CN110880062A - A Method for Determining Condition Maintenance Time of Power Distribution Equipment - Google Patents

A Method for Determining Condition Maintenance Time of Power Distribution Equipment Download PDF

Info

Publication number
CN110880062A
CN110880062A CN201911049328.1A CN201911049328A CN110880062A CN 110880062 A CN110880062 A CN 110880062A CN 201911049328 A CN201911049328 A CN 201911049328A CN 110880062 A CN110880062 A CN 110880062A
Authority
CN
China
Prior art keywords
equipment
cloud
value
state
matrix
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.)
Granted
Application number
CN201911049328.1A
Other languages
Chinese (zh)
Other versions
CN110880062B (en
Inventor
欧阳健娜
高立克
俞小勇
周扬珺
李珊
梁朔
陈绍南
黄伟翔
秦丽文
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Electric Power Research Institute of Guangxi Power Grid Co Ltd
Original Assignee
Electric Power Research Institute of Guangxi Power Grid Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Electric Power Research Institute of Guangxi Power Grid Co Ltd filed Critical Electric Power Research Institute of Guangxi Power Grid Co Ltd
Priority to CN201911049328.1A priority Critical patent/CN110880062B/en
Publication of CN110880062A publication Critical patent/CN110880062A/en
Application granted granted Critical
Publication of CN110880062B publication Critical patent/CN110880062B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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 then calculating membership degrees between the health state cloud of the equipment to be evaluated and each health state grade cloud; 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 the 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

一种配电设备状态检修时间的确定方法A Method for Determining Condition Maintenance Time of Power Distribution Equipment

技术领域：Technical field:

本发明属于电力工程技术领域，具体涉及一种配电设备状态检修时间的确定方法。The invention belongs to the technical field of electric power engineering, and particularly relates to a method for determining the state maintenance time of power distribution equipment.

背景技术：Background technique:

配电网传统设备检修多为定期检修与事后检修，检修资源利用效率低且无法针对各类设备差异进行故障预防。为此，利用配电设备的历史运行数据与试验数据对其进行状态检修，既避免了检修的盲目性，又能充分利用检修资源，提高检修效率。The traditional equipment maintenance of distribution network is mostly regular maintenance and post-event maintenance. The utilization efficiency of maintenance resources is low and it is impossible to prevent faults according to the differences of various equipment. For this reason, using the historical operation data and test data of power distribution equipment to perform condition maintenance can not only avoid the blindness of maintenance, but also make full use of maintenance resources and improve maintenance efficiency.

发明内容:Invention content:

本发明提供一种配电设备状态检修时间的确定方法，能够解决配电网传统设备检修多为定期检修与事后检修，检修资源利用效率低且无法针对各类设备差异进行故障预防的问题。The invention provides a method for determining the state maintenance time of power distribution equipment, which can solve the problem that traditional equipment maintenance of power distribution network is mostly regular maintenance and post-event maintenance, the utilization efficiency of maintenance resources is low, and fault prevention cannot be performed according to the differences of various equipment.

为达到上述目的，本发明采用如下技术方案：To achieve the above object, the present invention adopts the following technical solutions:

一种配电设备状态检修时间的确定方法，方法步骤如下，步骤1，健康度划分设备等级：根据设备综合扣分值，统计同类同电压等级第z台设备健康度

Figure 485668DEST_PATH_IMAGE001
（z=1,2,…Z），根据健康度
Figure 809333DEST_PATH_IMAGE001
将设备健康状态划分等级；步骤2，熵权法确定设备状态评估指标权重并用云模型求隶属度：应用熵权法对设备状态评估指标赋权后通过云模型生成待评价设备的健康状态云图，再计算待评价设备健康状态云与各健康状态等级云之间的隶属度并得到隶属度向量；步骤3，基于长短期记忆网络的配电设备故障发生时间预测模型训练：先将样本设备状态评估指标中的设备运行记录指标进行归一化，形成矩阵
Figure 854650DEST_PATH_IMAGE002
，然后把样本设备的隶属度向量δ z 、已故障次数
Figure 515438DEST_PATH_IMAGE003
、矩阵
Figure 487068DEST_PATH_IMAGE004
作为输入样本，输入LSTM神经网络进行训练；A method for determining the state maintenance time of power distribution equipment, the method steps are as follows, step 1, the health degree is divided into equipment grade: according to the comprehensive deduction value of the equipment, the health degree of the zth equipment of the same voltage level is counted
Figure 485668DEST_PATH_IMAGE001
( z =1,2,… Z ), according to health
Figure 809333DEST_PATH_IMAGE001
Divide the equipment health status into grades; step 2, the entropy weight method determines the weight of the equipment status evaluation index and uses the cloud model to obtain the degree of membership: the entropy weight method is applied to weight the equipment status evaluation index, and the cloud model is used to generate the health status cloud map of the equipment to be evaluated, Then calculate the membership degree between the health status cloud of the equipment to be evaluated and each health status level cloud, and obtain the membership degree vector; Step 3, train the fault occurrence time prediction model of power distribution equipment based on the long short-term memory network: first evaluate the status of the sample equipment The equipment operation record indicators in the indicators are normalized to form a matrix
Figure 854650DEST_PATH_IMAGE002
, then the membership vector δ z of the sample device, the number of failures
Figure 515438DEST_PATH_IMAGE003
,matrix
Figure 487068DEST_PATH_IMAGE004
As an input sample, input the LSTM neural network for training;

步骤4，预测待评价设备的下一次故障发生时间，并确定检修时间：根据待评价设备的隶属度向量

Figure 840689DEST_PATH_IMAGE005
、已故障次数
Figure 779826DEST_PATH_IMAGE006
、设备运行记录指标归一化矩阵
Figure 509884DEST_PATH_IMAGE007
组成的矩阵输入已训练的LSTM神经网络计算得到待评价设备下一次故障发生时间的预测值
Figure 834555DEST_PATH_IMAGE008
，根据公式
Figure 500023DEST_PATH_IMAGE009
确定检修时间
Figure 785511DEST_PATH_IMAGE010
，其中，
Figure 679780DEST_PATH_IMAGE011
为计划检修时间，
Figure 734323DEST_PATH_IMAGE012
为检修所需时间，
Figure 570692DEST_PATH_IMAGE013
安全裕度时间。Step 4: Predict the next failure time of the equipment to be evaluated, and determine the maintenance time: according to the membership vector of the equipment to be evaluated
Figure 840689DEST_PATH_IMAGE005
, the number of failures
Figure 779826DEST_PATH_IMAGE006
, Equipment operation record index normalization matrix
Figure 509884DEST_PATH_IMAGE007
The formed matrix is input to the trained LSTM neural network to calculate the predicted value of the next failure time of the equipment to be evaluated.
Figure 834555DEST_PATH_IMAGE008
, according to the formula
Figure 500023DEST_PATH_IMAGE009
Determine the maintenance time
Figure 785511DEST_PATH_IMAGE010
,in,
Figure 679780DEST_PATH_IMAGE011
In order to plan the maintenance time,
Figure 734323DEST_PATH_IMAGE012
Time required for maintenance,
Figure 570692DEST_PATH_IMAGE013
Safety Margin Time.

进一步地，步骤1中根据健康度

Figure 343476DEST_PATH_IMAGE014
将设备健康状态划分为正常、注意、异常和严重四个等级分别用
Figure 539971DEST_PATH_IMAGE015
、
Figure 714601DEST_PATH_IMAGE016
、
Figure 456292DEST_PATH_IMAGE017
、
Figure 716372DEST_PATH_IMAGE018
表示。Further, in step 1, according to the health
Figure 343476DEST_PATH_IMAGE014
The equipment health status is divided into four levels: normal, attention, abnormal and serious.
Figure 539971DEST_PATH_IMAGE015
,
Figure 714601DEST_PATH_IMAGE016
,
Figure 456292DEST_PATH_IMAGE017
,
Figure 716372DEST_PATH_IMAGE018
express.

进一步地，由式

Figure 952443DEST_PATH_IMAGE019
计算4个健康状态等级的云数字特征，
Figure 981579DEST_PATH_IMAGE020
为健康状态等级区间最小值，
Figure 159751DEST_PATH_IMAGE021
为健康状态等级区间最大值，
Figure 641548DEST_PATH_IMAGE022
为设备健康状态等级f的期望值，
Figure 179845DEST_PATH_IMAGE023
为设备健康状态等级f的熵，
Figure 204433DEST_PATH_IMAGE024
为设备健康状态等级f的超熵，
Figure 678140DEST_PATH_IMAGE024
取0.01，其中，f=1,2,3,4，分别对应四个状态等级
Figure 381654DEST_PATH_IMAGE015
、
Figure 225107DEST_PATH_IMAGE016
、
Figure 104201DEST_PATH_IMAGE017
、
Figure 748809DEST_PATH_IMAGE018
。Further, by the formula
Figure 952443DEST_PATH_IMAGE019
Calculate cloud digital features for 4 health status levels,
Figure 981579DEST_PATH_IMAGE020
is the minimum value of the health state level interval,
Figure 159751DEST_PATH_IMAGE021
is the maximum value of the health state level interval,
Figure 641548DEST_PATH_IMAGE022
is the expected value of the equipment health status level f ,
Figure 179845DEST_PATH_IMAGE023
is the entropy of the device health status level f ,
Figure 204433DEST_PATH_IMAGE024
is the superentropy of the device health state level f ,
Figure 678140DEST_PATH_IMAGE024
Take 0.01, where f = 1, 2, 3, 4, corresponding to four state levels respectively
Figure 381654DEST_PATH_IMAGE015
,
Figure 225107DEST_PATH_IMAGE016
,
Figure 104201DEST_PATH_IMAGE017
,
Figure 748809DEST_PATH_IMAGE018
.

