrier voltage u₁、u₂), through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still usedi₁、i₂Represent that (or the voltage signal transferring to after data signal is still used u₁、u₂Represent), extract fault current i₁、i₂High frequency divisionAmount i_1h、i_2h(or false voltage u₁、u₂High fdrequency component u_1h、u_2h)；

Step 2: calculate the treating capacity of two high fdrequency components and the difference of two high fdrequency component treating capacities, and with these twoThe difference of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: by high fdrequency component i_1h、i_2h(or u_1h、u_2h) calculate their high fdrequency component treating capacity P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example,According to formulaCalculate i_1h、i_2hEnergy P₁、P₂, wherein n is positive integer, n desirable 200; FurtherCalculate the difference diff of high fdrequency component treating capacity₁₂＝P₁-P₂, and by P₁、P₂、diff₁₂Together as fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: by typical fault position, the typical fault of the inside of protected circuit CB, CD and bus C and perimeter(information of each fault sample comprises the input of fault signature, fault initial angle, transition resistance composition to fault sample under conditionVector) the input vector composition typical fault sample set of composition; Application typical fault sample set is trained machine learning;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, the region of machine learning automatic decision guilty culprit, failure judgementBe to occur in circuit CB above, or CD is upper, or bus C is upper, or outside protection zone;

The beneficial effect of technical scheme 9 is that protection zone is 3 times of conventional method, can greatly save investment like this, pressesConventional method configuration protection equipment can be realized dual protection and invest without increasing.

As the technical scheme 10 of circuit bus bar protecting method of the present invention, based on two fault current signals (or fault electricityPress signal) high fdrequency component treating capacity difference, fault initial angle and transition resistance protection be connected in two on same busThe method of circuit, object of protection is two circuit CB that are connected in same bus C and the CD in Fig. 1, can comprise following stepRapid:

Step 1: of respectively extracting fault current signal (or failure voltage signal) from protected two circuits is highFrequency component: the top that returns back out line CB and CD at two of bus C respectively constantly gathers fault current i₁、i₂(or false voltage u₁、u₂), through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still used i₁、i₂Represent(or the voltage signal transferring to after data signal is still used u₁、u₂Represent), extract fault current i₁、i₂High fdrequency component i_1h、i_2h(or false voltage u₁、u₂High fdrequency component u_1h、u_2h)；

Step 2: the high fdrequency component treating capacity of two fault-signals that calculation procedure 1 is extracted and two high fdrequency component treating capacitiesDifference, and build fault signature with the difference of two high fdrequency component treating capacities: by high fdrequency component i_1h、i_2h(or u_1h、u_2h) calculateGo out their high fdrequency component treating capacity P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the difference diff of high fdrequency component treating capacity₁₂＝P₁-P₂, and by high fdrequency component treating capacity difference diff₁₂As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by the typical fault position of the inside of protected circuit CB and CD and perimeterPut, (information of each fault sample comprises fault signature, fault initial angle, transition electricity for fault sample under typical fault conditionThe input vector of resistance composition) the input vector composition typical fault sample set of composition; Application typical fault sample set is to engineeringHabit is trained;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, the region of machine learning automatic decision guilty culprit, failure judgementBe to occur in circuit CB above, or CD is upper, or outside protection zone;

The beneficial effect of technical scheme 10 is that protection zone is 2 times of conventional method, can greatly save investment like this, pressesConventional method configuration protection equipment can be realized dual protection and invest without increasing.

As the technical scheme 11 of circuit bus bar protecting method of the present invention, based on two fault current signals (or fault electricityPress signal) high fdrequency component treating capacity difference, fault initial angle and transition resistance as one of fault direction judgment component protectionThe method of circuit, object of protection is the circuit CB in Fig. 1, can comprise following step:

Step 1: each fault current signal (or false voltage letter that extracts from being connected in two circuits of same busNumber) high fdrequency component: the top that returns back out line CB and CD at two of bus C respectively constantly gathers fault current i₁、i₂(or faultVoltage u₁、u₂), through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still usedi₁、i₂Represent that (or the voltage signal transferring to after data signal is still used u₁、u₂Represent), extract fault current i₁、i₂High frequency divisionAmount i_1h、i_2h(or false voltage u₁、u₂High fdrequency component u_1h、u_2h)；

Step 2: calculate the difference of high fdrequency component treating capacity of two fault-signals, and with two high fdrequency component treating capacitiesDifference, as the fault signature of failure judgement direction, builds fault signature: by high fdrequency component i_1h、i_2h(or u_1h、u_2h) calculate itHigh fdrequency component treating capacity P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the difference diff of high fdrequency component treating capacity₁₂＝P₁-P₂, and by the high fdrequency component treating capacity difference diff for failure judgement direction₁₂As the fault spy of failure judgement directionLevy;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, machine learning automatic decision fault direction; If local terminal protection is sentencedDisconnected fault direction is positive direction, to opposite end send " forward fault " signal, otherwise to opposite end send " negative sense fault " signal orPerson is not to opposite end transmitted signal; If local terminal protection failure judgement direction is being for just, and receive that opposite end protection sends " forward is formerBarrier " signal, be judged as protection zone internal fault, otherwise be judged to protection external area error.

As the technical scheme 12 of circuit bus bar protecting method of the present invention, based on two fault current signals (or fault electricityPress signal) high fdrequency component treating capacity ratio, fault initial angle and transition resistance as one of fault direction judgment component protectionThe method of circuit, object of protection is the circuit CB in Fig. 1, can comprise following step:

Step 1: each fault current signal (or false voltage letter that extracts from being connected in two circuits of same busNumber) high fdrequency component: the top that returns back out line CB and CD at two of bus C respectively constantly gathers fault current i₁、i₂(or faultVoltage u₁、u₂), through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still usedi₁、i₂Represent that (or the voltage signal transferring to after data signal is still used u₁、u₂Represent), extract fault current i₁、i₂High frequency divisionAmount i_1h、i_2h(or voltage u₁、u₂High fdrequency component u_1h、u_2h)；

Step 2: calculate the ratio of high fdrequency component treating capacity of two fault-signals, and with two high fdrequency component treating capacitiesRatio, as the fault signature of failure judgement direction, builds fault signature: by high fdrequency component i_1h、i_2h(or u_1h、u_2h) calculate itHigh fdrequency component treating capacity P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the ratio R atio of high fdrequency component treating capacity₁₂＝P₁/P₂, and by the high fdrequency component treating capacity ratio R atio for failure judgement direction₁₂As the fault of failure judgement directionFeature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

As the technical scheme 13 of circuit bus bar protecting method of the present invention, based on all outlet fault current signals of busThe method that high fdrequency component treating capacity, fault initial angle and the transition resistance of (or failure voltage signal) signal protects bus, it is right to protectResemble as the bus C in Fig. 1, can comprise following step:

Step 1: of extracting respectively fault current signal (or failure voltage signal) from all outlets of bus is highFrequency component: the top that is circuit CB, CD, CG at the top of all branch roads that is connected in bus C respectively constantly gathers fault electricityStream i₁、i₂、i₃(or false voltage u₁、u₂、u₃), through analog-to-digital conversion, analog signal is converted to data signal, transfer data signal toAfter current signal still use i₁、i₂、i₃Represent that (or the voltage signal transferring to after data signal is still used u₁、u₂、u₃Represent), carryGet fault current i₁、i₂、i₃High fdrequency component i_1h、i_2h、i_3h(or false voltage u₁、u₂、u₃High fdrequency component u_1h、u_2h、u_3h)；

Step 2: calculate the high fdrequency component treating capacity of the fault-signal of every circuit, and with the fault-signal of all circuitsHigh fdrequency component treating capacity builds fault signature: by high fdrequency component i_1h、i_2h、i_3h(or u_1h、u_2h、u_3h) calculate their high frequency divisionAmount treating capacity P₁、P₂、P₃; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculate i_1h、i_2h、i_3hEnergy P₁、P₂、P₃; And by P₁、P₂、P₃Together as eventBarrier feature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by protected circuit CB, CD, CG and the inside of bus C and the allusion quotation of perimeter(information of each fault sample comprises that fault signature, fault are initial for fault sample under type abort situation, typical fault conditionThe input vector of angle, transition resistance composition) the input vector composition typical fault sample set of composition; Application typical fault sample setMachine learning is trained;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, and machine learning is automatic accurate determines whether in protected bus districtFault.

