CN110880062B - Method for determining state maintenance time of power distribution equipment - Google Patents
Method for determining state maintenance time of power distribution equipment Download PDFInfo
- Publication number
- CN110880062B CN110880062B CN201911049328.1A CN201911049328A CN110880062B CN 110880062 B CN110880062 B CN 110880062B CN 201911049328 A CN201911049328 A CN 201911049328A CN 110880062 B CN110880062 B CN 110880062B
- Authority
- CN
- China
- Prior art keywords
- equipment
- index
- cloud
- state
- health
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012423 maintenance Methods 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000011156 evaluation Methods 0.000 claims abstract description 51
- 230000036541 health Effects 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 28
- 239000013598 vector Substances 0.000 claims abstract description 25
- 230000007787 long-term memory Effects 0.000 claims abstract description 5
- 230000006403 short-term memory Effects 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 46
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000003862 health status Effects 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 230000015654 memory Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 5
- 230000002441 reversible effect Effects 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 238000007689 inspection Methods 0.000 claims description 2
- 230000002265 prevention Effects 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 5
- 201000004569 Blindness Diseases 0.000 description 3
- 238000006467 substitution reaction Methods 0.000 description 2
- YCKRFDGAMUMZLT-UHFFFAOYSA-N Fluorine atom Chemical compound [F] YCKRFDGAMUMZLT-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 229910052731 fluorine Inorganic materials 0.000 description 1
- 239000011737 fluorine Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for determining state maintenance time of power distribution equipment, belongs to the technical field of power engineering, and solves the problems that conventional equipment maintenance of a power distribution network is mainly regular maintenance and after-repair, the utilization efficiency of maintenance resources is low, and fault prevention cannot be performed according to different equipment differences. The method comprises the steps of carrying out health state grade division according to equipment comprehensive deduction values, applying an entropy weight method to weight equipment state evaluation indexes, generating a health state cloud picture of equipment to be evaluated through a cloud model, and calculating membership degrees between health state clouds of the equipment to be evaluated and health state grade clouds; and finally, inputting the membership vector, the times of failure, the operation records and other data into a long-term and short-term memory network for training, so as to realize the prediction of the next failure occurrence time of the equipment to be evaluated and reasonably plan a maintenance plan. And the reasonable planning of the operation equipment maintenance plan is realized according to the predicted fault occurrence time, the maintenance resources are effectively and reasonably utilized, and the equipment fault risk is reduced.
Description
Technical Field
The invention belongs to the technical field of electric power engineering, and particularly relates to a method for determining state maintenance time of power distribution equipment.
Background
The traditional equipment maintenance of distribution network is mostly regular maintenance and overhaul afterwards, overhauls the low and can't carry out the fault prevention to all kinds of equipment differences of resource utilization efficiency. Therefore, the state maintenance is carried out on the distribution equipment by utilizing the historical operating data and the test data of the distribution equipment, the maintenance blindness is avoided, the maintenance resources can be fully utilized, and the maintenance efficiency is improved.
Disclosure of Invention
The invention provides a method for determining the state maintenance time of power distribution equipment, which can solve the problems that the conventional equipment maintenance of a power distribution network is mostly regular maintenance and after-repair, the utilization efficiency of maintenance resources is low, and the fault prevention can not be performed according to the difference of various equipment.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for determining the state maintenance time of power distribution equipment is characterized by comprising the following steps: the method comprises the following steps of 1, classifying equipment according to the health degree: according to the comprehensive deduction value of the equipment, the health degree g of the z-th equipment with the same class and voltage class is countedz(Z is 1,2, …, Z), g according to health degreezThe device health status is ranked. Step 2, an entropy weight method determines index weight and solves membership and membership vector by using a cloud model: and after weighting the equipment state evaluation index by using an entropy weight method, generating a health state cloud picture of the equipment to be evaluated through a cloud model, and calculating the membership degree and the membership degree vector between the health state cloud of the equipment to be evaluated and each health state grade cloud. Step 3, training a power distribution equipment fault occurrence time prediction model based on the long-term and short-term memory network: firstly, the operation record indexes of the sample equipment are normalized to form a matrix BzThen the membership vector delta of the sample devicezNumber of failed times lzMatrix BzAs input samples, the LSTM neural network is input for training. Step 4, predicting the next fault occurrence time of the equipment to be evaluated, and determining the maintenance time: according to a membership vector delta ' ═ delta ' of the equipment to be evaluated '1,δ′2,δ′3,δ′4]Inputting a matrix consisting of the failed times l 'and the running record index normalization matrix B' into a trained LSTM neural network to calculate to obtain a predicted value T 'of the next failure occurrence time of the equipment to be evaluated'tAccording to the formula Tz=Tplan+Top+Ts<T′tDetermining a time to overhaul TzWherein, TplanFor planning maintenance time, TopTime required for inspection, TsA safety margin time.