进一步地，步骤2中的熵权法步骤如下，Further, the steps of the entropy weight method in step 2 are as follows,

第一，构建设备状态评估指标矩阵：由n个评价对象m个二级指标构成的指标矩阵，如下，

Figure 329832DEST_PATH_IMAGE025
，
Figure 85299DEST_PATH_IMAGE026
，First, construct the equipment status evaluation index matrix: an index matrix composed of n evaluation objects and m secondary indexes, as follows,
Figure 329832DEST_PATH_IMAGE025
,
Figure 85299DEST_PATH_IMAGE026
,

其中，X为由

Figure 553320DEST_PATH_IMAGE027
个指标值构造的指标矩阵；X i 为指标矩阵中的第i个指标列向量，即n个评价对象的第i个评价指标组成的向量；为第i个评价对象的第j个指标值；x为指标集合，
Figure 103250DEST_PATH_IMAGE028
为指标集合中的第j个指标；m为指标个数；n为评价对象个数；Among them, X is the
Figure 553320DEST_PATH_IMAGE027
The index matrix constructed by the index values; X i is the ith index column vector in the index matrix, that is, the vector composed of the ith evaluation index of the n evaluation objects; it is the jth index value of the ith evaluation object; x is the set of indicators,
Figure 103250DEST_PATH_IMAGE028
is the jth indicator in the indicator set; m is the number of indicators; n is the number of evaluation objects;

第二，设备状态评估指标归一化处理：对正向指标和负向指标分别进行归一化处理，如下

Figure 46935DEST_PATH_IMAGE029
，Second, normalization processing of equipment status evaluation indicators: normalize the positive indicators and negative indicators respectively, as follows
Figure 46935DEST_PATH_IMAGE029
,

Figure 232191DEST_PATH_IMAGE030
，
Figure 232191DEST_PATH_IMAGE030
,

得到归一化设备状态评估指标矩阵，如下，The normalized equipment status evaluation index matrix is obtained, as follows,

Figure 944932DEST_PATH_IMAGE031
，
Figure 944932DEST_PATH_IMAGE031
,

第三，各设备状态评估指标的熵值计算：计算公式如下，Third, the entropy value calculation of each equipment status evaluation index: the calculation formula is as follows,

Figure 275551DEST_PATH_IMAGE032
，e j 为第j个评估指标的熵值，其中，
Figure 706532DEST_PATH_IMAGE033
，
Figure 928435DEST_PATH_IMAGE034
是第i个样本设备在第j个指标上得分相对于所有待评价对象在该指标上得分的占比，
Figure 761262DEST_PATH_IMAGE035
，
Figure 275551DEST_PATH_IMAGE032
, e j is the entropy value of the jth evaluation index, where,
Figure 706532DEST_PATH_IMAGE033
,
Figure 928435DEST_PATH_IMAGE034
is the ratio of the score of the i -th sample device on the j -th indicator relative to the score of all the objects to be evaluated on this indicator,
Figure 761262DEST_PATH_IMAGE035
,

第四，各设备状态评估指标的熵权计算：计算公式如下，Fourth, the entropy weight calculation of each equipment state evaluation index: the calculation formula is as follows,

Figure 528360DEST_PATH_IMAGE036
，w j 为第j个评估指标的熵权。
Figure 528360DEST_PATH_IMAGE036
, w j is the entropy weight of the jth evaluation index.

进一步地，步骤2中云模型生成云图主要步骤如下，Further, in step 2, the main steps of generating a cloud map from the cloud model are as follows:

第一，计算逆向云发生器：第j二级指标的期望

Figure 181059DEST_PATH_IMAGE037
、熵
Figure 973696DEST_PATH_IMAGE038
、超熵
Figure 129871DEST_PATH_IMAGE039
的计算公式如下，First, calculate the reverse cloud generator: the expectation of the second-level indicator
Figure 181059DEST_PATH_IMAGE037
,entropy
Figure 973696DEST_PATH_IMAGE038
, super entropy
Figure 129871DEST_PATH_IMAGE039
The calculation formula is as follows,

Figure 458084DEST_PATH_IMAGE040
，
Figure 473445DEST_PATH_IMAGE041
，
Figure 178096DEST_PATH_IMAGE042
，
Figure 844569DEST_PATH_IMAGE043
，
Figure 458084DEST_PATH_IMAGE040
,
Figure 473445DEST_PATH_IMAGE041
,
Figure 178096DEST_PATH_IMAGE042
,
Figure 844569DEST_PATH_IMAGE043
,

其中，S 2 是方差，P为指标样本数量，

Figure 343684DEST_PATH_IMAGE044
为二级指标值；结合各级相关指标云模型数字特征求得目标层等级云模型数字特征参数，计算式如下所示，
Figure 846340DEST_PATH_IMAGE045
，
Figure 89103DEST_PATH_IMAGE046
，
Figure 99829DEST_PATH_IMAGE047
；Among them, S2 is the variance , P is the number of index samples,
Figure 343684DEST_PATH_IMAGE044
is the second-level index value; the digital characteristic parameters of the cloud model of the target layer level are obtained by combining the digital characteristics of the cloud model of the relevant indicators at all levels. The calculation formula is as follows:
Figure 846340DEST_PATH_IMAGE045
,
Figure 89103DEST_PATH_IMAGE046
,
Figure 99829DEST_PATH_IMAGE047
;

第二，计算正向云发生器：由数字特征为

Figure 769845DEST_PATH_IMAGE048
的正向云发生器随机产生N个的云滴
Figure 759798DEST_PATH_IMAGE049
，具体步骤为：Second, calculate the forward cloud generator: by the digital features as
Figure 769845DEST_PATH_IMAGE048
The forward cloud generator randomly generates N cloud droplets
Figure 759798DEST_PATH_IMAGE049
, the specific steps are:

a，以

Figure 540672DEST_PATH_IMAGE050
为期望，
Figure 181738DEST_PATH_IMAGE051
为标准差，生成正态分布随机数
Figure 22655DEST_PATH_IMAGE052
；a, with
Figure 540672DEST_PATH_IMAGE050
for expectation,
Figure 181738DEST_PATH_IMAGE051
is the standard deviation, generating a normally distributed random number
Figure 22655DEST_PATH_IMAGE052
;

b，以

Figure 499904DEST_PATH_IMAGE053
为期望，
Figure 84469DEST_PATH_IMAGE054
为标准差，生成正态分布随机数
Figure 81506DEST_PATH_IMAGE055
；b, with
Figure 499904DEST_PATH_IMAGE053
for expectation,
Figure 84469DEST_PATH_IMAGE054
is the standard deviation, generating a normally distributed random number
Figure 81506DEST_PATH_IMAGE055
;

c，以

Figure 93324DEST_PATH_IMAGE052
、
Figure 792290DEST_PATH_IMAGE055
为变量，代入公式
Figure 570759DEST_PATH_IMAGE056
产生云滴
Figure 796204DEST_PATH_IMAGE049
；c, with
Figure 93324DEST_PATH_IMAGE052
,
Figure 792290DEST_PATH_IMAGE055
For the variable, substitute the formula
Figure 570759DEST_PATH_IMAGE056
cloud droplets
Figure 796204DEST_PATH_IMAGE049
;

d，重复步骤a至c，直至产生N个云滴为止，并根据N个云滴绘制云模型图。d. Repeat steps a to c until N cloud droplets are generated, and draw a cloud model diagram according to the N cloud droplets.

进一步地，样本设备云图与第f朵等级云图的交点有K个云滴，取K个云滴的隶属度值的均值作为该设备状态值的隶属度，如式所示，

Figure 854290DEST_PATH_IMAGE057
，其中，f=1,2,3,4，则第z台样本设备的隶属度向量
Figure 430764DEST_PATH_IMAGE058
。Further, there are K cloud droplets at the intersection of the cloud map of the sample device and the cloud map of the fth level, and the mean value of the membership values of the K cloud droplets is taken as the membership degree of the state value of the device, as shown in the formula:
Figure 854290DEST_PATH_IMAGE057
, where f = 1,2,3,4, then the membership vector of the zth sample device
Figure 430764DEST_PATH_IMAGE058
.

进一步地，步骤3中LSTM神经网络的训练算法为反向传播算法，分为三步，第一，结合权重矩阵进行前向计算LSTM记忆模块的输出值，第二，反向计算每个记忆模块的误差项，第三，根据相应的残差项，计算第一步中所用的每一个权重矩阵的梯度，对权重矩阵进行更新。Further, the training algorithm of the LSTM neural network in step 3 is a back-propagation algorithm, which is divided into three steps. First, the output value of the LSTM memory module is forwardly calculated in combination with the weight matrix, and second, each memory module is calculated in reverse. The error term of , and thirdly, according to the corresponding residual term, the gradient of each weight matrix used in the first step is calculated, and the weight matrix is updated.

进一步地，当每个样本设备的样本数据均输入LSTM神经网络进行训练后，模型训练完毕；或者设置训练误差小于1e-06时模型训练结束。Further, when the sample data of each sample device is input into the LSTM neural network for training, the model training is completed; or the model training ends when the training error is set to be less than 1e-06.

进一步地，前向计算LSTM记忆模块的输出值即样本设备下一次故障发生时间的真实值

Figure 983231DEST_PATH_IMAGE059
的计算公式如下，
Figure 63182DEST_PATH_IMAGE060
，
Figure 557749DEST_PATH_IMAGE061
，Further, the output value of the forward calculation LSTM memory module is the real value of the next fault occurrence time of the sample device
Figure 983231DEST_PATH_IMAGE059
The calculation formula is as follows,
Figure 63182DEST_PATH_IMAGE060
,
Figure 557749DEST_PATH_IMAGE061
,

Figure 621520DEST_PATH_IMAGE062
，
Figure 476212DEST_PATH_IMAGE063
，
Figure 410670DEST_PATH_IMAGE064
，
Figure 621520DEST_PATH_IMAGE062
,
Figure 476212DEST_PATH_IMAGE063
,
Figure 410670DEST_PATH_IMAGE064
,