As the technical scheme 14 of circuit bus bar protecting method of the present invention, based on the fault current signal of an outlet of busThe high fdrequency component treating capacity of (or failure voltage signal) and row ripple direction, fault initial angle and a bus of transition resistance protectionWith the method for a circuit that is connected in this bus, object of protection is circuit CB and the bus C in Fig. 1, can comprise following stepRapid:

Step 1: extract the high fdrequency component of fault current signal (or voltage signal) from an outlet of bus: at quiltThe C end of protection circuit CB, constantly gathers fault current i₁(or false voltage u₁), through analog-to-digital conversion, analog signal is converted to numberWord signal, the current signal transferring to after data signal is still used i₁Represent that (or the voltage signal transferring to after data signal is still used u₁Represent), extract fault current i₁High fdrequency component i_h(or voltage u₁High fdrequency component u_h)；

Step 2: calculate high fdrequency component treating capacity, the failure judgement row wave line of propagation of fault-signal, and with fault-signalHigh fdrequency component treating capacity and fault traveling wave direction build fault signature: by high fdrequency component i_h(or u_h) calculate its high fdrequency componentTreating capacity P₁; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculating fault is highFrequency component i_hEnergy, wherein n is positive integer; Regulation positive direction is that row ripple is transmitted to circuit by bus, i.e. fault direction dirct quiltJust be defined as, represent to be dirct=1 with 1; Negative direction is that row ripple is transmitted to bus by circuit, and this fault direction is confirmed as bearing,Represent to be dirct=2 with 2; By fault high fdrequency component energy P₁, fault direction dirct is together as fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by protected circuit CB and the inside of bus C and the typical fault of perimeter(information of each fault sample comprises fault signature, fault initial angle, transition to fault sample under position, typical fault conditionThe input vector of resistance composition) the input vector composition typical fault sample set of composition; Application typical fault sample set is to machineStudy is trained;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, the region of machine learning automatic decision guilty culprit, failure judgementTo occur in circuit CB above, or on bus C, or outside protection zone;

The beneficial effect of technical scheme 14 is that the method is to belong to bus distribution protecting method, only need be from a circuitFault current in extraction fault transient information, without just carrying out fault judgement with other protected location exchange messages,Each protected location based on this principle has independence completely, and protected location breaks down or be out of service, noAffect the normal operation of other protected location, this protection is applicable to being arranged on switchyard scene, meets protective relaying device and dispersesTransfer the development trend requirement that is installed to switchyard; And protection zone is 2 times of conventional method.

As the technical scheme 15 of circuit bus bar protecting method of the present invention, based on fault current signal (or false voltage letterNumber) two high fdrequency component treating capacities and the difference of two high fdrequency component treating capacities, fault initial angle and transition resistance protection, a method for circuit, object of protection is the circuit CB in Fig. 1, can comprise following step:

Step 1: two high fdrequency components extracting fault current signal (or failure voltage signal) from protected circuit: existThe C end of protected circuit CB, constantly gathers fault current i₁(or electric fault is pressed u₁), through analog-to-digital conversion, analog signal is converted toData signal, the current signal transferring to after data signal is still used i₁Represent (or to transfer voltage signal after data signal to stillUse u₁Represent), extract fault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2(or false voltage u₁The high frequency of 2 frequency rangesComponent u_h1、u_h2)；

Step 2: calculate the treating capacity of two high fdrequency components and the difference of two high fdrequency component treating capacities, and with these twoThe difference of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: calculate high fdrequency component i_h1、i_h2(or u_h1、u_h2) high fdrequency component treating capacity, be expressed as P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example,According to formulaCalculate i_h1、i_h2Energy P₁、P₂; Further calculate two high fdrequency component processingThe difference diff of amount₁₂＝P₁-P₂, and by P₁、P₂、diff₁₂As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement: comprise fault signature, fault initial angle, transition resistance in physical fault situation are formedVector be input in the machine learning having trained, whether machine learning automatic decision fault in protection zone;

As the technical scheme 16 of circuit bus bar protecting method of the present invention, based on two fault current signals (or fault electricityPress signal) high fdrequency component treating capacity and the ratio of two high fdrequency component treating capacities, fault initial angle and transition resistance protection oneBar bus and the method that is connected in two circuits on this bus, object of protection is circuit CB, CD and the bus C in Fig. 1, canComprise following step:

Step 1: fault current signal (or the false voltage that extracts protected two circuits that are connected in same busSignal) high fdrequency component: the top that returns back out line CB and CD at two of bus C respectively constantly gathers fault current i₁、i₂(or thereforeBarrier voltage u₁、u₂), extract fault current i₁、i₂High fdrequency component i_1h、i_2h(or false voltage u₁、u₂High fdrequency component u_1h、u_2h)；

Step 2: calculate the treating capacity of two high fdrequency components and the ratio of two high fdrequency component treating capacities, and with these twoThe ratio of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: by high fdrequency component i_1h、i_2h(or u_1h、u_2h) calculate their high fdrequency component treating capacity P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example,According to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the ratio of high fdrequency component treating capacityValue Ratio₁₂＝P₁/P₂, and by P₁、P₂、Ratio₁₂Together as fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: by typical fault position, the typical fault of the inside of protected circuit CB, CD and bus C and perimeter(information of each fault sample comprises the input of fault signature, fault initial angle, transition resistance composition to fault sample under conditionVector) composition typical fault sample set; Application typical fault sample set is trained machine learning;

Step 7: fault judgement: comprise fault signature, fault initial angle, transition resistance in physical fault situation are formedVector be input in the machine learning having trained, the region of machine learning automatic decision guilty culprit, failure judgement is to send outRaw on circuit CB, or CD is upper, or bus C is upper, or outside protection zone;

As the technical scheme 17 of circuit bus bar protecting method of the present invention, based on two fault current signals (or fault electricityPress signal) high fdrequency component treating capacity ratio, fault initial angle and transition resistance protection be connected in two on same busThe method of circuit, object of protection is two circuit CB that are connected in same bus C and the CD in Fig. 1, can comprise following stepRapid:

Step 2: the high fdrequency component treating capacity of two fault-signals that calculation procedure 1 is extracted and two high fdrequency component treating capacitiesRatio, and build fault signature with the ratio of two high fdrequency component treating capacities: by high fdrequency component i_1h、i_2h(or u_1h、u_2h) calculateGo out their high fdrequency component treating capacity P₁、P₂; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the ratio R atio of high fdrequency component treating capacity₁₂＝P₁/P₂, and by high fdrequency component treating capacity ratio R atio₁₂As fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by the typical fault position of the inside of protected circuit CB and CD and perimeterPut, (information of each fault sample comprises fault signature, fault initial angle, transition electricity for fault sample under typical fault conditionThe input vector of resistance composition) composition typical fault sample set; Application typical fault sample set is trained machine learning;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition resistance in physical fault situationThe vector of composition is input in the machine learning having trained, the region of machine learning automatic decision guilty culprit, failure judgementBe to occur in circuit CB above, or CD is upper, or outside protection zone. As the technical scheme of circuit bus bar protecting method of the present invention18, high fdrequency component treating capacity, the fault of the fault current signal (or failure voltage signal) of all outlets based on a busInitial angle and transition resistance protect the method for bus, and object of protection is by the bus C in Fig. 1 is connected all circuits, under can comprisingState step:

Step 1: extract respectively one of fault current signal (or failure voltage signal) from all outlets of a busIndividual high fdrequency component: the top that is circuit CB, CD, CG at the top of all branch roads that is connected in bus C respectively constantly gather thereforeBarrier current i₁、i₂、i₃(or false voltage u₁、u₂、u₃), through analog-to-digital conversion, analog signal is converted to data signal, transfer numeral toCurrent signal after signal is still used i₁、i₂、i₃Represent that (or the voltage signal transferring to after data signal is still used u₁、u₂、u₃TableShow), extract fault current i₁、i₂、i₃High fdrequency component i_1h、i_2h、i_3h(or false voltage u₁、u₂、u₃High fdrequency component u_1h、u_2h、u_3h)；

Step 2: calculate the high fdrequency component treating capacity of the fault-signal of every circuit, and with the fault-signal of all circuitsHigh fdrequency component treating capacity builds fault signature: by high fdrequency component i_1h、i_2h、i_3h(or u_1h、u_2h、u_3h) calculate their high frequency divisionAmount treating capacity P₁、P₂、P₃; With the one of high component treating capacity, the energy of high fdrequency component is example, according to formulaCalculate i_1h、i_2h、i_3hEnergy P₁、P₂、P₃; And by P₁、P₂、P₃Conduct togetherFault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by the inside of protected circuit CB, CD, CG and perimeter (perimeter asBus C, B, G, D, circuit AB, DE) typical fault position typical fault condition under the fault sample (letter of each fault sampleBreath comprises the input vector of fault signature, fault initial angle, transition resistance composition) composition typical fault sample set; Application typical caseThe machine learning of fault sample set pair is trained;

Step 7: fault judgement: comprise fault signature, fault initial angle, transition resistance in physical fault situation are formedVector be input in the machine learning having trained, machine learning is automatic accurate judges whether that fault is to occur in circuit CBUpper, or on circuit CD, or on circuit CG, or outside protection zone.