Further, in step 1, according to the health degree gzThe equipment is divided into four grades of normal, attention, abnormal and serious, which are respectively usedf1、f2、f3、f4And (4) showing.
Further, by the formulaCloud number feature calculating 4 health status levels, afIs the minimum value of the health status grade interval, bfMaximum value of the health status grade interval, ExfIs an expected value of a device health level f, EnfEntropy, H, of the device health level fefSuper entropy, H, for a device health level fefTake 0.01, where f is 1,2,3,4, corresponding to four state levels f1、f2、f3、f4。
Further, the entropy weight method in step 2 includes the following steps, and firstly, an equipment state evaluation index matrix is constructed: an index matrix composed of m secondary indexes, which are n evaluation objects, is as follows, X ═ Xij)n×m=(X1,X2,...Xi,...,Xm),x={x1,x2,...,xi,...,xm},
Wherein, X is an index matrix constructed by n multiplied by m index values; xiThe evaluation index is the ith index column vector in the index matrix, namely the vector formed by the ith evaluation indexes of the n evaluation objects; x is a radical of a fluorine atomijThe j index value is the ith evaluation object; x is index set, xjIs the jth index in the index set; m is the index number; n is the number of evaluation objects; secondly, the equipment state evaluation index normalization processing: the positive indicators and the negative indicators are respectively normalized as follows,
thirdly, calculating the entropy value of each equipment state evaluation index: the formula for calculating the entropy of the jth evaluation index is as follows,
ejentropy for the jth evaluation index, k being 1/ln (n), pijIs the ratio of the score of the ith sample device on the jth index relative to the scores of all the objects to be evaluated on the index,
fourthly, calculating the entropy weight of each equipment state evaluation index: the entropy weight calculation formula of the jth evaluation index is as follows,
Further, the cloud model generating cloud picture in step 2 mainly comprises the following steps of firstly, calculating a reverse cloud generator: expectation of j-th secondary index ExjEntropy EnjHyper entropy HejThe formula of (c) is as follows, wherein P is the index sample number, xiThe index value is a secondary index value; the digital characteristic parameters of the cloud model of the level of the target layer are obtained by combining the digital characteristics of the cloud model of the related indexes of each level, the calculation formula is shown as follows,
second, a forward cloud generator is calculated: is characterized by a number ofC(Ex,En,He) The positive cloud generator randomly generates N cloud droplets (x)i,yi) The method comprises the following specific steps:
a, with EnTo expectation, HeGenerating a normal distribution random number E 'as a standard deviation'n;
b, with ExL E 'to expect |'nI is standard deviation to generate normal distribution random number xi;
c, by E'n、xiAs a variable, into the formula yi=exp(-(xi-Ex)2/2(E'n)2) Producing cloud droplets (x)i,yi);
And d, repeating the steps a to c until N cloud droplets are generated, and drawing a cloud model graph according to the N cloud droplets.
Furthermore, K cloud drops are arranged at the intersection point of the cloud picture of the sample device and the f-th level cloud picture, the mean value of the membership degree values of the K cloud drops is taken as the membership degree of the state value of the device, as shown in the formula,wherein, f is 1,2,3,4, the membership vector of the z-th sample device
Further, the LSTM training algorithm in step 3 is a back propagation algorithm, and includes three steps, first, calculating an output value of the LSTM memory module forward in combination with the weight matrix, second, calculating an error term of each memory module backward, and third, calculating a gradient of each weight matrix used in the first step according to the corresponding residual term, and updating the weight matrix.
Further, after the sample data of each sample device is input into the LSTM neural network for training, the model training is finished; or the model training is finished when the training error is set to be less than 1 e-6.
Further, the output value of the LSTM memory module, namely the true time of next fault occurrence of the sample equipment, is calculated in the forward directionValue TtIs as follows, ft=σ(Wfxxt+WfhΤt-1+bf),it=σ(Wixxt+WihTt-1+bi),ot=σ(Woxxt+WohTt-1+bo),Tt=ot·tanh(ct) (ii) a σ is sigmoid function, xt=[δz,lz,Bz]Is the input matrix at time t, Wfx、Wix、WoxRepresentation and current input xtWeight matrix of multiplications, Wfh、Wih、Woh、WchIndicating the output value T at the time T-1t-1A multiplied weight matrix. bf、bi、bc、boRespectively as the offset terms of the forgetting gate, the input gate, the state unit and the output gate, ft、it、otActivation functions of forgetting gate, input gate, output gate, respectively, ct、Is a vector of state units and instantaneous states; t istThe current output of the LSTM is the real value of the next failure occurrence time of the sample device.