Figure 76138DEST_PATH_IMAGE065
；
Figure 361625DEST_PATH_IMAGE066
为sigmoid函数，
Figure 255894DEST_PATH_IMAGE067
是t时刻的输入矩阵，
Figure 310438DEST_PATH_IMAGE068
、
Figure 146807DEST_PATH_IMAGE069
、
Figure 654012DEST_PATH_IMAGE070
、
Figure 850507DEST_PATH_IMAGE071
表示与当前输入
Figure 25136DEST_PATH_IMAGE072
相乘的权重矩阵，
Figure 766827DEST_PATH_IMAGE073
、
Figure 26907DEST_PATH_IMAGE074
、
Figure 262979DEST_PATH_IMAGE075
、
Figure 292115DEST_PATH_IMAGE076
表示与t-1时刻输出值
Figure 204707DEST_PATH_IMAGE077
相乘的权重矩阵，
Figure 952083DEST_PATH_IMAGE078
、
Figure 224802DEST_PATH_IMAGE079
、
Figure 108444DEST_PATH_IMAGE080
、
Figure 457517DEST_PATH_IMAGE081
分别为遗忘门、输入门、状态单元、输出门的偏置项，
Figure 426610DEST_PATH_IMAGE082
、
Figure 4484DEST_PATH_IMAGE083
、
Figure 8212DEST_PATH_IMAGE084
分别是遗忘门、输入门、输出门的激活函数，
Figure 262607DEST_PATH_IMAGE085
、
Figure 718996DEST_PATH_IMAGE086
是状态单元和即时状态的向量；
Figure 208883DEST_PATH_IMAGE087
为LSTM神经网络当前输出，是样本设备下一次故障发生时间的真实值。
Figure 76138DEST_PATH_IMAGE065
;
Figure 361625DEST_PATH_IMAGE066
is the sigmoid function,
Figure 255894DEST_PATH_IMAGE067
is the input matrix at time t ,
Figure 310438DEST_PATH_IMAGE068
,
Figure 146807DEST_PATH_IMAGE069
,
Figure 654012DEST_PATH_IMAGE070
,
Figure 850507DEST_PATH_IMAGE071
represents the current input
Figure 25136DEST_PATH_IMAGE072
multiplied weight matrix,
Figure 766827DEST_PATH_IMAGE073
,
Figure 26907DEST_PATH_IMAGE074
,
Figure 262979DEST_PATH_IMAGE075
,
Figure 292115DEST_PATH_IMAGE076
Represents the output value at time t- 1
Figure 204707DEST_PATH_IMAGE077
multiplied weight matrix,
Figure 952083DEST_PATH_IMAGE078
,
Figure 224802DEST_PATH_IMAGE079
,
Figure 108444DEST_PATH_IMAGE080
,
Figure 457517DEST_PATH_IMAGE081
are the bias terms of the forget gate, input gate, state unit, and output gate, respectively,
Figure 426610DEST_PATH_IMAGE082
,
Figure 4484DEST_PATH_IMAGE083
,
Figure 8212DEST_PATH_IMAGE084
are the activation functions of the forget gate, the input gate, and the output gate, respectively.
Figure 262607DEST_PATH_IMAGE085
,
Figure 718996DEST_PATH_IMAGE086
is the vector of state units and immediate states;
Figure 208883DEST_PATH_IMAGE087
is the current output of the LSTM neural network, which is the real value of the next failure time of the sample device.

与现有技术相比，本发明的有益效果是：Compared with the prior art, the beneficial effects of the present invention are:

本方法利用配电设备的历史运行数据与试验数据通过熵权法和云模型处理并输入LSTM神经网络训练之后根据待评价设备的相应指标数据处理后输入LSTM神经网络以预测待评价设备下一次故障发生时间并确定检修时间，实现在运设备检修计划的合理规划，既避免了检修的盲目性，又能充分利用检修资源，提高检修效率，降低设备故障风险。This method uses the historical operation data and test data of power distribution equipment to be processed by entropy weight method and cloud model and input to LSTM neural network for training, and then processed according to the corresponding index data of the equipment to be evaluated, and then input to the LSTM neural network to predict the next failure of the equipment to be evaluated. The occurrence time and the maintenance time are determined to realize the reasonable planning of the maintenance plan of the equipment in operation, which not only avoids the blindness of the maintenance, but also makes full use of the maintenance resources, improves the maintenance efficiency, and reduces the risk of equipment failure.

附图说明：Description of drawings:

图1为本发明配电设备状态检修时间的确定方法的流程图；Fig. 1 is the flow chart of the method for determining the state maintenance time of power distribution equipment according to the present invention;

图2为本发明配电变压器状态评估指标体系；Fig. 2 is the distribution transformer state evaluation index system of the present invention;

图3为本发明配电线路状态评估指标体系；Fig. 3 is the distribution line state evaluation index system of the present invention;

图4为本发明LSTM记忆模块结构；Fig. 4 is the LSTM memory module structure of the present invention;

图5为本发明LSTM网络时序展开图。FIG. 5 is an expanded view of the LSTM network timing sequence of the present invention.

具体实施方式：Detailed ways:

为了使本发明要解决的技术问题、技术方案和优点更加清楚，下面将结合附图及具体实施例进行详细描述，将本发明的具体实施方案详细叙述如下：In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments, and the specific embodiments of the present invention will be described in detail as follows:

本发明提供一种配电设备状态检修时间的确定方法，方法流程如图1所示，所讨论的配电设备状态检修实际上是根据设备综合扣分值进行健康状态等级划分；应用熵权法对设备状态评估指标赋权后通过云模型生成待评价设备的健康状态云图，再计算待评价设备健康状态云与各健康状态等级云之间的隶属度；最后将隶属度向量、已故障次数、运行记录等数据输入长短期记忆网络进行训练，实现对待评价设备下一故障发生时间的预测。本发明方法的主要内容包括以下步骤：The present invention provides a method for determining the state maintenance time of power distribution equipment. The method flow is shown in Figure 1. The state maintenance of power distribution equipment in question is actually to divide the health state level according to the comprehensive deduction value of the equipment; the entropy weight method is applied. After weighting the equipment status evaluation index, the cloud model is used to generate the health status cloud map of the equipment to be evaluated, and then the membership degree between the health status cloud of the equipment to be evaluated and each health status level cloud is calculated; finally, the membership degree vector, the number of failures, Data such as operation records are input into the long-term and short-term memory network for training, so as to predict the occurrence time of the next failure of the equipment to be evaluated. The main content of the method of the present invention comprises the following steps:

步骤1：根据设备综合扣分值，进行健康状态等级划分。Step 1: Classify the health status according to the comprehensive deduction value of the equipment.

根据设备综合扣分值，统计同类同电压等级第z台设备健康度

Figure 191752DEST_PATH_IMAGE001
（z=1,2,…Z），根据健康度
Figure 7261DEST_PATH_IMAGE001
将设备健康等级划分为正常、注意、异常和严重四个等级，分别用
Figure 560733DEST_PATH_IMAGE015
、
Figure 854311DEST_PATH_IMAGE016
、
Figure 193151DEST_PATH_IMAGE017
、
Figure 179561DEST_PATH_IMAGE018
表示，对应取值范围分别是[80，100]、[60，80)、[40，60)、[0，40)。According to the comprehensive deduction value of the equipment, the health degree of the zth equipment of the same voltage level is calculated
Figure 191752DEST_PATH_IMAGE001
( z =1,2,… Z ), according to health
Figure 7261DEST_PATH_IMAGE001
The equipment health level is divided into four levels: normal, attention, abnormal and serious.
Figure 560733DEST_PATH_IMAGE015
,
Figure 854311DEST_PATH_IMAGE016
,
Figure 193151DEST_PATH_IMAGE017
,
Figure 179561DEST_PATH_IMAGE018
indicates that the corresponding value ranges are [80, 100], [60, 80), [40, 60), [0, 40).

步骤2：应用熵权法对设备状态评估指标赋权后通过云模型生成待评价设备的健康状态云图，再计算待评价设备健康状态云与各健康状态等级云之间的隶属度。对样本设备按式（4）、（5）归一化各指标，将归一化后的数值和对应权重输入逆向云发生器，按式（10）~（13）计算云数字特征，再通过正向云发生器生成云图。Step 2: After the entropy weight method is applied to weight the equipment status evaluation index, the cloud model is used to generate the health status cloud map of the equipment to be evaluated, and then the membership degree between the health status cloud of the equipment to be evaluated and each health status level cloud is calculated. For the sample equipment, normalize each index according to equations (4) and (5), input the normalized values and corresponding weights into the reverse cloud generator, and calculate the cloud digital features according to equations (10) to (13), and then pass The forward cloud generator generates a cloud map.

根据设备状态值范围[0，1]和设备健康状态等级的对应关系，

Figure 485909DEST_PATH_IMAGE015
、
Figure 583178DEST_PATH_IMAGE016
、
Figure 275059DEST_PATH_IMAGE017
、
Figure 166792DEST_PATH_IMAGE018
四个状态值范围分别为[0.8，1]、[0.6，0.8)、[0.4，0.6)、[0，0.4)。由式（1）计算4个健康状态等级的云数字特征，按步骤（2-2）生成对应云图，
Figure 85069DEST_PATH_IMAGE020
为健康状态等级区间最小值，
Figure 595816DEST_PATH_IMAGE021
为健康状态等级区间最大值，
Figure 17570DEST_PATH_IMAGE022
为设备健康状态等级f的期望值，
Figure 971882DEST_PATH_IMAGE023
为设备健康状态等级f的熵，
Figure 377456DEST_PATH_IMAGE024
为设备健康状态等级f的超熵，
Figure 957473DEST_PATH_IMAGE024
取0.01，其中，f=1，2，3，4，分别对应四个状态等级
Figure 420684DEST_PATH_IMAGE015
、
Figure 654219DEST_PATH_IMAGE016
、
Figure 156876DEST_PATH_IMAGE017
、
Figure 868480DEST_PATH_IMAGE018
，计算得出设备健康状态划分等级及相应云模型数字特征如表1所示，According to the corresponding relationship between the device state value range [0, 1] and the device health state level,
Figure 485909DEST_PATH_IMAGE015
,
Figure 583178DEST_PATH_IMAGE016
,
Figure 275059DEST_PATH_IMAGE017
,
Figure 166792DEST_PATH_IMAGE018
The four state value ranges are [0.8, 1], [0.6, 0.8), [0.4, 0.6), [0, 0.4). Calculate the cloud digital features of the four health status levels by formula (1), and generate the corresponding cloud map according to step (2-2),
Figure 85069DEST_PATH_IMAGE020
is the minimum value of the health state level interval,
Figure 595816DEST_PATH_IMAGE021
is the maximum value of the health state level interval,
Figure 17570DEST_PATH_IMAGE022
is the expected value of the equipment health status level f ,
Figure 971882DEST_PATH_IMAGE023
is the entropy of the device health status level f ,
Figure 377456DEST_PATH_IMAGE024
is the superentropy of the device health state level f ,
Figure 957473DEST_PATH_IMAGE024
Take 0.01, where f = 1, 2, 3, 4, corresponding to the four state levels respectively
Figure 420684DEST_PATH_IMAGE015
,
Figure 654219DEST_PATH_IMAGE016
,
Figure 156876DEST_PATH_IMAGE017
,
Figure 868480DEST_PATH_IMAGE018
, the classification level of equipment health status and the corresponding cloud model digital characteristics are shown in Table 1.