Just the fault initial angle of the circuit bus bar protecting method based on fault initial angle transition resistance and machine learning is calculatedMethod, the present invention is that the method at the calculating primary fault angle that solve the technical problem comprises the following steps:

Step 1: calculate the fault distance L of trouble point to protection installation place;

Step 2: the fault angle θ of detection failure moment protection installation place₀；

Step 3: the phase coefficient α of computational scheme;

Step 4: calculate fault initial angle θ_f，θ_f＝θ₀-α·L。

The beneficial effect of fault initial angle computational methods of the present invention is, the error of calculation of fault initial angle is very little, forOverhead transmission line, under general power frequency condition, phase coefficient α ≈ 0.06 spend/km, the error of calculating fault distance L is in 200m time,The error of calculation of fault initial angle is in 0.012 degree, and the error of calculating fault distance L is in 100m time, fault initial angleThe error of calculation in 0.006 degree.

Brief description of the drawings

Shown in Fig. 1 is 500kV EHV transmission network schematic diagram, and wherein, circuit AB is long is 180km, and circuit BC length is342km, circuit CD is long is 360km, and circuit DE length is 266km, and circuit CG length is 270km; S₁＝35GVA，S₂＝10GVA，S₃＝20GVA，S₄＝5GVA，S₅=12GVA; Line parameter circuit value X₁、R₁、C₁、X₀、R₀、C₀Be respectively X₁＝0.2783Ω/km，R₁＝0.0270Ω/km，X₀＝0.6494Ω/km，R₀＝0.1948Ω/km，C₁＝0.0127μF/km，C₀=0.0090 μ F/km; RespectivelyBus direct-to-ground capacitance is 6000PF.

Detailed description of the invention

Main with the one in the high fdrequency component treating capacity of fault-signal---the high fdrequency component of fault current signal belowEnergy is example, by reference to the accompanying drawings the specific embodiment of the present invention is described, but embodiments of the present invention is not limited to this.

The embodiment 1 (corresponding technical scheme 1) of circuit bus bar protecting method of the present invention, multiple high based on fault-signalFrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example), SVMs, fault initial angle and transition electricityThe method of a circuit of resistance protection, object of protection is the circuit CB in Fig. 1 (a), can comprise following step:

Step 1: extract two or three of fault current signal or more than three multiple high fdrequency components: protectedThe C end of circuit CB, constantly gathers fault current i₁, through analog-to-digital conversion, analog signal is converted to data signal, transfer numeral letter toCurrent signal after number is still used i₁Represent, extract fault current i₁The high fdrequency component i of 3 frequency ranges_h1、i_h2、i_h3; Sample frequencyDesirable 200kHz, fault current high fdrequency component i_h1、i_h2、i_h3Desirable 12.5kHz～25kHz, 25kHz～50kHz, 50kHz respectively～100kHz；

Step 2: calculate the high fdrequency component treating capacity of the above multiple high fdrequency components of two or three or three, and with this twoIndividual or more than three or three multiple high fdrequency component treating capacities build fault signatures: high fdrequency component treating capacity is with high fdrequency component energyAmount is for example, according to formula Calculate fault high fdrequency component i_h1、i_h2、i_h3EnergyP₁、P₂、P₃, and by P₁、P₂、P₃Together as fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is the input vector of---SVMs---as machine learning; Fault zone label can value be suitable real number;Regulation in the present embodiment, when troubles inside the sample space, label=1, fault zone, when external area error, label=2, fault zone;

The training of step 6: machine learning---SVMs---: by inner protected circuit CB and perimeterUnder the typical fault condition of typical fault position, comprise fault signature, fault initial angle, transition resistance, fault type, faulty sectionThe input vector composition typical fault sample set of territory label composition; Application typical fault sample set is to machine learning---support toAmount machine---train; The A phase ground short circuit of the present embodiment, the concentrated P of fault sample of machine learning training₁、P₂、P₃AsShown in table 1～4;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input to the machine learning---SVMs---having trainedIn, the region of machine learning---SVMs---automatic decision guilty culprit; In physical fault deterministic process, canSo that fault zone label is first assumed to 2, be protection external area error, the judgement of fault zone is with machine learning---support toBeing as the criterion of amount machine---judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB, P₁、P₂、P₃Be respectively 2071500,469682,325111, fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, judgementCorrectly;

The upper F of circuit CB under table 1 different faults condition₁Point be l apart from bus C₁P when fault₁，P₂And P₃Maximum

The upper F of circuit AB under table 2 different faults condition₂Point be l apart from bus B₂P when fault₁，P₂And P₃Maximum

The upper F of circuit CD under table 3 different faults condition₃Point be l apart from bus C₃P when fault₁，P₂And P₃Maximum

The upper F of circuit CG under table 4 different faults condition₄Point be l apart from bus C₄P when fault₁，P₂And P₃Maximum

The embodiment 2 (corresponding technical scheme 2) of circuit bus bar protecting method of the present invention, multiple high based on fault-signalFrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example), neutral net, fault initial angle and transition resistanceThe method of a circuit of protection, object of protection is the circuit CB in Fig. 1 (a), can comprise following step:

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is the input vector of---neutral net---as machine learning; Fault zone label can value be suitable real number; ?In the present embodiment, specify, when troubles inside the sample space, label=1, fault zone, when external area error, label=2, fault zone;

The training of step 6: machine learning---neutral net---: the allusion quotation by protected circuit CB inside with perimeterUnder type abort situation typical fault condition, comprise fault signature, fault initial angle, transition resistance, fault type, fault zoneThe input vector composition typical fault sample set of label composition; Application typical fault sample set is to machine learning---nerve netNetwork---train; The A phase ground short circuit of the present embodiment, the concentrated P of fault sample of machine learning training₁、P₂、P₃As table 1Shown in～4;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning---neutral net---having trained,The region of machine learning---neutral net---automatic decision guilty culprit; In physical fault deterministic process, can be by eventBarrier region label is first assumed to 2, is protection external area error, and the judgement of fault zone is with machine learning---neutral net---Being as the criterion of judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB, P₁、P₂、P₃Be respectively 2071500,469682,325111, fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, correct judgment.

The embodiment 3 (corresponding technical scheme 3) of circuit bus bar protecting method of the present invention is high based on of fault-signalFrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example), fault initial angle and a line of transition resistance protectionThe method on road, object of protection is the circuit CB in Fig. 1 (a), can comprise following step:

Step 1: the high fdrequency component of extracting fault current signal:, at the C of protected circuit CB end, constantly gather fault electricityStream i₁, through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still used i₁Represent,Extract fault current i₁High fdrequency component i_h; The desirable 200kHz of sample frequency, fault current high fdrequency component i_hDesirable 50kHz～100kHz；

Step 2: calculate the high fdrequency component treating capacity of high fdrequency component, and build fault spy with this high fdrequency component treating capacityLevy: high fdrequency component treating capacity is taking high fdrequency component energy as example, according to formulaCalculate fault high fdrequency component i_hEnergyAmount, with the energy P of fault high fdrequency component₁As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment, districtWhen internal fault, label=1, fault zone, when external area error, label=2, fault zone;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated P of fault sample of machine learning training₁As shown in table 1～4;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,Fault initial angle is 45 degree, P₁Be 2071500, transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, judgement justReally.