Compared with the prior art, the invention has the beneficial effects that:
according to the method, historical operating data and test data of the power distribution equipment are processed through an entropy weight method and a cloud model, input into an LSTM neural network for training, processed according to corresponding index data of the equipment to be evaluated and input into the LSTM neural network to predict next fault occurrence time of the equipment to be evaluated and determine maintenance time, reasonable planning of an in-operation equipment maintenance plan is achieved, maintenance blindness is avoided, maintenance resources can be fully utilized, maintenance efficiency is improved, and equipment fault risks are reduced.
Drawings
FIG. 1 is a flow chart of a method for determining a condition maintenance time of a power distribution device in accordance with the present invention;
FIG. 2 is a state evaluation index system for a distribution transformer in accordance with the present invention;
FIG. 3 is a diagram of a distribution line condition evaluation index system according to the present invention;
FIG. 4 illustrates an LSTM memory module structure according to the present invention;
FIG. 5 is an expanded view of the timing of the LSTM network of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following detailed description is made with reference to the accompanying drawings and specific examples, and specific embodiments of the present invention are described in detail as follows:
the invention provides a method for determining the state maintenance time of power distribution equipment, the flow of the method is shown in figure 1, the state maintenance of the power distribution equipment in question is actually graded into health states according to the comprehensive deduction value of the equipment; after an entropy weight method is applied to weight the equipment state evaluation index, a health state cloud picture of the equipment to be evaluated is generated through a cloud model, and then the membership degree between the health state cloud of the equipment to be evaluated and each health state grade cloud is calculated; and finally, inputting the membership vector, the times of failure, the operation records and other data into a long-term and short-term memory network for training, so as to realize the prediction of the next failure occurrence time of the equipment to be evaluated. The method mainly comprises the following steps:
step 1: and carrying out health state grade division according to the equipment comprehensive deduction value.
According to the comprehensive deduction value of the equipment, the health degree g of the z-th equipment with the same class and voltage class is countedz(Z: 1,2, …, Z), g according to health degreezThe health grade of the equipment is divided into four grades of normal, attention, abnormity and severity, and f is used for the four grades1、f2、f3、f4Expressed that the corresponding value ranges are [80,100, respectively]、[60,80)、[40,60)、[0,40)。
Step 2: and after the entropy weight method is used for weighting the equipment state evaluation index, generating a health state cloud picture of the equipment to be evaluated through a cloud model, and calculating the membership degree between the health state cloud of the equipment to be evaluated and each health state grade cloud.
Normalizing each index of the sample equipment according to the formulas (4) and (5), inputting the normalized numerical value and the corresponding weight into a reverse cloud generator, calculating the cloud digital characteristics according to the formulas (10) to (13), and generating a cloud picture by the forward cloud generator.
According to the range of the device state value [0,1]Corresponding to the level of health of the apparatus, f1、f2、f3、f4The four state value ranges are respectively [0.8,1 ]]0.6,0.8), 0.4,0.6, 0.4). Calculating the cloud digital characteristics of 4 health state grades by the formula (1), and generating a corresponding cloud picture according to the step (2-2), afIs the minimum value of the health status grade interval, bfMaximum value of the health status grade interval, ExfExpected value of the device health status level f, EnfEntropy, H, of the device health level fefHyper-entropy of the device health level f, according to EnfSize of (A), determined by empirical and repetitive tests, HefTake 0.01. Wherein, f is 1,2,3,4, which corresponds to four state levels f1、f2、f3、f4。
The health status classification grade of the equipment obtained by calculation and the corresponding cloud model digital characteristics are shown in table 1.
Table 1 cloud model-based digital feature device status rating scheme
(1) The entropy weight method comprises the following calculation steps:
(1-1) constructing an equipment state evaluation index matrix:
an index matrix composed of m secondary indexes of n evaluation objects is shown in formula (2).
X=(xij)n×m=(X1,X2,...Xi,...,Xm) (2)
x={x1,x2,...,xi,...,xm} (3)
Wherein, X is an index matrix constructed by n multiplied by m index values; xiThe evaluation index is the ith index column vector in the index matrix, namely the vector formed by the ith evaluation indexes of the n evaluation objects; x is the number ofijThe jth index value of the ith evaluation object; x is index set, xjIs the jth index in the index set; m is the index number; n is the number of evaluation targets.
(1-2) device status evaluation index normalization processing:
considering that the equipment state evaluation index system comprises both positive indexes and negative indexes (the higher the positive index value is, the better the negative index value is), respectively performing normalization processing on the positive indexes and the negative indexes, as shown in formula (4) and formula (5):
and (4) after the normalization processing is carried out on the positive indexes and the negative indexes, obtaining a normalized index matrix shown as the formula (6).
And (1-3) calculating entropy values of the equipment state evaluation indexes. The formula for calculating the entropy of the jth evaluation index is shown in (7):
wherein k is 1/ln (n), pijIs the ratio of the score of the ith sample device on the jth index to the scores of all the objects to be evaluated on the index, as shown in formula (8).