Figure 168223DEST_PATH_IMAGE019
（1）
Figure 168223DEST_PATH_IMAGE019
(1)

Figure 572659DEST_PATH_IMAGE088
Figure 572659DEST_PATH_IMAGE088

（1）熵权法计算步骤如下：(1) The calculation steps of the entropy weight method are as follows:

（1-1）构建设备状态评估指标矩阵：(1-1) Construct equipment status evaluation index matrix:

由n个评价对象m个二级指标构成的指标矩阵如（2）所示下，The index matrix composed of n evaluation objects and m secondary indexes is shown in (2) below,

Figure 687246DEST_PATH_IMAGE025
（2）
Figure 687246DEST_PATH_IMAGE025
(2)

Figure 77907DEST_PATH_IMAGE026
（3）
Figure 77907DEST_PATH_IMAGE026
(3)

其中，X为由

Figure 594339DEST_PATH_IMAGE027
个指标值构造的指标矩阵；X i 为指标矩阵中的第i个指标列向量，即n个评价对象的第i个评价指标组成的向量；为第i个评价对象的第j个指标值；x为指标集合，
Figure 559890DEST_PATH_IMAGE028
为指标集合中的第j个指标；m为指标个数；n为评价对象个数；Among them, X is the
Figure 594339DEST_PATH_IMAGE027
The index matrix constructed by the index values; X i is the ith index column vector in the index matrix, that is, the vector composed of the ith evaluation index of the n evaluation objects; it is the jth index value of the ith evaluation object; x is the set of indicators,
Figure 559890DEST_PATH_IMAGE028
is the jth indicator in the indicator set; m is the number of indicators; n is the number of evaluation objects;

（1-2）设备状态评估指标归一化处理：(1-2) Normalization of equipment status evaluation indicators:

考虑到设备状态评估指标体系中既包含正向指标，又包含负向指标（正向指标数值越高越好，负向指标数值越低越好），对正向指标和负向指标分别进行归一化处理，如公式（4）和（5）所示，Considering that the equipment status evaluation index system includes both positive indicators and negative indicators (the higher the value of the positive index, the better, and the lower the value of the negative index, the better), the positive indicators and the negative indicators are respectively normalized. Normalization, as shown in equations (4) and (5),

Figure 161773DEST_PATH_IMAGE029
（4）
Figure 161773DEST_PATH_IMAGE029
(4)

Figure 621704DEST_PATH_IMAGE030
（5）
Figure 621704DEST_PATH_IMAGE030
(5)

得到归一化设备状态评估指标矩阵，如（6）所示下，The normalized equipment state evaluation index matrix is obtained, as shown in (6),

Figure 258222DEST_PATH_IMAGE031
（6）
Figure 258222DEST_PATH_IMAGE031
(6)

（1-3）各设备状态评估指标的熵值计算：计算如公式（7）所示，(1-3) Calculation of entropy value of each equipment state evaluation index: the calculation is shown in formula (7),

Figure 896139DEST_PATH_IMAGE032
（7）
Figure 896139DEST_PATH_IMAGE032
(7)

e j 为第j个评估指标的熵值，其中，

Figure 719738DEST_PATH_IMAGE033
，
Figure 983360DEST_PATH_IMAGE034
是第i个样本设备在第j个指标上得分相对于所有待评价对象在该指标上得分的占比，如公式（8）所示 e j is the entropy value of the jth evaluation index, where,
Figure 719738DEST_PATH_IMAGE033
,
Figure 983360DEST_PATH_IMAGE034
is the ratio of the score of the i -th sample device on the j -th indicator relative to the score of all the objects to be evaluated on this indicator, as shown in formula (8)

Figure 943226DEST_PATH_IMAGE035
（8）
Figure 943226DEST_PATH_IMAGE035
(8)

（1-4）各设备状态评估指标的熵权计算：计算公式如（9）所示，(1-4) Calculation of entropy weight of each equipment status evaluation index: the calculation formula is shown in (9),

Figure 250579DEST_PATH_IMAGE036
（9）
Figure 250579DEST_PATH_IMAGE036
(9)

w j 为第j个评估指标的熵权。 w j is the entropy weight of the jth evaluation index.

（2）云模型主要步骤如下：对样本设备按式、归一化各指标，将归一化后的数值和对应权重输入逆向云发生器，计算云数字特征，再通过正向云发生器生成云图。(2) The main steps of the cloud model are as follows: press the formula and normalize the indicators for the sample equipment, input the normalized values and corresponding weights into the reverse cloud generator, calculate the cloud digital features, and then generate them through the forward cloud generator Cloud map.

（2-1）计算逆向云发生器：第j二级指标的期望

Figure 827054DEST_PATH_IMAGE037
、熵
Figure 628788DEST_PATH_IMAGE038
、超熵
Figure 708740DEST_PATH_IMAGE039
的计算公式如下，(2-1) Calculate the reverse cloud generator: the expectation of the second level indicator
Figure 827054DEST_PATH_IMAGE037
,entropy
Figure 628788DEST_PATH_IMAGE038
, super entropy
Figure 708740DEST_PATH_IMAGE039
The calculation formula is as follows,

Figure 954039DEST_PATH_IMAGE040
(10)
Figure 954039DEST_PATH_IMAGE040
(10)

Figure 752230DEST_PATH_IMAGE041
(11)
Figure 752230DEST_PATH_IMAGE041
(11)

Figure 92076DEST_PATH_IMAGE042
(12)
Figure 92076DEST_PATH_IMAGE042
(12)

Figure 26534DEST_PATH_IMAGE043
(13)
Figure 26534DEST_PATH_IMAGE043
(13)

其中，S 2 是方差，P为指标样本数量，

Figure 816635DEST_PATH_IMAGE044
为二级指标值；结合各级相关指标云模型数字特征求得目标层等级云模型数字特征参数，计算式如下所示，Among them, S2 is the variance , P is the number of index samples,
Figure 816635DEST_PATH_IMAGE044
is the second-level index value; the digital characteristic parameters of the cloud model of the target layer level are obtained by combining the digital characteristics of the cloud model of the relevant indicators at all levels. The calculation formula is as follows:

Figure 961178DEST_PATH_IMAGE045
(14)
Figure 961178DEST_PATH_IMAGE045
(14)

Figure 494927DEST_PATH_IMAGE046
(15)
Figure 494927DEST_PATH_IMAGE046
(15)

Figure 424837DEST_PATH_IMAGE047
(16)
Figure 424837DEST_PATH_IMAGE047
(16)

（2-2）计算正向云发生器：由数字特征为

Figure 120261DEST_PATH_IMAGE048
的正向云发生器随机产生N个的云滴
Figure 519143DEST_PATH_IMAGE049
，具体步骤为：(2-2) Calculate the forward cloud generator: composed of digital features as
Figure 120261DEST_PATH_IMAGE048
The forward cloud generator randomly generates N cloud droplets
Figure 519143DEST_PATH_IMAGE049
, the specific steps are:

（2-2-1）以

Figure 325425DEST_PATH_IMAGE050
为期望，
Figure 109842DEST_PATH_IMAGE051
为标准差，生成正态分布随机数
Figure 241746DEST_PATH_IMAGE052
；(2-2-1) with
Figure 325425DEST_PATH_IMAGE050
for expectation,
Figure 109842DEST_PATH_IMAGE051
is the standard deviation, generating a normally distributed random number
Figure 241746DEST_PATH_IMAGE052
;

（2-2-2）以

Figure 360880DEST_PATH_IMAGE053
为期望，
Figure 236432DEST_PATH_IMAGE054
为标准差，生成正态分布随机数
Figure 140934DEST_PATH_IMAGE055
；(2-2-2) with
Figure 360880DEST_PATH_IMAGE053
for expectation,
Figure 236432DEST_PATH_IMAGE054
is the standard deviation, generating a normally distributed random number
Figure 140934DEST_PATH_IMAGE055
;

（2-2-3）以

Figure 912581DEST_PATH_IMAGE052
、
Figure 659957DEST_PATH_IMAGE055
为变量，代入公式
Figure 230878DEST_PATH_IMAGE056
产生云滴
Figure 724308DEST_PATH_IMAGE049
；(2-2-3) with
Figure 912581DEST_PATH_IMAGE052
,
Figure 659957DEST_PATH_IMAGE055
For the variable, substitute the formula
Figure 230878DEST_PATH_IMAGE056
cloud droplets
Figure 724308DEST_PATH_IMAGE049
;

（2-2-4）重复步骤(2-2-1)至(2-2-3)，直至产生N个云滴为止，并根据N个云滴绘制云模型图。(2-2-4) Repeat steps (2-2-1) to (2-2-3) until N cloud droplets are generated, and draw a cloud model diagram according to the N cloud droplets.

(2-2-5) 样本设备云图与第f朵等级云图的交点有K个云滴，取K个云滴的隶属度值的均值作为该设备状态值的隶属度，如式(17)所示，(2-2-5) There are K cloud droplets at the intersection of the cloud map of the sample device and the cloud map of the fth level, and the average value of the membership values of the K cloud droplets is taken as the membership degree of the state value of the device, as shown in formula (17) Show,

Figure 932435DEST_PATH_IMAGE057
(17)
Figure 932435DEST_PATH_IMAGE057
(17)

其中，f=1,2,3,4，则第z台样本设备的隶属度向量

Figure 635949DEST_PATH_IMAGE058
。Among them, f = 1,2,3,4, then the membership vector of the zth sample device
Figure 635949DEST_PATH_IMAGE058
.

步骤3：基于长短期记忆网络的配电设备故障发生时间预测模型训练，Step 3: Training of a prediction model for fault occurrence time of power distribution equipment based on long short-term memory network,

将样本设备的运行记录指标（设备状态评估指标体系中的“运行记录B”类指标，如图2、图3所示）按照公式~归一化，形成矩阵

Figure 977937DEST_PATH_IMAGE002
,然后把样本设备的隶属度向量
Figure 716086DEST_PATH_IMAGE089
、已故障次数
Figure 236061DEST_PATH_IMAGE003
、运行记录指标归一化矩阵
Figure 692450DEST_PATH_IMAGE004
作为输入样本，输入LSTM神经网络进行训练（其拓扑结构如图4所示）。The operation record indicators of the sample equipment (the "operation record B" category indicators in the equipment status evaluation index system, as shown in Figure 2 and Figure 3) are normalized according to the formula ~ to form a matrix
Figure 977937DEST_PATH_IMAGE002
, and then take the membership vector of the sample device
Figure 716086DEST_PATH_IMAGE089
, the number of failures
Figure 236061DEST_PATH_IMAGE003
, operation record index normalization matrix
Figure 692450DEST_PATH_IMAGE004
As input samples, input the LSTM neural network for training (its topology is shown in Figure 4).