The embodiment 4 (corresponding technical scheme 4) of circuit bus bar protecting method of the present invention is high based on of fault-signalThe method of frequency component instantaneous amplitude integration, fault initial angle and a circuit of transition resistance protection, object of protection is in Fig. 1 (a)Circuit CB, can comprise following step:

Step 1: the high fdrequency component of extracting fault current signal:, at the C of protected circuit CB end, constantly gather fault electricityStream i₁, through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still used i₁Represent;The desirable 200kHz of sample frequency;

Step 2: the high fdrequency component treating capacity or instantaneous amplitude integration or instantaneous amplitude and or the instantaneous width that calculate high fdrequency componentValue, and with this high fdrequency component treating capacity or instantaneous amplitude integration or instantaneous amplitude and or instantaneous amplitude build fault signature: withThe one of high component treating capacity, high fdrequency component energy instantaneous amplitude integration is example, application Hilbert-Huang transform calculates faultCurrent i₁The instantaneous amplitude IA of high fdrequency component, by formulaIA being carried out on a period of time to integration obtainsTo the instantaneous amplitude integration IOIA of high fdrequency component, using fault high fdrequency component instantaneous amplitude integration IOIA as fault signature; Wherein τFor integration time constant number, desirable 0.001s;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated IOIA of the fault sample of machine learning training is as shown in table 5～8;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,IOIA is 2053, and fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, judgement justReally;

The upper F of circuit CB under table 5 different faults condition₁Point be l apart from bus C₁The maximum of IOIA when fault

The upper F of circuit AB under table 6 different faults condition₂Point be l apart from bus B₂The maximum of IOIA when fault

The upper F of circuit CD under table 7 different faults condition₃Point be l apart from bus C₃The maximum of IOIA when fault

The upper F of circuit CG under table 8 different faults condition₄Point be l apart from bus C₄The maximum of IOIA when fault

The embodiment 5 (corresponding technical scheme 5) of circuit bus bar protecting method of the present invention, the entropy based on fault-signal, eventThe method of barrier initial angle and a circuit of transition resistance protection, object of protection is the circuit CB in Fig. 1 (a), can comprise followingStep:

Step 1: extract fault current signal:, at the C of protected circuit CB end, constantly gather fault current i₁, through mouldAnalog signal is converted to data signal by number conversion, and the current signal transferring to after data signal is still used i₁Represent; Sample frequencyDesirable 200kHz;

Step 2: calculate the entropy of fault-signal, and build fault signature with this entropy: calculate the entropy of fault-signal, such as rootAccording to formulaCalculate signal i₁Entropy E (s), wherein s_nRepresent signal i₁Projection coefficient in an Orthogonal Wavelets; And using the entropy E (s) of fault-signal as fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated E (s) of the fault sample of machine learning training is as shown in table 9～12;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,Fault initial angle is 45 degree, and E (s) is 5794, and transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, judgement justReally;

The upper F of circuit CB under table 9 different faults condition₁Point be l apart from bus C₁The maximum of E (s) when fault

The upper F of circuit AB under table 10 different faults condition₂Point be l apart from bus B₂The maximum of E (s) when fault

The upper F of circuit CD under table 11 different faults condition₃Point be l apart from bus C₃The maximum of E (s) when fault

The upper F of circuit CG under table 12 different faults condition₄Point be l apart from bus C₄The maximum of E (s) when fault

The embodiment 6 (corresponding technical scheme 6) of circuit bus bar protecting method of the present invention is high based on two of fault-signalFrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) difference, fault initial angle and transition resistance protection oneThe method of bar circuit, object of protection is the circuit CB in Fig. 1 (a), can comprise following step:

Step 1: two high fdrequency components extracting fault current signal:, at the C of protected circuit CB end, constantly gather eventBarrier current i₁, through analog-to-digital conversion, analog signal being converted to data signal, the current signal transferring to after data signal is still used i₁TableShow, extract fault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2; The desirable 200kHz of sample frequency, transient current high fdrequency componenti_h1、i_h2Desirable 25kHz～50kHz, 50kHz～100kHz respectively;

Step 2: calculate the difference of high fdrequency component treating capacity of two high fdrequency components, and with the difference of high fdrequency component treating capacityBuild fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example, according to formulaMeterCalculate i_h1、i_h2Energy P₁、P₂; Further calculate the difference diff of two high-frequency energies₁₂＝P₁-P₂, and by high-frequency energy differencediff₁₂As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated diff of fault sample of machine learning training₁₂As shown in table 13～16;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,diff₁₂Be 111800, fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, judgementCorrectly;

The upper F of circuit CB under table 13 different faults condition₁Point be l apart from bus C₁Diff when fault₁₂Maximum

The upper F of circuit AB under table 14 different faults condition₂Point be l apart from bus B₂Diff when fault₁₂Maximum

The upper F of circuit CD under table 15 different faults condition₃Point be l apart from bus C₃Diff when fault₁₂Maximum

The upper F of circuit CG under table 16 different faults condition₄Point be l apart from bus C₄Diff when fault₁₂Maximum

The embodiment 7 (corresponding technical scheme 7) of circuit bus bar protecting method of the present invention is high based on two of fault-signalFrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) ratio, fault initial angle and transition resistance protection oneThe method of bar circuit, object of protection is the circuit CB in Fig. 1 (a), can comprise following step:

Step 2: calculate the ratio of high fdrequency component treating capacity of two high fdrequency components, and with the ratio of high fdrequency component treating capacityBuild fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example, according to formulaMeterCalculate i_h1、i_h2Energy P₁、P₂; Further calculate the ratio R atio of two high-frequency energies₁₂＝P₁/P₂, and by high-frequency energy ratioRatio₁₂As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodimentShort circuit, the concentrated Ratio of fault sample of machine learning training₁₂As shown in table 17～20;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,Ratio₁₂Be 214730, fault initial angle is 45 degree, transition resistance is 50 Europe, judged result is: circuit CB troubles inside the sample space, sentenceDisconnected correct;

The upper F of circuit CB under table 17 different faults condition₁Point be l apart from bus C₁Ratio when fault₁₂Maximum

The upper F of circuit AB under table 18 different faults condition₂Point be l apart from bus B₂Ratio when fault₁₂Maximum

The upper F of circuit CD under table 19 different faults condition₃Point be l apart from bus C₃Ratio when fault₁₂Maximum

The upper F of circuit CG under table 20 different faults condition₄Point be l apart from bus C₄Ratio when fault₁₂Maximum

The embodiment 8 (corresponding technical scheme 8) of circuit bus bar protecting method of the present invention is high based on two of fault-signalRatio, the event of frequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) and two high fdrequency component treating capacitiesThe method of barrier initial angle and a circuit of transition resistance protection, object of protection is the circuit CB in Fig. 1 (a), can comprise followingStep:

Step 1: two high fdrequency components extracting fault current signal from protected circuit: at the C of protected circuit CBEnd, constantly gathers fault current i₁, through analog-to-digital conversion, analog signal is converted to data signal, transfer the electric current after data signal toSignal is still used i₁Represent, extract fault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2; The desirable 200kHz of sample frequency, thereforeBarrier electric current high fdrequency component i_h1、i_h2Desirable 25kHz～50kHz, 50kHz～100kHz respectively;

Step 2: calculate the treating capacity of two high fdrequency components and the ratio of two high fdrequency component treating capacities, and with these twoThe ratio of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: high fdrequency component treating capacity is with high fdrequency componentEnergy is example, according to formula Calculate i_h1、i_h2Energy P₁、P₂; Further calculate two high frequenciesThe ratio R atio of energy₁₂＝P₁/P₂, and by P₁、P₂、Ratio₁₂As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated P of fault sample of machine learning training₁、P₂Shown in table 1～4, Ratio₁₂As shown in table 17～20;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,P₁、P₂Be respectively 2071500,469682, Ratio₁₂Be 214730, fault initial angle is 45 degree, and transition resistance is 50 Europe, judgementResult is: circuit CB troubles inside the sample space, correct judgment;

The embodiment 9 (corresponding technical scheme 9) of circuit bus bar protecting method of the present invention, based on the height of two fault-signalsFrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) and both differences, fault initial angle and transitionBus of resistance protection and be connected in the method for two circuits on this bus, object of protection be circuit CB in Fig. 1 (a),CD and bus C, can comprise following step:

Step 1: the high fdrequency component of extracting the fault current signal of two circuits that are connected in same bus: exist respectivelyThe top that two of bus C returns back out line CB and CD constantly gathers fault current i₁、i₂, through analog-to-digital conversion, analog signal is converted to numberWord signal, the current signal transferring to after data signal is still used i₁、i₂Represent, extract fault current i₁、i₂High fdrequency component i_1h、i_2h; The desirable 200kHz of sample frequency, fault current high fdrequency component i_1h、i_2hDesirable 50kHz～100kHz;