And (1-4) calculating the entropy weight of each equipment state evaluation index. The entropy weight calculation formula of the j-th evaluation index is shown in (9).
(2) The cloud model comprises the following main steps: normalizing each index of the sample equipment according to the formulas (4) and (5), inputting the normalized numerical value and the corresponding weight into a reverse cloud generator, calculating the cloud numerical characteristics, and generating a cloud picture through the forward cloud generator
(2-1) calculating a reverse cloud generator
Expectation of j-th level index ExjEntropy EnjEntropy of HejThe calculation formula of (a) is as follows:
wherein P is the index sample number, xiThe index value is a second-level index value.
Can calculate each two according to entropy weight methodLevel index weight wjAnd combining the digital characteristics of the cloud model of the relevant indexes at each level to obtain the digital characteristic parameters of the cloud model of the level of the target layer. The calculation formula is as follows:
(2-2) calculating a forward cloud Generator
Characterised by the number C (E)x,En,He) The positive cloud generator randomly generates N cloud droplets (x)i,yi). The method comprises the following specific steps:
(2-2-1) with EnTo expectation, HeGenerating a normal distribution random number E 'as a standard deviation'n;
(2-2-2) with ExL E 'to expect |'nI is standard deviation to generate normal distribution random number xi;
(2-2-3) with E'n、xiAs a variable, into the formula yi=exp(-(xi-Ex)2/2(E'n)2) Producing cloud droplets (x)i,yi);
(2-2-4) repeating the steps (2-2-1) to (2-2-3) until N cloud drops are generated, and drawing a cloud model map according to the N cloud drops.
(2-2-5) if K cloud drops exist at the intersection point of the cloud picture of the sample equipment and the f-th level cloud picture, taking the mean value of the membership values of the K cloud drops as the membership of the equipment state value, as shown in the formula
And step 3: the power distribution equipment fault occurrence time prediction model training based on the long-term and short-term memory network,
normalizing the operation record index (the index of the 'operation record B' class in the equipment state evaluation index system, as shown in figures 2 and 3) of the sample equipment according to formulas (4) to (5) to form a matrix BzThen the membership degree vector of the sample deviceNumber of failed times lzRunning record index normalization matrix BzAs input samples, the LSTM neural network is input for training (the topology is shown in FIG. 4).
The training algorithm of the LSTM neural network is a back propagation algorithm and is divided into the following three steps, and when the sample data of each sample device is input into the LSTM neural network for training, the model training is finished; or when the training error is set to be less than 1e-6, the model training is finished, the training error is the total training error of the sample set which is trained each time, the training error is not the training error of single sample data, and 1e-6 is a value obtained by experience and repeatability tests.
(3-1) calculating the output value of the LSTM memory module in the forward direction;
ft=σ(Wfxxt+WfhΤt-1+bf) (18)
it=σ(Wixxt+WihTt-1+bi) (19)
ot=σ(Woxxt+WohTt-1+bo) (22)
Tt=ot·tanh(ct) (23)
wherein sigma is sigmoid function, xtIs the input matrix at time t, formed by membership vectorsNumber of failed times lzRunning record index normalization matrix BzComposition i.e. xt=[δz,lz,Bz];Wfx、Wix、Wox、WcxRepresentation and current input xtWeight matrix of multiplications, Wfh、Wih、Woh、WchIndicating the output value T at the time T-1t-1A multiplied weight matrix. bf、bi、bc、boRespectively as the offset terms of the forgetting gate, the input gate, the state unit and the output gate, ft、it、otActivation functions of forgetting gate, input gate, output gate, respectively, ct、Is a vector of state units and instantaneous states; t istThe current output of the LSTM is the true value of the next time the sample device fails.
(3-2) calculating the error term delta of each memory module reversely, wherein the error term of the LSTM propagates along two directions. The error term is propagated reversely along the time, namely the error term at the T-1 moment is calculated, and the output value of the LSTM at the T moment is TtDefining the error term at time t asWhere E is a loss function, thenFrom the formula of the full derivativeAboutδf,t、δi,t、δo,tThe relationship of the four items is that,δf,t、δi,t、δo,tdefined as follows, the symbol "·" denotes multiplication by elements:
δo,t=δt·tanh(ct)·ot·(1-ot) (24)
δf,t=δt·ot·(1-tanh(ct)2)·ct-1·ft·(1-ft) (25)
the substitution of each item is solved by using a partial derivative formula to solve the following problems:
equation (28) is the equation for the error to propagate back to the previous time instant, and thus the equation for the error term to propagate back to any k time instant can be derived:
defining the current layer as the l-th layer, the error term of the l-1 layer is the inverse of the weighted input of the error function to the l-1 layer neurons, i.e.Input of LSTMfl-1Representing the activation function of the l-1 layer. Equation (30) is obtained using the fully-derived equation, which is a calculation to pass the error to the previous layer.