LSTM神经网络的训练算法为反向传播算法，分为以下三步，且当每个样本设备的样本数据均输入LSTM神经网络进行训练后，模型训练完毕；或者设置训练误差小于1e-06时模型训练完毕，而训练误差是每次进行训练的样本集的总训练误差，不是单条样本数据的训练误差，而且1e-06是一个由经验和重复性试验所获得的值。The training algorithm of the LSTM neural network is the back-propagation algorithm, which 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 completed; or the model is set when the training error is less than 1e-06 The training is completed, and the training error is the total training error of the sample set for each training, not the training error of a single sample data, and 1e-06 is a value obtained by experience and repeated experiments.

（3-1）前向计算LSTM记忆模块的输出值；(3-1) Calculate the output value of the LSTM memory module forward;

Figure 808435DEST_PATH_IMAGE060
(18)
Figure 808435DEST_PATH_IMAGE060
(18)

Figure 666670DEST_PATH_IMAGE061
(19)
Figure 666670DEST_PATH_IMAGE061
(19)

Figure 216600DEST_PATH_IMAGE062
(20)
Figure 216600DEST_PATH_IMAGE062
(20)

Figure 770072DEST_PATH_IMAGE063
(21)
Figure 770072DEST_PATH_IMAGE063
(twenty one)

Figure 329230DEST_PATH_IMAGE064
(22)
Figure 329230DEST_PATH_IMAGE064
(twenty two)

Figure 901025DEST_PATH_IMAGE065
(23)
Figure 901025DEST_PATH_IMAGE065
(twenty three)

Figure 887436DEST_PATH_IMAGE066
其中为sigmoid函数，
Figure 193783DEST_PATH_IMAGE067
是t时刻的输入矩阵，
Figure 291052DEST_PATH_IMAGE068
、
Figure 592721DEST_PATH_IMAGE069
、
Figure 641710DEST_PATH_IMAGE070
、
Figure 28829DEST_PATH_IMAGE071
表示与当前输入
Figure 805156DEST_PATH_IMAGE072
相乘的权重矩阵，
Figure 492489DEST_PATH_IMAGE073
、
Figure 679757DEST_PATH_IMAGE074
、
Figure 757434DEST_PATH_IMAGE075
、
Figure 822604DEST_PATH_IMAGE076
表示与t-1时刻输出值
Figure 98865DEST_PATH_IMAGE077
相乘的权重矩阵，
Figure 597979DEST_PATH_IMAGE078
、
Figure 100636DEST_PATH_IMAGE079
、
Figure 343398DEST_PATH_IMAGE080
、
Figure 864378DEST_PATH_IMAGE081
分别为遗忘门、输入门、状态单元、输出门的偏置项，
Figure 534394DEST_PATH_IMAGE082
、
Figure 117822DEST_PATH_IMAGE083
、
Figure 39642DEST_PATH_IMAGE084
分别是遗忘门、输入门、输出门的激活函数，
Figure 290495DEST_PATH_IMAGE085
、
Figure 745792DEST_PATH_IMAGE086
是状态单元和即时状态的向量；
Figure 82095DEST_PATH_IMAGE087
为LSTM神经网络当前输出，是样本设备下一次故障发生时间的真实值。
Figure 887436DEST_PATH_IMAGE066
where is the sigmoid function,
Figure 193783DEST_PATH_IMAGE067
is the input matrix at time t ,
Figure 291052DEST_PATH_IMAGE068
,
Figure 592721DEST_PATH_IMAGE069
,
Figure 641710DEST_PATH_IMAGE070
,
Figure 28829DEST_PATH_IMAGE071
represents the current input
Figure 805156DEST_PATH_IMAGE072
multiplied weight matrix,
Figure 492489DEST_PATH_IMAGE073
,
Figure 679757DEST_PATH_IMAGE074
,
Figure 757434DEST_PATH_IMAGE075
,
Figure 822604DEST_PATH_IMAGE076
Represents the output value at time t- 1
Figure 98865DEST_PATH_IMAGE077
multiplied weight matrix,
Figure 597979DEST_PATH_IMAGE078
,
Figure 100636DEST_PATH_IMAGE079
,
Figure 343398DEST_PATH_IMAGE080
,
Figure 864378DEST_PATH_IMAGE081
are the bias terms of the forget gate, input gate, state unit, and output gate, respectively,
Figure 534394DEST_PATH_IMAGE082
,
Figure 117822DEST_PATH_IMAGE083
,
Figure 39642DEST_PATH_IMAGE084
are the activation functions of the forget gate, the input gate, and the output gate, respectively.
Figure 290495DEST_PATH_IMAGE085
,
Figure 745792DEST_PATH_IMAGE086
is the vector of state units and immediate states;
Figure 82095DEST_PATH_IMAGE087
is the current output of the LSTM neural network, which is the real value of the next failure time of the sample device.

（3-2）反向计算每个记忆模块的误差项

Figure 401081DEST_PATH_IMAGE090
，LSTM的误差项沿两个方向进行传播。误差项沿时间反向传播，即计算出t-1时刻的误差项，在t时刻，LSTM的输出值为
Figure 647386DEST_PATH_IMAGE091
，定义t时刻的误差项为
Figure 659204DEST_PATH_IMAGE092
，其中E为损失函数，则
Figure 545120DEST_PATH_IMAGE093
。根据全导数公式可求得
Figure 667797DEST_PATH_IMAGE094
关于
Figure 784920DEST_PATH_IMAGE095
、
Figure 967640DEST_PATH_IMAGE096
、
Figure 419481DEST_PATH_IMAGE097
、
Figure 345848DEST_PATH_IMAGE098
四项的关系，
Figure 816013DEST_PATH_IMAGE095
、
Figure 169634DEST_PATH_IMAGE096
、
Figure 797186DEST_PATH_IMAGE097
、
Figure 261666DEST_PATH_IMAGE098
定义如下，符号“
Figure 337069DEST_PATH_IMAGE099
”表示按元素相乘：(3-2) Reversely calculate the error term of each memory module
Figure 401081DEST_PATH_IMAGE090
, the error term of the LSTM propagates in both directions. The error term propagates back in time, that is, the error term at time t- 1 is calculated. At time t , the output value of LSTM is
Figure 647386DEST_PATH_IMAGE091
, the error term at time t is defined as
Figure 659204DEST_PATH_IMAGE092
, where E is the loss function, then
Figure 545120DEST_PATH_IMAGE093
. According to the full derivative formula, it can be obtained
Figure 667797DEST_PATH_IMAGE094
about
Figure 784920DEST_PATH_IMAGE095
,
Figure 967640DEST_PATH_IMAGE096
,
Figure 419481DEST_PATH_IMAGE097
,
Figure 345848DEST_PATH_IMAGE098
four relationship,
Figure 816013DEST_PATH_IMAGE095
,
Figure 169634DEST_PATH_IMAGE096
,
Figure 797186DEST_PATH_IMAGE097
,
Figure 261666DEST_PATH_IMAGE098
Defined as follows, the notation "
Figure 337069DEST_PATH_IMAGE099
” means element-wise multiplication:

Figure 596012DEST_PATH_IMAGE100
(24)
Figure 596012DEST_PATH_IMAGE100
(twenty four)

Figure 133697DEST_PATH_IMAGE101
(25)
Figure 133697DEST_PATH_IMAGE101
(25)

Figure 605130DEST_PATH_IMAGE102
(26)
Figure 605130DEST_PATH_IMAGE102
(26)

Figure 784307DEST_PATH_IMAGE103
(27)
Figure 784307DEST_PATH_IMAGE103
(27)

利用偏导公式求出各项代入可求得：Using the partial derivative formula to find the substitution of each item can be obtained:

Figure 745310DEST_PATH_IMAGE104
(28)
Figure 745310DEST_PATH_IMAGE104
(28)

式为误差反向传播到上一时刻的公式，由此可得出误差项反向传播到任意k时刻的公式：The formula is the formula for the back propagation of the error to the previous moment, from which the formula for the back propagation of the error term to any k moment can be obtained:

Figure 393460DEST_PATH_IMAGE105
(29)
Figure 393460DEST_PATH_IMAGE105
(29)

定义当前为第

Figure 465322DEST_PATH_IMAGE106
层，则
Figure 470DEST_PATH_IMAGE107
层的误差项为误差函数对
Figure 132374DEST_PATH_IMAGE107
层神经元加权输入的倒数，即
Figure 2241DEST_PATH_IMAGE108
。LSTM的输入
Figure 268007DEST_PATH_IMAGE109
，
Figure 31563DEST_PATH_IMAGE110
表示
Figure 209735DEST_PATH_IMAGE107
层的激活函数。使用全导公式可得式（30），即为将误差传递到上一层的计算。define the current
Figure 465322DEST_PATH_IMAGE106
layer, then
Figure 470DEST_PATH_IMAGE107
The error term of the layer is the error function pair
Figure 132374DEST_PATH_IMAGE107
The reciprocal of the layer neuron weighted input, i.e.
Figure 2241DEST_PATH_IMAGE108
. Input to LSTM
Figure 268007DEST_PATH_IMAGE109
,
Figure 31563DEST_PATH_IMAGE110
express
Figure 209735DEST_PATH_IMAGE107
layer activation function. Equation (30) can be obtained by using the full derivative formula, which is the calculation of transferring the error to the previous layer.

Figure 691532DEST_PATH_IMAGE111
(30)
Figure 691532DEST_PATH_IMAGE111
(30)

（3-3）根据相应的残差项，计算（3-1）中每个权重矩阵的梯度，对权重矩阵进行更新，公式如下：(3-3) Calculate the gradient of each weight matrix in (3-1) according to the corresponding residual item, and update the weight matrix. The formula is as follows:

Figure 731294DEST_PATH_IMAGE112
(31)
Figure 731294DEST_PATH_IMAGE112
(31)

Figure 614937DEST_PATH_IMAGE113
(32)
Figure 614937DEST_PATH_IMAGE113
(32)

Figure 964009DEST_PATH_IMAGE114
(33)
Figure 964009DEST_PATH_IMAGE114
(33)

Figure 933102DEST_PATH_IMAGE115
(34)
Figure 933102DEST_PATH_IMAGE115
(34)

Figure 275091DEST_PATH_IMAGE116
(35)
Figure 275091DEST_PATH_IMAGE116
(35)

Figure 385478DEST_PATH_IMAGE117
(36)
Figure 385478DEST_PATH_IMAGE117
(36)

Figure 171031DEST_PATH_IMAGE118
(37)
Figure 171031DEST_PATH_IMAGE118
(37)

Figure 361841DEST_PATH_IMAGE119
(38)
Figure 361841DEST_PATH_IMAGE119
(38)

Figure 117308DEST_PATH_IMAGE120
(39)
Figure 117308DEST_PATH_IMAGE120
(39)

Figure 834597DEST_PATH_IMAGE121
(40)
Figure 834597DEST_PATH_IMAGE121
(40)

Figure 384527DEST_PATH_IMAGE122
(41)
Figure 384527DEST_PATH_IMAGE122
(41)

Figure 203578DEST_PATH_IMAGE123
(42)
Figure 203578DEST_PATH_IMAGE123
(42)

步骤4：预测待评价设备的下一次故障发生时间，确定检修时间。Step 4: Predict the next failure time of the equipment to be evaluated, and determine the maintenance time.