Step 2: calculate the treating capacity of two high fdrequency components and the difference of two high fdrequency component treating capacities, and with these twoThe difference of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: high fdrequency component treating capacity is with high fdrequency componentEnergy is example, according to formula Calculate i_1h、i_2hEnergy P₁、P₂; Further calculate high-frequency energyDifference diff₁₂＝P₁-P₂, and by P₁、P₂、diff₁₂Together as fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment: quiltWhen protection circuit CB troubles inside the sample space, label=1, fault zone; When protected circuit CD troubles inside the sample space, label=2, fault zone;When the upper fault of protected bus C, label=3, fault zone; When protection external area error, label=4, fault zone;

Step 6: the training of machine learning: by protected circuit CB, CD and the inside of bus C and the typical case of perimeter eventUnder barrier position typical fault condition, comprise fault signature, fault initial angle, transition resistance, fault type, fault zone labelThe input vector composition typical fault sample set of composition; Application typical fault sample set is trained machine learning; This enforcementThe A phase ground short circuit of example, the concentrated P of fault training sample of machine learning₁、P₂、diff₁₂As shown in table 21～26;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 4, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,P₁、P₂、diff₁₂Be respectively 305862,728.2,305320, fault initial angle is 45 degree, and transition resistance is 50 Europe, judged resultFor: circuit CB troubles inside the sample space, correct judgment;

The upper F of circuit CB under table 21 different faults condition₁Point be l apart from bus C₁P when fault₁、P₂、P₃And diff₁₂Maximum

The upper F of circuit AB under table 22 different faults condition₂Point be l apart from bus B₂P when fault₁、P₂And diff₁₂?Large value

The upper F of circuit CD under table 23 different faults condition₃Point be l apart from bus C₃P when fault₁、P₂And diff₁₂?Large value

The upper F of circuit DE under table 24 different faults condition₅Point be l apart from bus D₅P when fault₁、P₂And diff₁₂?Large value

The upper F of table 25 different faults condition Down Highway C₆P when point failure₁、P₂、P₃And diff₁₂Maximum

The upper F of circuit CG under table 26 different faults condition₄Point be l apart from bus C₄P when fault₁、P₂And diff₁₂?Large value

The embodiment 10 (corresponding technical scheme 10) of circuit bus bar protecting method of the present invention, based on two fault-signalsHigh fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) difference, fault initial angle and transition resistance protectionArticle two, the method for circuit, object of protection is circuit CB and the CD in Fig. 1 (a), can comprise following step:

Step 1: each high fdrequency component of extracting fault current signal from protected two circuits: respectively at bus CTwo tops that return back out line CB and CD constantly gather fault current i₁、i₂, through analog-to-digital conversion, analog signal is converted to numeral letterNumber, the current signal transferring to after data signal is still used i₁、i₂Represent, extract fault current i₁、i₂High fdrequency component i_1h、i_2h; AdoptThe desirable 200kHz of sample frequency, the high fdrequency component i of fault current signal_1h、i_2hDesirable 50kHz～100kHz;

Step 2: calculate the high fdrequency component treating capacity of two fault-signals and the difference of two high fdrequency component treating capacities, and withThe difference of two high fdrequency component treating capacities builds fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example, i.e. basisFormulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the difference diff of high-frequency energy₁₂＝P₁-P₂, and by high-frequency energy difference diff₁₂As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment: quiltWhen protection circuit CB troubles inside the sample space, label=1, fault zone; When protected circuit CD troubles inside the sample space, label=2, fault zone;When protection external area error, label=3, fault zone;

Step 6: the training of machine learning: by the typical fault position allusion quotation of protected circuit CB and CD inside and perimeterUnder type fault condition, comprise the defeated of fault signature, fault initial angle, transition resistance, fault type, fault zone label compositionIncoming vector composition typical fault sample set; Application typical fault sample set is trained machine learning; The A of the present embodiment joinsGround short circuit, the concentrated diff of fault sample of machine learning training₁₂As shown in table 21～24,26;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 3, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,diff₁₂Be 305320, fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB troubles inside the sample space, judgementCorrectly;

The embodiment 11 (corresponding technical scheme 11) of circuit bus bar protecting method of the present invention, based on two fault-signalsHigh fdrequency component treating capacity (high fdrequency component treating capacity is using high fdrequency component energy as example) difference, fault initial angle and transition resistance asThe method of a circuit of fault direction judgment component protection, object of protection is the circuit CB in Fig. 1 (a), can comprise following stepRapid:

Step 1: each high fdrequency component of extracting fault current signal from being connected in two circuits of same bus: divideThe top that does not return back out line CB and CD at two of bus C constantly gathers fault current i₁、i₂, through analog-to-digital conversion, analog signal is changedFor data signal, the current signal transferring to after data signal is still used i₁、i₂Represent, extract fault current i₁、i₂High fdrequency componenti_1h、i_2h; The desirable 200kHz of sample frequency, fault current high fdrequency component i_1h、i_2hDesirable 50kHz～100kHz;

Step 2: calculate the difference of high fdrequency component treating capacity of two fault-signals, and with two high fdrequency component treating capacitiesDifference is as the fault signature of failure judgement direction, builds fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example,According to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the difference of high-frequency energydiff₁₂＝P₁-P₂, and by the high-frequency energy difference diff for failure judgement direction₁₂As the fault spy of failure judgement directionLevy; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment: quiltWhen protection circuit CB positive direction fault, label=1, fault zone; When protected circuit CB reverse direction failure, fault zone label＝2；

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated diff of fault sample of machine learning training₁₂As shown in table 21～23,26;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionFault direction; In physical fault deterministic process, fault zone label first can be assumed to 2, be protection reverse direction failure,The judgement of fault direction is as the criterion with machine learning judgement; If local terminal protection failure judgement direction is positive direction, to opposite endSend " forward fault " signal, otherwise to opposite end transmission " negative sense fault " signal or not to opposite end transmitted signal; If local terminalProtection failure judgement direction is being for just, and receives that opposite end protects " forward fault " signal of sending, is judged as in protection zone formerBarrier, otherwise be judged to protection external area error; The situation of the present embodiment is: the upper A phase earth fault of circuit CB, diff₁₂Be 305320, thereforeBarrier initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB positive direction fault, correct judgment;

The embodiment 12 (corresponding technical scheme 12) of circuit bus bar protecting method of the present invention, based on two fault-signalsHigh fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) ratio, fault initial angle and transition resistance are doneFor the method for a circuit of fault direction judgment component protection, object of protection is the circuit CB in Fig. 1 (a), can comprise followingStep:

Step 2: calculate the ratio of high fdrequency component treating capacity of two fault-signals, and with two high fdrequency component treating capacitiesRatio is as the fault signature of failure judgement direction, builds fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example,According to formulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the ratio of high-frequency energyRatio₁₂＝P₁/P₂, and by the high-frequency energy ratio R atio for failure judgement direction₁₂As the fault of failure judgement directionFeature; Wherein n is positive integer, desirable 200; In the time of CB positive direction fault, P₁Much larger than P₂, Ratio₁₂Much larger than 1; When CB anti-When direction fault, P₁Much smaller than P₂, Ratio₁₂Much smaller than 1;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by protected circuit CB, CD, CG and the inside of bus C and the allusion quotation of perimeterUnder type abort situation typical fault condition, comprise fault signature, fault initial angle, transition resistance, fault type, fault zoneThe input vector composition typical fault sample set of label composition; Application typical fault sample set is trained machine learning;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionFault direction; In physical fault deterministic process, fault zone label first can be assumed to 2, be protection reverse direction failure,The judgement of fault direction is as the criterion with machine learning judgement; If local terminal protection failure judgement direction is positive direction, to opposite endSend " forward fault " signal, otherwise to opposite end transmission " negative sense fault " signal or not to opposite end transmitted signal; If local terminalProtection failure judgement direction is being for just, and receives that opposite end protects " forward fault " signal of sending, is judged as in protection zone formerBarrier, otherwise be judged to protection external area error; The situation of the present embodiment is: the upper A phase earth fault of circuit CB, Ratio₁₂Much larger than 1, thereforeBarrier initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: circuit CB positive direction fault, correct judgment.