(3-3) calculating the gradient of each weight matrix in (3-1) according to the corresponding residual error term, and updating the weight matrix, wherein the formula is as follows:
and 4, step 4: and predicting the next fault occurrence time of the equipment to be evaluated, and determining the overhaul time.
Generating a health state cloud picture of the equipment to be evaluated according to the cloud model in the step 2, and calculating a membership degree vector delta ' ═ delta ' of the equipment to be evaluated according to the formula (17) '1,δ′2,δ′3,δ′4](ii) a Normalizing the operation record index (shown in fig. 2 and 3) of the equipment to be evaluated according to formulas (4) to (5) to form a matrix B ', and then, converting the membership degree vector delta ' of the equipment to be evaluated into [ delta '1,δ′2,δ′3,δ′4]The failed times l ' and the running record index normalization matrix B ' form a matrix x 't=[δ',l',B']Substitution type (18) - (23) are calculated to obtain a predicted value T 'of the next failure occurrence time of the equipment to be evaluated'tThe maintenance time T is calculated according to the formula (43)z,
Tz=Tplan+Top+Ts<T′t (43)
Wherein, TplanFor planning maintenance time, TopThe time required for maintenance and the safety margin time T for ensuring the reliability of the power supplys。
According to the method, the historical operation data and the test data of the power distribution equipment are utilized to predict the next fault occurrence time of the equipment to be evaluated and determine the maintenance time, so that the reasonable planning of the maintenance plan of the in-operation equipment is realized, the maintenance blindness is avoided, the maintenance resources can be fully utilized, the maintenance efficiency is improved, and the equipment fault risk is reduced.
The above description of embodiments of the invention with reference to the drawings is not intended to be limiting, and those skilled in the art can make many variations without departing from the spirit of the invention and the scope of the appended claims, which fall within the scope of the invention.
Claims (3)
1. A method for determining the state maintenance time of power distribution equipment is characterized by comprising the following steps: the method comprises the following steps of,
step 1, the health degree is divided into equipment grades: according to the comprehensive deduction value of the equipment, the health degree g of the z-th equipment with the same class and voltage class is countedz(Z is 1,2, …, Z), g according to health degreezGrading the health state of the equipment;
according to the degree of health gzThe health state of the equipment is divided into four grades of normal, attention, abnormal and serious, and f is used for the health state of the equipment1、f2、f3、f4Represents;
is composed ofCloud number feature calculating 4 health status levels, afIs the minimum value of the health status grade interval, bfMaximum value of the health status grade interval, ExfIs an expected value of a device health level f, EnfEntropy, H, of the device health level fefAs the state of health of the apparatus, etcSuper entropy of stage f, HefTake 0.01, where f is 1,2,3,4, corresponding to four state levels f1、f2、f3、f4;
Step 2, determining the equipment state evaluation index weight by an entropy weight method, and solving the membership by a cloud model: after weighting the equipment state evaluation index by applying an entropy weight method, generating a health state cloud picture of the equipment to be evaluated through a cloud model, and then calculating the membership between the health state cloud of the equipment to be evaluated and each health state grade cloud to obtain a membership vector;
the entropy weight method is carried out as follows,
firstly, constructing an equipment state evaluation index matrix: an index matrix composed of m secondary indexes, which are n evaluation objects, is as follows, X ═ Xij)n×m=(X1,X2,...Xi,...,Xm),x={x1,x2,...,xi,...,xm},
Wherein, X is an index matrix constructed by n multiplied by m index values; xiThe evaluation index is the ith index column vector in the index matrix, namely the vector formed by the ith evaluation indexes of the n evaluation objects; x is the number ofijThe j index value is the ith evaluation object; x is index set, xjIs the jth index in the index set; m is the index number; n is the number of evaluation objects;
secondly, the equipment state evaluation index normalization processing: the positive indicators and the negative indicators are normalized respectively, as follows,
a normalized device status evaluation index matrix is obtained, as follows,
thirdly, calculating the entropy value of each equipment state evaluation index: the calculation formula is as follows,
ejentropy for the jth evaluation index, k being 1/ln (n), pijIs the ratio of the score of the ith sample device on the jth index relative to the scores of all the objects to be evaluated on the index,