根据步骤2中云模型生成待评价设备的健康状态云图，由式（17）可计算出待评价设备的隶属度向量

Figure 762736DEST_PATH_IMAGE005
；将待评价设备的运行记录指标（如图2、图3所示）按照公式（4）~（5）进行归一化，形成矩阵
Figure 101575DEST_PATH_IMAGE007
，然后把由待评价设备的隶属度向量
Figure 822407DEST_PATH_IMAGE005
、已故障次数
Figure 987809DEST_PATH_IMAGE006
、运行记录指标归一化矩阵
Figure 960444DEST_PATH_IMAGE007
组成的矩阵代入式（18）~（23），计算得到待评价设备下一次故障发生时间的预测值
Figure 793271DEST_PATH_IMAGE008
，按公式（43）计算得到检修时间
Figure 809637DEST_PATH_IMAGE010
，According to the cloud model in step 2, the health state cloud map of the device to be evaluated is generated, and the membership vector of the device to be evaluated can be calculated by formula (17).
Figure 762736DEST_PATH_IMAGE005
; Normalize the operation record indicators of the equipment to be evaluated (as shown in Figure 2 and Figure 3) according to formulas (4) to (5) to form a matrix
Figure 101575DEST_PATH_IMAGE007
, and then put the membership vector of the device to be evaluated
Figure 822407DEST_PATH_IMAGE005
, the number of failures
Figure 987809DEST_PATH_IMAGE006
, operation record index normalization matrix
Figure 960444DEST_PATH_IMAGE007
Substitute the formed matrix into equations (18)~(23), and calculate the predicted value of the next fault occurrence time of the equipment to be evaluated.
Figure 793271DEST_PATH_IMAGE008
, the maintenance time is calculated according to formula (43)
Figure 809637DEST_PATH_IMAGE010
,

Figure 462335DEST_PATH_IMAGE009
（43）
Figure 462335DEST_PATH_IMAGE009
(43)

其中，

Figure 238662DEST_PATH_IMAGE011
为计划检修时间，
Figure 660416DEST_PATH_IMAGE011
为检修所需时间，为了确保供电可靠性，还需要考虑一个安全裕度时间
Figure 988629DEST_PATH_IMAGE013
。in,
Figure 238662DEST_PATH_IMAGE011
In order to plan the maintenance time,
Figure 660416DEST_PATH_IMAGE011
For the time required for maintenance, in order to ensure the reliability of the power supply, a safety margin time needs to be considered
Figure 988629DEST_PATH_IMAGE013
.

根据本方法利用配电设备的历史运行数据与试验数据预测待评价设备下一次故障发生时间并确定检修时间，实现在运设备检修计划的合理规划，既避免了检修的盲目性，又能充分利用检修资源，提高检修效率，降低设备故障风险。According to this method, the historical operation data and test data of power distribution equipment are used to predict the next fault occurrence time of the equipment to be evaluated and determine the maintenance time, so as to realize the reasonable planning of the maintenance plan of the equipment in operation, which not only avoids the blindness of maintenance, but also makes full use of Maintenance resources, improve maintenance efficiency, and reduce the risk of equipment failure.

上面结合附图对本发明的实施例进行的描述不是限制性的，本领域的普通技术人员在本发明的启示下，在不脱离本发明宗旨和权利要求所保护的范围情况下，还可做出很多形式，这些均属于本发明的保护之内。The above description of the embodiments of the present invention in conjunction with the accompanying drawings is not restrictive, and those of ordinary skill in the art can also make In many forms, these fall within the protection of the present invention.

Claims (9)