The embodiment 13 (corresponding technical scheme 13) of circuit bus bar protecting method of the present invention, based on all outlets of busHigh fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example), fault initial angle and the transition electricity of fault-signalThe method of resistance protection bus, object of protection is the bus C in Fig. 1 (a), can comprise following step:

Step 1 a: high fdrequency component extracting respectively fault current signal from all outlets of bus: connecting respectivelyThe top that is connected to all branch roads of bus C is that the top of circuit CB, CD, CG constantly gathers fault current i₁、i₂、i₃, turn through modulusThe analog signal of changing commanders is converted to data signal, and the current signal transferring to after data signal is still used i₁、i₂、i₃Represent, extract faultCurrent i₁、i₂、i₃High fdrequency component i_1h、i_2h、i_3h; The desirable 200kHz of sample frequency, fault current high fdrequency component i_1h、i_2h、i_3hCanGet 50kHz～100kHz;

Step 2: calculate the high fdrequency component treating capacity of the fault-signal of every circuit, and with the fault-signal of all circuitsHigh fdrequency component treating capacity builds fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example, according to formulaCalculate i_1h、i_2h、i_3hEnergy P₁、P₂、P₃; And by P₁、P₂、P₃Together as eventBarrier feature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment: quiltWhen the upper fault of protection bus C, label=1, fault zone; When region fault outside protected bus C, fault zone label=2；

Step 6: the training of machine learning: by protected circuit CB, CD, CG and the inside of bus C and the allusion quotation of perimeterUnder type abort situation typical fault condition, comprise fault signature, fault initial angle, transition resistance, fault type, fault zoneThe input vector composition typical fault sample set of label composition; Application typical fault sample set is trained machine learning; ThisThe A phase ground short circuit of embodiment, the concentrated P of fault training sample of machine learning₁、P₂、P₃As shown in table 21,23,25,26;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of bus C,P₁、P₂、P₃Difference 305862,728.2,728.8, fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: busC troubles inside the sample space, correct judgment.

The embodiment 14 (corresponding technical scheme 14) of circuit bus bar protecting method of the present invention, based on an outlet of busAt the beginning of the high fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) of fault-signal and row ripple direction, faultBeginning angle and transition resistance are protected a bus and are connected in the method for a circuit of this bus, and object of protection is in Fig. 1 (a)Circuit CB and bus C, can comprise following step:

Step 1: extract the high fdrequency component of fault current signal from an outlet of bus: protected circuit CB'sC end, constantly gathers fault current i₁, through analog-to-digital conversion, analog signal is converted to data signal, transfer the electricity after data signal toStream signal is still used i₁Represent, extract fault current i₁High fdrequency component i_h; The desirable 200kHz of sample frequency, fault current high frequencyComponent i_hDesirable 50kHz～100kHz;

Step 2: calculate high fdrequency component treating capacity, the failure judgement row wave line of propagation of fault-signal, and with fault-signalHigh fdrequency component treating capacity and fault traveling wave direction build fault signature: high fdrequency component treating capacity taking high fdrequency component energy as example,According to formulaCalculate fault high fdrequency component i_hEnergy, wherein n is positive integer; Regulation positive direction is that row ripple is by busBe transmitted to circuit, negative direction is that row ripple is transmitted to bus by circuit, i⁺And i^-Be respectively the direct wave along positive direction on circuit, andAlong reciprocal backward-travelling wave, signal S₁And S₂Represent respectively amplitude and and the backward-travelling wave of the direct wave that monitorsAmplitude and, and be expressed asWherein τ is the length time of integration, defines a ratioValue isIf ratio λ is less than threshold value k₀(0<k₀< 2), fault direction dirct is just confirmed as, and represents to be with 1Dirct=1, otherwise this fault direction is confirmed as bearing, and represents to be dirct=2 with 2; By fault high fdrequency component energy P₁, thereforeDirct is together as fault signature for barrier direction; In the present embodiment, the desirable 0.001s of integration time constant τ;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment, motherWhen the upper fault of line C, label=1, fault zone, when the upper fault of circuit CB, label=2, fault zone, when external area error, faulty sectionLabel=3, territory;

Step 6: the training of machine learning: by protected circuit CB and the inside of bus C and the typical fault of perimeterUnder the typical fault condition of position, comprise fault signature, fault initial angle, transition resistance, fault type, fault zone set of tagsThe input vector composition typical fault sample set becoming; Application typical fault sample set is trained machine learning; The present embodimentA phase ground short circuit, the concentrated P of fault training sample of machine learning₁As shown in table 21～23,25,26; Female in the present embodimentWhen the upper fault of line C, circuit CD, circuit CG, fault direction dirct is being for just, i.e. dirct=1, and when circuit CB fault, fault directionDirct is negative, i.e. dirct=2;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 3, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of circuit CB,P₁Be 305862, fault direction is being for just, fault initial angle is 45 degree, transition resistance is 50 Europe, judged result is: in circuit CB districtFault, correct judgment. The embodiment 15 (corresponding technical scheme 15) of circuit bus bar protecting method of the present invention, based on fault-signalTwo high fdrequency component treating capacities (high fdrequency component treating capacity is taking high fdrequency component energy as example) and two high fdrequency component treating capacitiesDifference, fault initial angle and the method for a circuit of transition resistance protection, object of protection is the circuit CB in Fig. 1 (a), canComprise following step:

Step 1: two high fdrequency components extracting fault current signal from protected circuit: at the C of protected circuit CBEnd, constantly gathers fault current i₁, through analog-to-digital conversion, analog signal is converted to data signal, transfer the electric current after data signal toSignal is still used i₁Represent, extract fault current i₁The high fdrequency component i of 2 frequency ranges_h1、i_h2; The desirable 200kHz of sample frequency,Fault current high fdrequency component i_h1、i_h2Desirable 25kHz～50kHz, 50kHz～100kHz respectively;

Step 2: calculate the treating capacity of two high fdrequency components and the difference of two high fdrequency component treating capacities, and with these twoThe difference of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: high fdrequency component treating capacity is with high fdrequency componentEnergy is example, according to formula Calculate i_h1、i_h2Energy P₁、P₂; Further calculate two high frequenciesThe difference diff of energy₁₂＝P₁-P₂, and by P₁、P₂、diff₁₂As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: the typical event in typical fault position protected circuit CB is inner and perimeterThe input that comprises fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under barrier condition toAmount composition typical fault sample set; Application typical fault sample set is trained machine learning; The A phase ground connection of the present embodiment is shortRoad, the concentrated P of fault sample of machine learning training₁、P₂Shown in table 1～4, diff₁₂As shown in table 13～16;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 2, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: circuit CB troubles inside the sample space, judgementCorrectly;

The embodiment 16 (corresponding technical scheme 16) of circuit bus bar protecting method of the present invention, based on two fault-signalsHigh fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) and both ratio, fault initial angle and mistakeCross bus of resistance protection and the method that is connected in two circuits on this bus, object of protection is the circuit in Fig. 1 (a)CB, CD and bus C, can comprise following step:

Step 2: calculate the treating capacity of two high fdrequency components and the ratio of two high fdrequency component treating capacities, and with these twoThe ratio of high fdrequency component treating capacity and two high fdrequency component treating capacities builds fault signature: high fdrequency component treating capacity is with high fdrequency componentEnergy is example, according to formula Calculate i_1h、i_2hEnergy P₁、P₂; Further calculate high-frequency energyRatio R aito₁₂＝P₁/P₂, and by P₁、P₂、Ratio₁₂Together as fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by protected circuit CB, CD and the inside of bus C and the typical case of perimeter eventUnder barrier position typical fault condition, comprise fault signature, fault initial angle, transition resistance, fault type, fault zone labelThe input vector composition typical fault sample set of composition; Application typical fault sample set is trained machine learning;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 4, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: circuit CB troubles inside the sample space, judgementCorrectly;

The embodiment 17 (corresponding technical scheme 17) of circuit bus bar protecting method of the present invention, based on two fault-signalsHigh fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example) ratio, fault initial angle and transition resistance protectionArticle two, the method for circuit, object of protection is circuit CB and the CD in Fig. 1 (a), can comprise following step:

Step 2: calculate the high fdrequency component treating capacity of two fault-signals and the ratio of two high fdrequency component treating capacities, and withThe ratio of two high fdrequency component treating capacities builds fault signature: high fdrequency component treating capacity is taking high fdrequency component energy as example, i.e. basisFormulaCalculate i_1h、i_2hEnergy P₁、P₂; Further calculate the ratio R atio of high-frequency energy₁₂＝P₁/P₂, and by high-frequency energy ratio R atio₁₂As fault signature; Wherein n is positive integer, desirable 200;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 6: the training of machine learning: by the typical fault position allusion quotation of protected circuit CB and CD inside and perimeterUnder type fault condition, comprise the defeated of fault signature, fault initial angle, transition resistance, fault type, fault zone label compositionIncoming vector composition typical fault sample set; Application typical fault sample set is trained machine learning;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 3, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: circuit CB troubles inside the sample space, judgementCorrectly;

The embodiment 18 (corresponding technical scheme 18) of circuit bus bar protecting method of the present invention, based on all outlets of busHigh fdrequency component treating capacity (high fdrequency component treating capacity is taking high fdrequency component energy as example), fault initial angle and the transition electricity of fault-signalThe method of resistance protection bus, object of protection is outlet CB, CD, the CG of the bus C in Fig. 1 (a), can comprise following step:

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: will comprise that fault signature, fault initial angle, transition resistance, fault type, fault zone label formVector is as the input vector of machine learning; Fault zone label can value be suitable real number; Regulation in the present embodiment: thereforeWhen barrier occurs to occur in respectively circuit CB, CD, CG, corresponding fault zone label, is respectively 1,2,3; Event outside protected areaWhen barrier, label=4, fault zone;

Step 6: the training of machine learning: by the typical fault position of the inside of protected circuit CB, CD, CG and perimeterPut, comprise fault signature, fault initial angle, transition resistance, fault type, fault zone label composition under typical fault conditionInput vector composition typical fault sample set; Application typical fault sample set is trained machine learning; The A of the present embodimentPhase ground short circuit, the concentrated P of fault sample of machine learning training₁、P₂、P₃As shown in table 21,23,25,26;

Step 7: physical fault judgement: will comprise fault signature, fault initial angle, transition electricity in physical fault situationThe vector of resistance, fault type, fault zone label composition is input in the machine learning having trained, machine learning automatic decisionThe region of guilty culprit; In physical fault deterministic process, fault zone label first can be assumed to 4, be outside protection zoneFault, the judgement of fault zone is as the criterion with machine learning judgement; The situation of the present embodiment is: the upper A phase earth fault of bus C,P₁、P₂、P₃Difference 305862,728.2,728.8, fault initial angle is 45 degree, and transition resistance is 50 Europe, and judged result is: protectionExternal area error, correct judgment.

The present invention is about the embodiment of the method at calculating primary fault angle, as the upper F of circuit BC₁Point failure, calculates initial eventBarrier angle comprises the following steps:

Step 1: calculate trouble point F₁To the fault distance L of protection installation place, obtain L=341km;

Step 2: the fault angle θ of detection failure moment protection installation place₀, obtain θ₀=65.88 degree;

Step 3: the phase coefficient α of computational scheme,Wherein x₁、b₁Be respectively circuitThe reactance of unit length and susceptance;

Step 4: calculate fault initial angle θ_f，θ_f＝θ₀-α L=65.88-0.06123*341 ≈ 45 spends.

By reference to the accompanying drawings the preferred technical scheme of the present invention and embodiment are described in detail above, but the invention is not restricted toTechnique scheme and embodiment, in the ken possessing, can also not depart from the present invention those skilled in the artUnder the prerequisite of design, make a variety of changes. Fault high fdrequency component treating capacity can comprise high frequency for other except described in aboveThe amount of information, the structure of fault signature can multiple version, thinks that other except explanation is described comprises high-frequency informationFault signature. The feature of maximum of the present invention and Spirit Essence are based on fault initial angle, transition resistance and machine learning intelligenceThe situation of ground failure judgement, eliminates fault initial angle and the impact of transition resistance on transient protection, comprises reliability to improve. ItsIt is any does not deviate from change, the modification done under Spirit Essence of the present invention and principle, substitute, combination, simplify the skill obtainingArt scheme, is equivalent substitute mode.

Claims

1. the circuit bus bar protecting method based on fault initial angle transition resistance and machine learning, the method comprises following stepRapid:

(1) extract fault high fdrequency component;

(2) build the fault signature that comprises fault high-frequency information;

(3) discriminating fault types and Fault Phase Selection;

(4) utilize fault signature to carry out fault judgement;

It is characterized in that: utilizing before fault signature carries out fault judgement, calculate transition resistance and fault initial angle; To compriseThe vector of fault signature, fault initial angle, transition resistance composition, as the input vector of machine learning, is first used typical fault sampleSet pair machine learning is trained, then the application training machine learning failure judgement of having got well; Described method can comprise following stepRapid:

Step 1: the high fdrequency component of extracting fault-signal;

Step 2: build the fault signature that comprises fault high-frequency information;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: structure comprises the input vector of the machine learning of fault signature, fault initial angle, transition resistance;

Step 6: the training of machine learning, application typical fault sample set is trained machine learning; Machine learning trainsAfter, carrying out fault while judging, this step can be omitted, skip;

Step 7: fault judgement, fault judgement is carried out in the machine learning that application training has been got well.

2. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the multiple high frequency division of fault-signalThe method of high fdrequency component treating capacity, SVMs, fault initial angle and a circuit of transition resistance protection of amount, described methodCan comprise following step:

Step 1: two or three of the fault-signal of extraction protected circuit or three above multiple high fdrequency components;

Step 2: calculate the high fdrequency component treating capacity of the above multiple high fdrequency components of two or three or three, and with these two orThree or three above multiple high fdrequency component treating capacities structure fault signatures;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: structure comprise fault signature, fault initial angle, the machine learning of transition resistance or the input of SVMs toAmount;

Step 6: the training of machine learning or SVMs, application typical fault sample set is to machine learning or SVMsTrain; After machine learning trains, carrying out fault while judging, this step can be omitted, skip;

Step 7: fault judgement, whether the machine learning that application training has been got well or SVMs failure judgement be in protection zone.

3. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the multiple high frequency division of fault-signalThe method of high fdrequency component treating capacity, neutral net, fault initial angle and a circuit of transition resistance protection of amount, described method canTo comprise following step:

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 5: structure comprises fault signature, fault initial angle, the machine learning of transition resistance or the input vector of neutral net;

Step 6: the training of machine learning or neutral net, application typical fault sample set carries out machine learning or neutral netTraining; After machine learning trains, carrying out fault while judging, this step can be omitted, skip;

Step 7: fault judgement, whether the machine learning that application training has been got well or neutral net failure judgement be in protection zone.

4. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on a high frequency division of fault-signalThe method of high fdrequency component treating capacity, fault initial angle and a circuit of transition resistance protection of amount, under described method can compriseState step:

Step 1 a: high fdrequency component extracting the fault-signal of protected circuit;

Step 2: calculate the high fdrequency component treating capacity of high fdrequency component, and build fault signature with this high fdrequency component treating capacity;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, whether the machine learning failure judgement that application training has been got well is in protection zone.

5. according to the method for circuit bus protection described in claim 1, it is characterized in that: the high fdrequency component wink based on fault-signalTime amplitude integration or instantaneous amplitude and or the method for instantaneous amplitude, fault initial angle and a circuit of transition resistance protection, described inMethod can comprise following step:

Step 1: the high fdrequency component of extracting the fault-signal of protected circuit;

Step 2: calculate the high fdrequency component treating capacity of high fdrequency component or instantaneous amplitude integration or instantaneous amplitude and or instantaneous amplitude, andWith this high fdrequency component treating capacity or instantaneous amplitude integration or instantaneous amplitude and or instantaneous amplitude build fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

6. according to the method for circuit bus protection described in claim 1, it is characterized in that: at the beginning of entropy based on fault-signal, faultThe method of beginning angle and a circuit of transition resistance protection, described method can comprise following step:

Step 1: the fault-signal that extracts protected circuit;

Step 2: calculate the entropy of fault-signal, and build fault signature with this entropy;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

7. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on two high frequency divisions of fault-signalThe method of high fdrequency component treating capacity difference, fault initial angle and a circuit of transition resistance protection of amount, described method can be wrappedContain following step:

Step 1: two high fdrequency components extracting the fault-signal of protected circuit;

Step 2: calculate the difference of the high fdrequency component treating capacity of two high fdrequency components, and build with the difference of high fdrequency component treating capacityFault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, whether the machine learning failure judgement that application training has been got well is on protected circuit.

8. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on two high frequency divisions of fault-signalThe method of high fdrequency component treating capacity ratio, fault initial angle and a circuit of transition resistance protection of amount, described method can be wrappedContain following step:

Step 2: calculate the ratio of the high fdrequency component treating capacity of two high fdrequency components, and build with the ratio of high fdrequency component treating capacityFault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

9. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on two high frequency divisions of fault-signalThe high fdrequency component treating capacity of amount, the ratio of two high fdrequency component treating capacities, fault initial angle and a circuit of transition resistance protectionMethod, described method can comprise following step:

Step 2: calculate the high fdrequency component treating capacity of two high fdrequency components and the ratio of two high fdrequency component treating capacities, and with thisThe ratio of two high fdrequency component treating capacities and two high fdrequency component treating capacities builds fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

10. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the high frequency of two fault-signalsDifference, fault initial angle and a mother of transition resistance protection of the high fdrequency component treating capacity of component treating capacity and two fault-signalsLine and the method that is connected in two circuits on this bus, described method can comprise following step:

Step 1: each high fdrequency component of extracting one of fault-signal from protected two circuits that are connected in same bus;

Step 2: the high fdrequency component treating capacity of two fault-signals that calculation procedure 1 is extracted and the high frequency of two fault-signalsThe difference of component treating capacity, and with the high fdrequency component treating capacity of these two fault-signals and the difference of two high fdrequency component treating capacitiesBuild fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, there is the region at place in the machine learning failure judgement that application training has been got well; Failure judgement is to send outRaw on protected bus, or on a protected circuit, or outside protection zone.

11. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the high frequency of two fault-signalsComponent treating capacity difference, fault initial angle and transition resistance protection are connected in the method for two circuits on same bus, instituteThe method of stating can comprise following step:

Step 1: each high fdrequency component extracting fault-signal from protected two circuits;

Step 2: the high fdrequency component treating capacity of two fault-signals that calculation procedure 1 is extracted and two high fdrequency component treating capacities poorValue, and build fault signature with the difference of two high fdrequency component treating capacities;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, there is the region at place in the machine learning failure judgement that application training has been got well; Failure judgement is to send outRaw on a protected circuit, or outside protection zone.

12. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the high frequency of two fault-signalsComponent treating capacity difference is as the method for fault direction judgment component, fault initial angle and a circuit of transition resistance protection, instituteThe method of stating can comprise following step:

Step 1: each high fdrequency component extracting fault-signal from being connected in two circuits of same bus;

Step 2: calculate the difference of high fdrequency component treating capacity of two fault-signals, and with the difference of two high fdrequency component treating capacitiesAs the fault signature of failure judgement direction;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, fault direction is judged in the machine learning that first application training has been got well, by the fault side of local terminal judgementTo sending opposite end to, receive the fault direction that opposite end sends simultaneously; Again according to the fault direction combination of local terminal protection judgementThe fault direction of opposite end protection judgement, comprehensively determines whether troubles inside the sample space.

13. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the high frequency of two fault-signalsComponent treating capacity ratio is as the method for fault direction judgment component, fault initial angle and a circuit of transition resistance protection, instituteThe method of stating can comprise following step:

Step 2: calculate the ratio of high fdrequency component treating capacity of two fault-signals, and with the ratio of two high fdrequency component treating capacitiesAs the fault signature of failure judgement direction;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

14. according to the method for circuit bus protection described in claim 1, it is characterized in that: the event of all outlets based on busThe method of high fdrequency component treating capacity, fault initial angle and the transition resistance protection bus of barrier signal, under described method can compriseState step:

Step 1: extract respectively high fdrequency component of fault-signal from all outlets of bus;

Step 2: calculate the high fdrequency component treating capacity of the fault-signal of every circuit, and with the high frequency of the fault-signal of all circuitsComponent treating capacity structure fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, the machine learning that application training has been got well determines whether protected bus troubles inside the sample space.

15. according to the method for circuit bus protection described in claim 1, it is characterized in that: the event of an outlet based on busHigh fdrequency component treating capacity and fault traveling wave direction, fault initial angle and the transition resistance of barrier signal are protected a bus and are connectedIn the method for a circuit of this bus, described method can comprise following step:

Step 1: extract a high fdrequency component of fault-signal from an outlet of bus;

Step 2: calculate high fdrequency component treating capacity, the failure judgement row wave line of propagation of fault-signal, and with the high frequency of fault-signalComponent treating capacity and fault traveling wave directional structure vectorical structure fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, the machine learning failure judgement region that application training has been got well; Failure judgement is to occur in instituteOn the bus of protection, or on protected circuit, or outside protection zone.

16. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on two high frequencies of fault-signalThe high fdrequency component treating capacity of component, the difference of two high fdrequency component treating capacities, fault initial angle and a line of transition resistance protectionThe method on road, described method can comprise following step:

Step 2: calculate the high fdrequency component treating capacity of two high fdrequency components and the difference of two high fdrequency component treating capacities, and with thisThe difference of two high fdrequency component treating capacities and two high fdrequency component treating capacities builds fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

17. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the high frequency of two fault-signalsRatio, fault initial angle and a mother of transition resistance protection of the high fdrequency component treating capacity of component treating capacity and two fault-signalsLine and the method that is connected in two circuits on this bus, described method can comprise following step:

Step 1: each high fdrequency component of extracting a fault-signal from protected two circuits that are connected in same bus;

Step 2: the high fdrequency component treating capacity of two fault-signals that calculation procedure 1 is extracted and the high frequency of two fault-signalsThe ratio of component treating capacity, and with the ratio of high fdrequency component treating capacity and two high fdrequency component treating capacities of these two fault-signalsBuild fault signature;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

18. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on the high frequency of two fault-signalsComponent treating capacity ratio, fault initial angle and transition resistance protection are connected in the method for two circuits on same bus, instituteThe method of stating can comprise following step:

Step 2: the high fdrequency component treating capacity of two fault-signals that calculation procedure 1 is extracted and the ratio of two high fdrequency component treating capacitiesValue, and build fault signature with the ratio of two high fdrequency component treating capacities;

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

19. according to the method for circuit bus protection described in claim 1, it is characterized in that: based on all outlets of a busThe high fdrequency component treating capacity, fault initial angle of fault-signal and the method for all outlets of transition resistance protection bus, described sideMethod can comprise following step:

Step 3: discriminating fault types and Fault Phase Selection;

Step 4: calculate transition resistance and fault initial angle;

Step 7: fault judgement, the machine learning failure judgement region that application training has been got well; Failure judgement is to occur in instituteOn a circuit of protection, or outside protection zone.

20. according to the method for the circuit bus protection described in claim 1 to 15 any one, it is characterized in that: machine learning isSVMs or neutral net or genetic algorithm or K arest neighbors k-NearestNeighbor-KNN sorting algorithm orK-Means algorithm or C4.5 algorithm, Apriori algorithm, greatest hope EM--Expectation – Maximization algorithm,PageRank algorithm, Adaboost algorithm, NaiveBayes algorithm, classification and regression tree CART--ClassificationandRegressionTrees algorithm.

21. according to the method for the circuit bus protection described in claim 1 to 15 any one, it is characterized in that: high fdrequency component placeReason amount be high fdrequency component energy or high fdrequency component instantaneous amplitude integration or high fdrequency component instantaneous amplitude and or high fdrequency component instantaneousAmplitude or high fdrequency component entropy or high fdrequency component complexity or high fdrequency component Singularity Degree or high fdrequency component modulus maximum.

22. according to the method for the circuit bus protection described in claim 1 to 15 any one, it is characterized in that: high fdrequency component isElectric current high fdrequency component or voltage high fdrequency component; Fault-signal is fault current signal or failure voltage signal; Fault high-frequency informationFor the high fdrequency component treating capacity of fault current signal or failure voltage signal or the ratio of high fdrequency component treating capacity or high frequency divisionAmount treating capacity difference or be the capable ripple direction of fault-signal or fault-signal direction or fault-signal high fdrequency component; Occur inFault sample composition typical fault under typical fault position, the typical fault condition of inside, protection zone, perimeter, protection zoneSample set, each fault sample information comprises the input vector of fault signature, fault initial angle, transition resistance composition.

23. 1 kinds for realizing a kind of computational methods of the fault initial angle of circuit bus bar protecting method described in claim 1, bagDraw together step:

Step 3: the phase coefficient α of computational scheme;

Step 4: calculate fault initial angle θ_f，θ_f＝θ₀-α·L。