fourthly, calculating the entropy weight of each equipment state evaluation index: the calculation formula is as follows,
wjentropy weights for the jth evaluation index;
the cloud model generates the cloud picture mainly by the following steps,
first, calculate the inverse cloud generator: expectation of j-th level index ExjEntropy EnjEntropy of HejThe formula of (c) is as follows,
wherein S is2Is the variance, P is the number of index samples, xiThe index value is a secondary index value; the digital characteristic parameters of the cloud model of the level of the target layer are obtained by combining the digital characteristics of the cloud model of the related indexes of each level, the calculation formula is shown as follows,
second, calculate a forward cloud generator: characterised by the number C (E)x,En,He) The positive cloud generator randomly generates N cloud droplets (x)i,yi) The method comprises the following specific steps:
a, with EnTo expectation, HeGenerating a normal distribution random number E 'as a standard deviation'n;
b, with ExL E 'to expect |'nI is standard deviation to generate normal distribution random number xi;
c, and E'n、xiAs variables, into the formula yi=exp(-(xi-Ex)2/2(E'n)2) Producing cloud droplets (x)i,yi);
d, repeating the steps a to c until N cloud drops are generated, and drawing a cloud model graph according to the N cloud drops;
k cloud drops are arranged at the intersection point of the cloud picture of the sample device and the f-th level cloud picture, the mean value of the membership degree values of the K cloud drops is taken as the membership degree of the device state value, as shown in the formula,wherein, f is 1,2,3,4, the membership vector of the z-th sample device
Step 3, training a power distribution equipment fault occurrence time prediction model based on the long-term and short-term memory network: firstly, normalizing the equipment operation record indexes in the sample equipment state evaluation indexes to form a matrix BzThen the membership vector delta of the sample devicezFailure times lzMatrix BzInputting the LSTM neural network as an input sample for training;
the training algorithm of the LSTM neural network is a back propagation algorithm and comprises three steps, namely, firstly, the output value of an LSTM memory module is calculated in a forward direction by combining a weight matrix, secondly, the error item of each memory module is calculated in a reverse direction, and thirdly, the gradient of each weight matrix used in the first step is calculated according to the corresponding residual error item, and the weight matrix is updated;
step 4, predicting the next fault occurrence time of the equipment to be evaluated, and determining the maintenance time: according to a membership degree vector delta ' ═ delta ' of the equipment to be evaluated '1,δ'2,δ'3,δ'4]Inputting a matrix consisting of the failed times l 'and the equipment operation record index normalization matrix B' into the trained LSTM neural network to calculate to obtain a predicted value T 'of the next failure occurrence time of the equipment to be evaluated'tAccording to the formula Tz=Tplan+Top+Ts<T'tDetermining a time to overhaul TzWherein, TplanFor planning maintenance time, TopTime required for inspection, TsA safety margin time.
2. The method of determining electrical distribution equipment condition repair time of claim 1, wherein: after the sample data of each sample device is input into the LSTM neural network for training, the model training is finished; or finishing the model training when the training error is set to be less than 1 e-6.
3. The method of determining electrical distribution equipment condition repair time of claim 1, wherein: calculating the output value of the LSTM memory module in the forward direction, namely the true value T of the next fault occurrence time of the sample equipmenttIs as follows, ft=σ(Wfxxt+WfhΤt-1+bf),it=σ(Wixxt+WihTt-1+bi), ot=σ(Woxxt+WohTt-1+bo),Tt=ot·tanh(ct) (ii) a σ is sigmoid function, xt=[δz,lz,Bz]Is the input matrix at time t, Wfx、Wix、Wox、WcxRepresentation and current input xtA multiplied weight matrix, Wfh、Wih、Woh、WchIndicating the output value T at the time T-1t-1Multiplied weight matrix, bf、bi、bc、boRespectively as the offset terms of the forgetting gate, the input gate, the state unit and the output gate, ft、it、otActivation functions of forgetting gate, input gate, output gate, respectively, ct、Is a vector of state units and instantaneous states; t is a unit oftThe current output of the LSTM neural network is the real value of the next fault occurrence time of the sample device.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911049328.1A CN110880062B (en) | 2019-10-31 | 2019-10-31 | Method for determining state maintenance time of power distribution equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911049328.1A CN110880062B (en) | 2019-10-31 | 2019-10-31 | Method for determining state maintenance time of power distribution equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110880062A CN110880062A (en) | 2020-03-13 |