1.一种配电设备状态检修时间的确定方法，其特征在于：方法步骤如下，1. a method for determining the state maintenance time of power distribution equipment, is characterized in that: the method steps are as follows, 步骤1，健康度划分设备等级：根据设备综合扣分值，统计同类同电压等级第z台设备健康度
Figure 351615DEST_PATH_IMAGE001
（z=1,2,…Z），根据健康度
Figure 416785DEST_PATH_IMAGE001
将设备健康状态划分等级；Step 1. Classify equipment by health degree: According to the comprehensive deduction value of the equipment, count the health degree of the zth equipment of the same voltage level.
Figure 351615DEST_PATH_IMAGE001
( z =1,2,… Z ), according to health
Figure 416785DEST_PATH_IMAGE001
Classify device health status; 步骤2，熵权法确定设备状态评估指标权重并用云模型求隶属度：应用熵权法对设备状态评估指标赋权后通过云模型生成待评价设备的健康状态云图，再计算待评价设备健康状态云与各健康状态等级云之间的隶属度并得到隶属度向量；Step 2, the entropy weight method determines the weight of the equipment status evaluation index and uses the cloud model to obtain the degree of membership: After the entropy weight method is used to weight the equipment status evaluation index, the cloud model is used to generate the health state cloud map of the equipment to be evaluated, and then the health state of the equipment to be evaluated is calculated. The membership degree between the cloud and each health status level cloud and get the membership degree vector; 步骤3，基于长短期记忆网络的配电设备故障发生时间预测模型训练：先将样本设备状态评估指标中的设备运行记录指标进行归一化，形成矩阵
Figure 161887DEST_PATH_IMAGE002
，然后把样本设备的隶属度向量δ z 、已故障次数
Figure 67526DEST_PATH_IMAGE003
、矩阵
Figure 694817DEST_PATH_IMAGE004
作为输入样本，输入LSTM神经网络进行训练；Step 3: Training of the prediction model for fault occurrence time of power distribution equipment based on long short-term memory network: first normalize the equipment operation record index in the sample equipment state evaluation index to form a matrix
Figure 161887DEST_PATH_IMAGE002
, then the membership vector δ z of the sample device, the number of failures
Figure 67526DEST_PATH_IMAGE003
,matrix
Figure 694817DEST_PATH_IMAGE004
As an input sample, input the LSTM neural network for training; 步骤4，预测待评价设备的下一次故障发生时间，并确定检修时间：根据待评价设备的隶属度向量
Figure 62213DEST_PATH_IMAGE005
、已故障次数
Figure 130663DEST_PATH_IMAGE006
、设备运行记录指标归一化矩阵
Figure 800679DEST_PATH_IMAGE007
组成的矩阵输入已训练的LSTM神经网络计算得到待评价设备下一次故障发生时间的预测值
Figure 541364DEST_PATH_IMAGE008
，根据公式
Figure 791080DEST_PATH_IMAGE009
确定检修时间
Figure 41933DEST_PATH_IMAGE010
，其中，
Figure 23795DEST_PATH_IMAGE011
为计划检修时间，
Figure 625678DEST_PATH_IMAGE012
为检修所需时间，
Figure 334877DEST_PATH_IMAGE013
安全裕度时间。Step 4: Predict the next failure time of the equipment to be evaluated, and determine the maintenance time: according to the membership vector of the equipment to be evaluated
Figure 62213DEST_PATH_IMAGE005
, the number of failures
Figure 130663DEST_PATH_IMAGE006
, Equipment operation record index normalization matrix
Figure 800679DEST_PATH_IMAGE007
The formed matrix is input to the trained LSTM neural network to calculate the predicted value of the next failure time of the equipment to be evaluated.
Figure 541364DEST_PATH_IMAGE008
, according to the formula
Figure 791080DEST_PATH_IMAGE009
Determine the maintenance time
Figure 41933DEST_PATH_IMAGE010
,in,
Figure 23795DEST_PATH_IMAGE011
In order to plan the maintenance time,
Figure 625678DEST_PATH_IMAGE012
Time required for maintenance,
Figure 334877DEST_PATH_IMAGE013
Safety Margin Time. 2.根据权利要求1所述的配电设备状态检修时间的确定方法，其特征在于：步骤1中根据健康度
Figure 705815DEST_PATH_IMAGE014
将设备健康状态划分为正常、注意、异常和严重四个等级分别用
Figure 593000DEST_PATH_IMAGE015
、
Figure 682179DEST_PATH_IMAGE016
、
Figure 430954DEST_PATH_IMAGE017
、
Figure 656399DEST_PATH_IMAGE018
表示。2. The method for determining the state maintenance time of power distribution equipment according to claim 1, characterized in that: in step 1, according to the degree of health
Figure 705815DEST_PATH_IMAGE014
The equipment health status is divided into four levels: normal, attention, abnormal and serious.
Figure 593000DEST_PATH_IMAGE015
,
Figure 682179DEST_PATH_IMAGE016
,
Figure 430954DEST_PATH_IMAGE017
,
Figure 656399DEST_PATH_IMAGE018
express. 3.根据权利要求2所述的配电设备状态检修时间的确定方法，其特征在于：由式
Figure 980064DEST_PATH_IMAGE019
计算4个健康状态等级的云数字特征，
Figure 290960DEST_PATH_IMAGE020
为健康状态等级区间最小值，
Figure 341961DEST_PATH_IMAGE021
为健康状态等级区间最大值，
Figure 687492DEST_PATH_IMAGE022
为设备健康状态等级f的期望值，
Figure 244375DEST_PATH_IMAGE023
为设备健康状态等级f的熵，
Figure 917933DEST_PATH_IMAGE024
为设备健康状态等级f的超熵，
Figure 913571DEST_PATH_IMAGE024
取0.01，其中，f=1,2,3,4，分别对应四个状态等级
Figure 751425DEST_PATH_IMAGE015
、
Figure 479210DEST_PATH_IMAGE016
、
Figure 764698DEST_PATH_IMAGE017
、
Figure 908234DEST_PATH_IMAGE018
。3. The method for determining the state maintenance time of power distribution equipment according to claim 2, characterized in that: by the formula
Figure 980064DEST_PATH_IMAGE019
Calculate cloud digital features for 4 health status levels,
Figure 290960DEST_PATH_IMAGE020
is the minimum value of the health state level interval,
Figure 341961DEST_PATH_IMAGE021
is the maximum value of the health state level interval,
Figure 687492DEST_PATH_IMAGE022
is the expected value of the equipment health status level f ,
Figure 244375DEST_PATH_IMAGE023
is the entropy of the device health status level f ,
Figure 917933DEST_PATH_IMAGE024
is the superentropy of the device health state level f ,
Figure 913571DEST_PATH_IMAGE024
Take 0.01, where f = 1, 2, 3, 4, corresponding to four state levels respectively
Figure 751425DEST_PATH_IMAGE015
,
Figure 479210DEST_PATH_IMAGE016
,
Figure 764698DEST_PATH_IMAGE017
,
Figure 908234DEST_PATH_IMAGE018
. 4.根据权利要求1所述的配电设备状态检修时间的确定方法，其特征在于：步骤2中的熵权法步骤如下，4. the method for determining the state maintenance time of power distribution equipment according to claim 1, is characterized in that: the entropy weight method step in step 2 is as follows, 第一，构建设备状态评估指标矩阵：由n个评价对象m个二级指标构成的指标矩阵，如下，
Figure 228357DEST_PATH_IMAGE025
，
Figure 313994DEST_PATH_IMAGE026
，First, construct the equipment status evaluation index matrix: an index matrix composed of n evaluation objects and m secondary indexes, as follows,
Figure 228357DEST_PATH_IMAGE025
,
Figure 313994DEST_PATH_IMAGE026
, 其中，X为由
Figure 290040DEST_PATH_IMAGE027
个指标值构造的指标矩阵；X i 为指标矩阵中的第i个指标列向量，即n个评价对象的第i个评价指标组成的向量；为第i个评价对象的第j个指标值；x为指标集合，
Figure 96322DEST_PATH_IMAGE028
为指标集合中的第j个指标；m为指标个数；n为评价对象个数；Among them, X is the
Figure 290040DEST_PATH_IMAGE027
The index matrix constructed by the index values; X i is the ith index column vector in the index matrix, that is, the vector composed of the ith evaluation index of the n evaluation objects; it is the jth index value of the ith evaluation object; x is the set of indicators,
Figure 96322DEST_PATH_IMAGE028
is the jth indicator in the indicator set; m is the number of indicators; n is the number of evaluation objects; 第二，设备状态评估指标归一化处理：对正向指标和负向指标分别进行归一化处理，如下
Figure 146318DEST_PATH_IMAGE029
，Second, normalization processing of equipment status evaluation indicators: normalize the positive indicators and negative indicators respectively, as follows
Figure 146318DEST_PATH_IMAGE029
,
Figure 278222DEST_PATH_IMAGE030
，
Figure 278222DEST_PATH_IMAGE030
, 得到归一化设备状态评估指标矩阵，如下，The normalized equipment status evaluation index matrix is obtained, as follows,
Figure 164400DEST_PATH_IMAGE031
，
Figure 164400DEST_PATH_IMAGE031
, 第三，各设备状态评估指标的熵值计算：计算公式如下，Third, the entropy value calculation of each equipment status evaluation index: the calculation formula is as follows,
Figure 977636DEST_PATH_IMAGE032
，e j 为第j个评估指标的熵值，其中，
Figure 6772DEST_PATH_IMAGE033
，
Figure 450522DEST_PATH_IMAGE034
是第i个样本设备在第j个指标上得分相对于所有待评价对象在该指标上得分的占比，
Figure 932319DEST_PATH_IMAGE035
，
Figure 977636DEST_PATH_IMAGE032
, e j is the entropy value of the jth evaluation index, where,
Figure 6772DEST_PATH_IMAGE033
,
Figure 450522DEST_PATH_IMAGE034
is the ratio of the score of the i -th sample device on the j -th indicator relative to the score of all the objects to be evaluated on this indicator,
Figure 932319DEST_PATH_IMAGE035
, 第四，各设备状态评估指标的熵权计算：计算公式如下，Fourth, the entropy weight calculation of each equipment state evaluation index: the calculation formula is as follows,
Figure 549245DEST_PATH_IMAGE036
，w j 为第j个评估指标的熵权。
Figure 549245DEST_PATH_IMAGE036
, w j is the entropy weight of the jth evaluation index. 5.根据权利要求4所述的配电设备状态检修时间的确定方法，其特征在于：步骤2中云模型生成云图主要步骤如下，5. The method for determining the state maintenance time of power distribution equipment according to claim 4, characterized in that: in step 2, the main steps of cloud model generation cloud map are as follows, 第一，计算逆向云发生器：第j二级指标的期望
Figure 557522DEST_PATH_IMAGE037
、熵
Figure 296807DEST_PATH_IMAGE038
、超熵
Figure 672425DEST_PATH_IMAGE039
的计算公式如下，First, calculate the reverse cloud generator: the expectation of the second-level indicator
Figure 557522DEST_PATH_IMAGE037
,entropy
Figure 296807DEST_PATH_IMAGE038
, super entropy
Figure 672425DEST_PATH_IMAGE039
The calculation formula is as follows,
Figure 250299DEST_PATH_IMAGE040
，
Figure 457290DEST_PATH_IMAGE041
，
Figure 101898DEST_PATH_IMAGE042
，
Figure 699232DEST_PATH_IMAGE043
，
Figure 250299DEST_PATH_IMAGE040
,
Figure 457290DEST_PATH_IMAGE041
,
Figure 101898DEST_PATH_IMAGE042
,
Figure 699232DEST_PATH_IMAGE043
, 其中，S 2 是方差，P为指标样本数量，
Figure 454699DEST_PATH_IMAGE044
为二级指标值；结合各级相关指标云模型数字特征求得目标层等级云模型数字特征参数，计算式如下所示，
Figure 250616DEST_PATH_IMAGE045
，
Figure 190759DEST_PATH_IMAGE046
，
Figure 400024DEST_PATH_IMAGE047
；Among them, S2 is the variance , P is the number of index samples,
Figure 454699DEST_PATH_IMAGE044
is the second-level index value; the digital characteristic parameters of the cloud model of the target layer level are obtained by combining the digital characteristics of the cloud model of the relevant indicators at all levels. The calculation formula is as follows:
Figure 250616DEST_PATH_IMAGE045
,
Figure 190759DEST_PATH_IMAGE046
,
Figure 400024DEST_PATH_IMAGE047
; 第二，计算正向云发生器：由数字特征为
Figure 834547DEST_PATH_IMAGE048
的正向云发生器随机产生N个的云滴
Figure 547288DEST_PATH_IMAGE049
，具体步骤为：Second, calculate the forward cloud generator: by the digital features as
Figure 834547DEST_PATH_IMAGE048
The forward cloud generator randomly generates N cloud droplets
Figure 547288DEST_PATH_IMAGE049
, the specific steps are: a，以
Figure 471382DEST_PATH_IMAGE050
为期望，
Figure 794041DEST_PATH_IMAGE051
为标准差，生成正态分布随机数
Figure 891310DEST_PATH_IMAGE052
；a, with
Figure 471382DEST_PATH_IMAGE050
for expectation,
Figure 794041DEST_PATH_IMAGE051
is the standard deviation, generating a normally distributed random number
Figure 891310DEST_PATH_IMAGE052
; b，以
Figure 599503DEST_PATH_IMAGE053
为期望，
Figure 694498DEST_PATH_IMAGE054
为标准差，生成正态分布随机数
Figure 612776DEST_PATH_IMAGE055
；b, with
Figure 599503DEST_PATH_IMAGE053
for expectation,
Figure 694498DEST_PATH_IMAGE054
is the standard deviation, generating a normally distributed random number
Figure 612776DEST_PATH_IMAGE055
; c，以
Figure 903949DEST_PATH_IMAGE052
、
Figure 325703DEST_PATH_IMAGE055
为变量，代入公式
Figure 857178DEST_PATH_IMAGE056
产生云滴
Figure 872539DEST_PATH_IMAGE049
；c, with
Figure 903949DEST_PATH_IMAGE052
,
Figure 325703DEST_PATH_IMAGE055
For the variable, substitute the formula
Figure 857178DEST_PATH_IMAGE056
cloud droplets
Figure 872539DEST_PATH_IMAGE049
; d，重复步骤a至c，直至产生N个云滴为止，并根据N个云滴绘制云模型图。d. Repeat steps a to c until N cloud droplets are generated, and draw a cloud model diagram according to the N cloud droplets. 6.根据权利要求5所述的配电设备状态检修时间的确定方法，其特征在于：样本设备云图与第f朵等级云图的交点有K个云滴，取K个云滴的隶属度值的均值作为该设备状态值的隶属度，如式所示，
Figure 842769DEST_PATH_IMAGE057
，其中，f=1,2,3,4，则第z台样本设备的隶属度向量
Figure 10707DEST_PATH_IMAGE058
。6. The determination method of power distribution equipment condition maintenance time according to claim 5, it is characterized in that: the intersection of sample equipment cloud image and the f -th grade cloud image has K cloud droplets, take the membership degree value of K cloud droplets. The mean value is used as the membership degree of the state value of the equipment, as shown in the formula,
Figure 842769DEST_PATH_IMAGE057
, where f = 1,2,3,4, then the membership vector of the zth sample device
Figure 10707DEST_PATH_IMAGE058
. 7.根据权利要求1所述的配电设备状态检修时间的确定方法，其特征在于：步骤3中LSTM神经网络的训练算法为反向传播算法，分为三步，第一，结合权重矩阵进行前向计算LSTM记忆模块的输出值，第二，反向计算每个记忆模块的误差项，第三，根据相应的残差项，计算第一步中所用的每一个权重矩阵的梯度，对权重矩阵进行更新。7. the determination method of power distribution equipment condition maintenance time according to claim 1, it is characterized in that: in step 3, the training algorithm of LSTM neural network is back propagation algorithm, is divided into three steps, first, carry out in conjunction with weight matrix Calculate the output value of the LSTM memory module forward. Second, calculate the error term of each memory module in reverse. Third, calculate the gradient of each weight matrix used in the first step according to the corresponding residual term. The matrix is updated. 8.根据权利要求7所述的配电设备状态检修时间的确定方法，其特征在于：当每个样本设备的样本数据均输入LSTM神经网络进行训练后，模型训练完毕；或者设置训练误差小于1e-06时模型训练结束。8. The method for determining the state maintenance time of power distribution equipment according to claim 7, characterized in that: when the sample data of each sample equipment is input into the LSTM neural network for training, the model training is completed; or the training error is set to be less than 1e Model training ends at -06. 9.根据权利要求7所述的配电设备状态检修时间的确定方法，其特征在于：前向计算LSTM记忆模块的输出值即样本设备下一次故障发生时间的真实值
Figure 509822DEST_PATH_IMAGE059
的计算公式如下，
Figure 340374DEST_PATH_IMAGE060
，
Figure 458503DEST_PATH_IMAGE061
，9. The method for determining the state maintenance time of power distribution equipment according to claim 7, characterized in that: the output value of the forward calculation LSTM memory module is the real value of the next fault occurrence time of the sample equipment
Figure 509822DEST_PATH_IMAGE059
The calculation formula is as follows,
Figure 340374DEST_PATH_IMAGE060
,
Figure 458503DEST_PATH_IMAGE061
,
Figure 120429DEST_PATH_IMAGE062
，
Figure 915078DEST_PATH_IMAGE063
，
Figure 29665DEST_PATH_IMAGE064
，
Figure 120429DEST_PATH_IMAGE062
,
Figure 915078DEST_PATH_IMAGE063
,
Figure 29665DEST_PATH_IMAGE064
,
Figure 13801DEST_PATH_IMAGE065
；
Figure 405599DEST_PATH_IMAGE066
为sigmoid函数，
Figure 246516DEST_PATH_IMAGE067
是t时刻的输入矩阵，
Figure 474498DEST_PATH_IMAGE068
、
Figure 59063DEST_PATH_IMAGE069
、
Figure 305367DEST_PATH_IMAGE070
、
Figure 972978DEST_PATH_IMAGE071
表示与当前输入
Figure 62157DEST_PATH_IMAGE072
相乘的权重矩阵，
Figure 325779DEST_PATH_IMAGE073
、
Figure 754486DEST_PATH_IMAGE074
、
Figure 937206DEST_PATH_IMAGE075
、
Figure 139779DEST_PATH_IMAGE076
表示与t-1时刻输出值
Figure 331726DEST_PATH_IMAGE077
相乘的权重矩阵，
Figure 287044DEST_PATH_IMAGE078
、
Figure 906244DEST_PATH_IMAGE079
、
Figure 907698DEST_PATH_IMAGE080
、
Figure 762391DEST_PATH_IMAGE081
分别为遗忘门、输入门、状态单元、输出门的偏置项，
Figure 900111DEST_PATH_IMAGE082
、
Figure 893475DEST_PATH_IMAGE083
、
Figure 805061DEST_PATH_IMAGE084
分别是遗忘门、输入门、输出门的激活函数，
Figure 338811DEST_PATH_IMAGE085
、
Figure 268721DEST_PATH_IMAGE086
是状态单元和即时状态的向量；
Figure 964144DEST_PATH_IMAGE087
为LSTM神经网络当前输出，是样本设备下一次故障发生时间的真实值。
Figure 13801DEST_PATH_IMAGE065
;
Figure 405599DEST_PATH_IMAGE066
is the sigmoid function,
Figure 246516DEST_PATH_IMAGE067
is the input matrix at time t ,
Figure 474498DEST_PATH_IMAGE068
,
Figure 59063DEST_PATH_IMAGE069
,
Figure 305367DEST_PATH_IMAGE070
,
Figure 972978DEST_PATH_IMAGE071
represents the current input
Figure 62157DEST_PATH_IMAGE072
multiplied weight matrix,
Figure 325779DEST_PATH_IMAGE073
,
Figure 754486DEST_PATH_IMAGE074
,
Figure 937206DEST_PATH_IMAGE075
,
Figure 139779DEST_PATH_IMAGE076
Represents the output value at time t- 1
Figure 331726DEST_PATH_IMAGE077
multiplied weight matrix,
Figure 287044DEST_PATH_IMAGE078
,
Figure 906244DEST_PATH_IMAGE079
,
Figure 907698DEST_PATH_IMAGE080
,
Figure 762391DEST_PATH_IMAGE081
are the bias terms of the forget gate, input gate, state unit, and output gate, respectively,
Figure 900111DEST_PATH_IMAGE082
,
Figure 893475DEST_PATH_IMAGE083
,
Figure 805061DEST_PATH_IMAGE084
are the activation functions of the forget gate, the input gate, and the output gate, respectively.
Figure 338811DEST_PATH_IMAGE085
,
Figure 268721DEST_PATH_IMAGE086
is the vector of state units and immediate states;
Figure 964144DEST_PATH_IMAGE087
is the current output of the LSTM neural network, which is the real value of the next failure time of the sample device.
CN201911049328.1A 2019-10-31 2019-10-31 A Method for Determining Condition Maintenance Time of Power Distribution Equipment Active CN110880062B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911049328.1A CN110880062B (en) 2019-10-31 2019-10-31 A Method for Determining Condition 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 A Method for Determining Condition Maintenance Time of Power Distribution Equipment

Publications (2)

Publication Number Publication Date
CN110880062A true CN110880062A (en) 2020-03-13
CN110880062B 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 A Method for Determining Condition Maintenance Time of Power Distribution Equipment

Country Status (1)

Country Link
CN (1) CN110880062B (en)

Cited By (8)

* Cited by examiner, † Cited by third party
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
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
CN113221441A (en) * 2020-12-24 2021-08-06 山东鲁能软件技术有限公司 Method and device for health assessment of power plant equipment
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
CN114970904A (en) * 2022-07-26 2022-08-30 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
CN117571347A (en) * 2023-10-20 2024-02-20 河北白沙烟草有限责任公司 Equipment health state monitoring and evaluating method and device and electronic equipment
CN118536963A (en) * 2024-04-10 2024-08-23 国网冀北电力有限公司经济技术研究院 Transformer maintenance decision method and system based on state evaluation and fault rate correction
CN119443956A (en) * 2024-07-29 2025-02-14 广州昂立如新电气技术有限公司 A smart grid secondary equipment state intelligent assessment system and method

Citations (17)

* Cited by examiner, † Cited by third party
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
US20190138903A1 (en) * 2017-11-08 2019-05-09 International Business Machines Corporation Reducing the cost of n modular redundancy for neural networks
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

Patent Citations (17)

* Cited by examiner, † Cited by third party
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
US20190138903A1 (en) * 2017-11-08 2019-05-09 International Business Machines Corporation Reducing the cost of n modular redundancy for neural networks
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)

* Cited by examiner, † Cited by third party
Title
OUYANG JIANNA 等: "Modeling and Analysis of Microgrid Cluster Simulation Based on RTDS", 《2018 2ND IEEE CONFERENCE ON ENERGY INTERNET AND ENERGY SYSTEM INTEGRATION (EI2)》 *
WU LIFANG: "Study on the R TDS Simulation Platform of Parallel Injection Hybrid Active Power Filter", 《2018 2ND IEEE CONFERENCE ON ENERGY INTERNET AND ENERGY SYSTEM INTEGRATION (EI2)》 *
刘华新等: "基于融合理论的风电机组状态评价正态云模型", 《太阳能学报》 *
朱见伟 等: "基于云理论的变压器多重故障诊断及短期预测策略", 《电气技术》 *
梁朔 等: "基于风险评估的配电设备状态检修决策方法", 《电力系统及其自动化学报》 *
熊卫红等: "基于云理论及熵权法的变压器潜在故障风险评估方法", 《电力自动化设备》 *

Cited By (12)

* Cited by examiner, † Cited by third party
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
CN113221441A (en) * 2020-12-24 2021-08-06 山东鲁能软件技术有限公司 Method and device for health assessment of power plant 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
CN113408759B (en) * 2021-07-07 2025-02-18 大连理工大学 A preventive maintenance method for distributed energy supply system equipment based on entropy theory
CN114970904A (en) * 2022-07-26 2022-08-30 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
CN114970904B (en) * 2022-07-26 2022-11-29 中铁电气化勘测设计研究院有限公司 A method for digital adjustment of catenary operation and maintenance resources based on defect processing
CN117571347A (en) * 2023-10-20 2024-02-20 河北白沙烟草有限责任公司 Equipment health state monitoring and evaluating method and device and electronic equipment
CN117571347B (en) * 2023-10-20 2024-09-20 河北白沙烟草有限责任公司 Equipment health state monitoring and evaluating method and device and electronic equipment
CN118536963A (en) * 2024-04-10 2024-08-23 国网冀北电力有限公司经济技术研究院 Transformer maintenance decision method and system based on state evaluation and fault rate correction
CN119443956A (en) * 2024-07-29 2025-02-14 广州昂立如新电气技术有限公司 A smart grid secondary equipment state intelligent assessment system and method

Also Published As

Publication number Publication date
CN110880062B (en) 2022-07-08

Similar Documents

Publication Publication Date Title
CN110880062A (en) A Method for Determining Condition Maintenance Time of Power Distribution Equipment
CN104155574B (en) Distribution network failure sorting technique based on Adaptive Neuro-fuzzy Inference
El-Keib et al. Application of artificial neural networks in voltage stability assessment
CN112310980B (en) Safety and stability evaluation method and system for direct-current blocking frequency of alternating-current and direct-current series-parallel power grid
CN113591379B (en) A power system transient stability prevention and emergency coordination control auxiliary decision-making method
CN108038300A (en) Optical Fiber Condition Evaluation Method Based on Improved Membership Function and Neural Network
CN105117602A (en) Metering apparatus operation state early warning method
CN113705615B (en) A multi-stage equipment fault diagnosis method and system for electric vehicle charging process based on neural network
CN106780130A (en) A kind of evaluation method containing distributed photovoltaic power distribution network
CN109993665B (en) Online safety and stability assessment method, device and system for power system
CN107093895A (en) Online transient safe and stable appraisal procedure based on forecast failure collection automatic screening
CN105512808A (en) Power system transient stability assessment method based on big data
CN105354643A (en) Risk prediction evaluation method for wind power grid integration
CN110650040A (en) Information system security situation evaluation method based on correction matrix-entropy weight membership cloud
CN104361529A (en) Reliability detecting and evaluating method of power distribution system on basis of cloud model
CN107394772A (en) Consider that the power system blackstart of integration node weight recovers Multipurpose Optimal Method
CN109993346B (en) Microgrid voltage security assessment method based on chaotic time series and neural network
CN115712064A (en) Excitation system fault diagnosis method based on LSTM-CNN hybrid neural network
Zhou et al. Statistics-based method for large-scale group decision-making with incomplete linguistic distribution fuzzy information: Incorporating reliability and entropy
CN113204908A (en) Mechanical fault monitoring method for photoelectric composite submarine cable
CN109117651A (en) A kind of continuous data safety protecting method
CN115329911A (en) A safety correction method for power system with UPFC based on SAE binary model
CN105741184B (en) Transformer state evaluation method and device
CN114037209A (en) Method and device for comprehensive benefit analysis of distributed photovoltaic connected to DC power distribution system
CN109033561A (en) Mine ventilation system anti-disaster ability evaluation method and device

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