CN110880062B true CN110880062B (en) | 2022-07-08 |
Family
ID=69728229
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911049328.1A Active CN110880062B (en) | 2019-10-31 | 2019-10-31 | Method for determining state maintenance time of power distribution equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110880062B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111652395A (en) * | 2020-06-12 | 2020-09-11 | 成都国铁电气设备有限公司 | Health assessment method for high-speed railway contact network equipment |
CN113221441B (en) * | 2020-12-24 | 2022-12-02 | 山东鲁能软件技术有限公司 | Method and device for health assessment of power plant equipment |
CN112884174A (en) * | 2021-02-05 | 2021-06-01 | 上海市市政工程管理咨询有限公司 | Daily maintenance information management method and system for road |
CN113408759A (en) * | 2021-07-07 | 2021-09-17 | 大连理工大学 | Method for preventive maintenance of distributed energy supply system equipment based on entropy theory |
CN114970904B (en) * | 2022-07-26 | 2022-11-29 | 中铁电气化勘测设计研究院有限公司 | Digital adjustment method for contact network operation and maintenance resources based on defect processing |
Citations (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103177186A (en) * | 2013-03-21 | 2013-06-26 | 国家电网公司 | Electric return circuit fault probability forecasting method |
CN103279671A (en) * | 2013-06-03 | 2013-09-04 | 南京大学 | Urban water disaster risk prediction method based on RBF (radial basis function) neural network-cloud model |
CN105590146A (en) * | 2016-02-29 | 2016-05-18 | 上海带来科技有限公司 | Power plant device intelligent prediction overhaul method and power plant device intelligent prediction overhaul system based on big data |
CN106056218A (en) * | 2016-05-11 | 2016-10-26 | 国电南瑞科技股份有限公司 | Equipment monthly maintenance scheduling optimization method considering overload and transient stability constraint |
CN106251042A (en) * | 2016-07-19 | 2016-12-21 | 广西电网有限责任公司电力科学研究院 | A kind of method that distribution transformer is carried out health state evaluation |
CN206312210U (en) * | 2016-08-11 | 2017-07-07 | 中国南方电网有限责任公司电网技术研究中心 | State evaluation system of power distribution network equipment |
CN106961349A (en) * | 2017-02-20 | 2017-07-18 | 江苏大学 | A kind of sensor fault identifying system and method based on data fusion |
CN106997513A (en) * | 2017-04-11 | 2017-08-01 | 广西电网有限责任公司电力科学研究院 | The distribution net equipment state evaluation system analyzed based on big data |
CN107769972A (en) * | 2017-10-25 | 2018-03-06 | 武汉大学 | A kind of power telecom network equipment fault Forecasting Methodology based on improved LSTM |
CN108037378A (en) * | 2017-10-26 | 2018-05-15 | 上海交通大学 | Running state of transformer Forecasting Methodology and system based on long memory network in short-term |
CN109490726A (en) * | 2018-11-23 | 2019-03-19 | 广西电网有限责任公司南宁供电局 | Electric power transformer insulated state evaluating method based on Clouds theory |
CN109583520A (en) * | 2018-12-27 | 2019-04-05 | 云南电网有限责任公司玉溪供电局 | A kind of state evaluating method of cloud model and genetic algorithm optimization support vector machines |
CN110070060A (en) * | 2019-04-26 | 2019-07-30 | 天津开发区精诺瀚海数据科技有限公司 | A kind of method for diagnosing faults of bearing apparatus |
CN110210169A (en) * | 2019-06-14 | 2019-09-06 | 中铁高新工业股份有限公司 | A kind of shield machine failure prediction method based on LSTM |
CN110264053A (en) * | 2019-06-10 | 2019-09-20 | 广西电网有限责任公司电力科学研究院 | A kind of distribution network reliability evaluation method for considering Strategies of Maintenance and failure rate being influenced |
CN110348513A (en) * | 2019-07-10 | 2019-10-18 | 北京华电天仁电力控制技术有限公司 | A kind of Wind turbines failure prediction method based on deep learning |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11599795B2 (en) * | 2017-11-08 | 2023-03-07 | International Business Machines Corporation | Reducing the cost of n modular redundancy for neural networks |
-
2019
- 2019-10-31 CN CN201911049328.1A patent/CN110880062B/en active Active
Patent Citations (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103177186A (en) * | 2013-03-21 | 2013-06-26 | 国家电网公司 | Electric return circuit fault probability forecasting method |
CN103279671A (en) * | 2013-06-03 | 2013-09-04 | 南京大学 | Urban water disaster risk prediction method based on RBF (radial basis function) neural network-cloud model |
CN105590146A (en) * | 2016-02-29 | 2016-05-18 | 上海带来科技有限公司 | Power plant device intelligent prediction overhaul method and power plant device intelligent prediction overhaul system based on big data |
CN106056218A (en) * | 2016-05-11 | 2016-10-26 | 国电南瑞科技股份有限公司 | Equipment monthly maintenance scheduling optimization method considering overload and transient stability constraint |
CN106251042A (en) * | 2016-07-19 | 2016-12-21 | 广西电网有限责任公司电力科学研究院 | A kind of method that distribution transformer is carried out health state evaluation |
CN206312210U (en) * | 2016-08-11 | 2017-07-07 | 中国南方电网有限责任公司电网技术研究中心 | State evaluation system of power distribution network equipment |
CN106961349A (en) * | 2017-02-20 | 2017-07-18 | 江苏大学 | A kind of sensor fault identifying system and method based on data fusion |
CN106997513A (en) * | 2017-04-11 | 2017-08-01 | 广西电网有限责任公司电力科学研究院 | The distribution net equipment state evaluation system analyzed based on big data |
CN107769972A (en) * | 2017-10-25 | 2018-03-06 | 武汉大学 | A kind of power telecom network equipment fault Forecasting Methodology based on improved LSTM |
CN108037378A (en) * | 2017-10-26 | 2018-05-15 | 上海交通大学 | Running state of transformer Forecasting Methodology and system based on long memory network in short-term |
CN109490726A (en) * | 2018-11-23 | 2019-03-19 | 广西电网有限责任公司南宁供电局 | Electric power transformer insulated state evaluating method based on Clouds theory |
CN109583520A (en) * | 2018-12-27 | 2019-04-05 | 云南电网有限责任公司玉溪供电局 | A kind of state evaluating method of cloud model and genetic algorithm optimization support vector machines |
CN110070060A (en) * | 2019-04-26 | 2019-07-30 | 天津开发区精诺瀚海数据科技有限公司 | A kind of method for diagnosing faults of bearing apparatus |
CN110264053A (en) * | 2019-06-10 | 2019-09-20 | 广西电网有限责任公司电力科学研究院 | A kind of distribution network reliability evaluation method for considering Strategies of Maintenance and failure rate being influenced |
CN110210169A (en) * | 2019-06-14 | 2019-09-06 | 中铁高新工业股份有限公司 | A kind of shield machine failure prediction method based on LSTM |
CN110348513A (en) * | 2019-07-10 | 2019-10-18 | 北京华电天仁电力控制技术有限公司 | A kind of Wind turbines failure prediction method based on deep learning |
Non-Patent Citations (6)
Title |
---|
Modeling and Analysis of Microgrid Cluster Simulation Based on RTDS;OUYANG Jianna 等;《2018 2nd IEEE Conference on Energy Internet and Energy System Integration (EI2)》;20181220;第1-5页 * |
Study on the R TDS Simulation Platform of Parallel Injection Hybrid Active Power Filter;Wu Lifang;《2018 2nd IEEE Conference on Energy Internet and Energy System Integration (EI2)》;20181220;第1-7页 * |
基于云理论及熵权法的变压器潜在故障风险评估方法;熊卫红等;《电力自动化设备》;20180802(第08期);第125-130,146页 * |
基于云理论的变压器多重故障诊断及短期预测策略;朱见伟 等;《电气技术》;20160610(第3期);第3-4,7页 * |
基于融合理论的风电机组状态评价正态云模型;刘华新等;《太阳能学报》;20181028(第10期);第2891-2900页 * |
基于风险评估的配电设备状态检修决策方法;梁朔 等;《电力系统及其自动化学报》;20181030;第31卷(第1期);第47-52页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110880062A (en) | 2020-03-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110880062B (en) | Method for determining state maintenance time of power distribution equipment | |
Zheng et al. | A novel equivalent model of active distribution networks based on LSTM | |
Geeganage et al. | Application of energy-based power system features for dynamic security assessment | |
Sobbouhi et al. | Transient stability prediction of power system; a review on methods, classification and considerations | |
Genc et al. | Decision tree-based preventive and corrective control applications for dynamic security enhancement in power systems | |
CN102074955B (en) | Method based on knowledge discovery technology for stability assessment and control of electric system | |
CN110417011B (en) | Online dynamic security assessment method based on mutual information and iterative random forest | |
Wahab et al. | Fast transient stability assessment of large power system using probabilistic neural network with feature reduction techniques | |
Dai et al. | Real-time prediction of event-driven load shedding for frequency stability enhancement of power systems | |
CN102622510A (en) | System and method for quantitative management of software defects | |
CN107330573B (en) | State evaluation method and device for photovoltaic system key equipment | |
CN108053077A (en) | A kind of short-term wind speed forecasting method and system based on two type T-S fuzzy models of section | |
Shan et al. | Reliability analysis of power distribution network based on PSO-DBN | |
CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
Bhatt et al. | Static security assessment of power system using radial basis function neural network module | |
Abiyev | Fuzzy wavelet neural network for prediction of electricity consumption | |
Zhang | Short‐Term Power Load Forecasting Based on SAPSO‐CNN‐LSTM Model considering Autocorrelated Errors | |
Selvi et al. | Artificial neural network approach for on-line ATC estimation in deregulated power system | |
CN113919162B (en) | Voltage sag risk early warning method based on simulation and multi-source measured data fusion | |
CN115712064A (en) | Excitation system fault diagnosis method based on LSTM-CNN hybrid neural network | |
Xu et al. | A RBF network for short—Term Load forecast on microgrid | |
CN115423174A (en) | Short-term wind power prediction method considering fluctuation process division | |
Zhao et al. | A fast and accurate transient stability assessment method based on deep learning: Wecc case study | |
Citroen et al. | Moroccan long term electricity demand forecasting using Wavelet neural Networks | |
Bai et al. | Contingency-based voltage stability monitoring via neural network with multi-level feature fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |