A kind of distribution net equipment health degree dynamic diagnosis method of meter and trust evaluation
Technical field
The present invention relates to distribution net equipment technical field, the distribution net equipment health degree of specifically a kind of meter and trust evaluation moves
State diagnostic method.
Background technology
Distribution Network Equipment is the final link to customer power supply, is played an important role in power network, thus distribution network
Safety and stability just seems especially important.With economic development and urbanization process, the scale of China's power distribution network also becomes increasingly
Greatly, users are also more and more stricter for the reliability requirement of power supply.Compared to high-voltage fence, Distribution Network Equipment voltage class
It is low, number of devices is more, grid structure is complicated, on-line monitoring equipment imperfection.So the state of Distribution Network Equipment can be according to setting
Standby history data, history test report and patrol record etc. establish evaluation model to be assessed.
State Grid Corporation of China has formulated distribution net equipment evaluation criterion at present, but the criterion of these models is generally all only
Can be compared with the threshold value of a certain fixed threshold or fixed proportion.Because the evaluated equipment overwhelming majority in distribution is state
Good, if applying mechanically fixed model criteria, it is possible that (many decades) interior all devices assessment result all exists for a long time
Within defined threshold, no equipment can overhaul;Then as equipment aging in batch, the assessment result of unexpected large number quipments is all in several years
It is interior to exceed defined threshold, cause in a short time maintenance workload increased dramatically, overhaul of the equipments work can not normally carry out.And distribution is set
Standby substantial amounts, the diagnosis criterion of its state should be compared with the Statistical Distribution of same category of device (lateral comparison), or
With (longitudinal comparison) compared with the historical data regularity of distribution of equipment itself, whether equipment is judged with respect to change level according to it
It is worth noting.
The content of the invention
The present invention provides a kind of distribution net equipment health degree dynamic diagnosis method of meter and trust evaluation, credible by data
Degree is assessed, the fusion of different criterion evaluation results, solves the data collision occurred during collecting device state variable and data are superfluous
Remaining phenomenon, while by dynamic regulation, the self-renewing of assessment models is realized, to adapt to continue to develop the power distribution network of change.
A kind of distribution net equipment health degree dynamic diagnosis method of meter and trust evaluation, it is characterised in that including following step
Suddenly:
Step (1):Obtain the primitive character collection of monitoring power distribution amount:
Data acquisition and monitor control system based on power distribution network, collect the monitoring variable that existing power distribution network can provide, shape
Into primitive character collection database, the primitive character set owner will include:Historical operational information, fault message, bad condition letter
Breath, routine test information;
Step (2):Determine that important state becomes by carrying out correlation analysis to the information of faulty equipment and non-faulting equipment
Amount:
Fault source tracing method is used to faulty equipment, the every terms of information of faulty equipment is determined in primitive character collection database,
By carrying out correlation analysis with the information of non-faulting equipment, information in database is screened, obtains same fail result phase
The larger characteristic information of closing property, and determine that it is important state variables;
Step (3):The data of important state variables to being determined in step (2) carry out input quality pretreatment;
Step (4):Important state variables after step (3) processing are given a mark;
Step (5):Marking result to important state variables is inputted into evaluation model, and draws evaluation result.
Further, in addition to step (6):Evaluation procedure under non-full information, it is specially:
When evaluating certain equipment, if certain variable data missing that should be in input step (5), it will influence equipment and comment
The degree of accuracy of valency and confidence level, therefore non-full information evaluation method is used, the variable of missing is replaced using big data statistical result,
Make evaluation result relatively more accurate.
Further, include the determination and amendment of step (7) dynamic parameter, be specially:
The important state variables feature of input is once updated at interval of cycle time T, and produces new statistics ginseng
Number:Variance, average etc., statistical parameter old in former scoring functions is replaced with new statistical parameter, realizes the self-renewing of algorithm,
Dynamically scoring functions in evaluation algorithms are modified, and record the relevant parameter after changing, to reach evaluation model with setting
The purpose of self-recision for update.
Further, preprocess method is handled singular value, to normal including 3 δ rules described in the step (3)
The processing of data step, data smoothing processing and data interpolation processing:
3 δ rules are to singular value processing:The normal state point of difference between the same continuous measurement point of data series two is obtained first
Cloth parameter, i.e. its sample average and sample standard deviation, and verify whether its distribution meets normal distribution law, then with this sample
Average and sample standard deviation go to filter legacy data, when 3 times bigger than sample standard deviation of the error between data and sample average with
When upper, it can be determined that can be replaced at this for singular value point, its value caused by interference by the average of 4 points before and after former data;
To the processing of normal data step:The change between the adjacent data of signal two is calculated first, is determined by 3 δ rules all
Transition, the time width between more adjacent two transition, if the time width between the two transition reaches preset value,
It is considered as the normal transition of system to be preserved;If time width is not up to preset value, it is considered as singular value and is handled by 3 δ rules;
Data smoothing processing:The weighted average of each 2 points totally five point datas is taken before and after step point, for step point edge
The data of position can use the method for reducing smooth region, retain legacy data as far as possible and accomplish corresponding smooth, formula is such as
Under:
Step both ends end points xi=xi;
Step both ends time end points
Step intermediate point
Data interpolating processing:All data are divided into two classes by interpolation method, and one kind is the association between different measurement points
Property stronger data, within time of measuring and spatial dimension, can mutually be converted between each thermometric point data, can be with being based on
The fitting of the time and space is filled a vacancy with interpolation;
Another kind of data are the poor data of the horizontal comparativity between different measurement points, and this kind of data, which is taken, linearly to be estimated
The method of calculation, that is, think that data are determined during loss by linear change, its two end data by given data.
Further, step (4) is for different equipment state variables, and scoring method is different, for that can not quantify
State variable use experience scoring;For can quantify but Different Individual due to manufacture etc. the different state variable of reason
Using functional arrangement scoring;For that can quantify and only be used statistical Butut scoring by the state variable of influence on system operation.
Further, the establishment step of evaluation model and internal computation flow are as follows described in step (5):
1. establish equipment state Comment gathers:It is good, and it is typically, suspicious, it is abnormal, dangerous;
2. the membership function of Comment gathers is established according to Fuzzy Criteria:
Kilter membership function:
Normal condition membership function:
Suspicious state membership function:
Abnormality membership function:
Precarious position membership function:
3. bring membership function i.e. five kinds of states of the above into after each important state variables are determined into score by scoring functions
Corresponding function, obtain probability assignment function of the variable to equipment state;
4. the evaluation result of different important state variables is merged to eliminate conflict
By step 2. middle membership function, the evaluation result of obtained every kind of variable is the variable for each state
Basic probability assignment (BPA) function m:2u→ [0,1], wherein:
M (A) represents the basic probability assignment to comment state A;U is identification framework, i.e., the set of all comments;
Followed by probability conflict composition rule:
After all variables are synthesized to comment shape probability of state assignment function, you can counted and all variables can
The equipment comment confidential interval of reliability;
Wherein m (ASynthesis) it is the synthesis of assigning probability of the different variables for comment state A, i, j represent i-th or j change
Amount, mi(X)、mj(Y) respectively represent i-th, j-th of variable for each comment state in identification framework U basic probability assignment letter
Number.
Further, the step (6) specifically includes:
1. the statistical distribution of the data is brought into marking formula, the score distribution situation F (X of the data are thus obtained
Point);
2. score distribution function is divided according to membership function institute by stages, and respectively to obtaining in each section
Distribution function F (X points) is divided to be integrated;
3. the integral result of integration segment is to characterize the shape probability of state corresponding to the integration segment;
4. then together bring the probability and other information source into fusion formula, you can obtain final equipment state confidence
Section;
It is 5. such as when missing data species is N, missing data in N is successively and existing when missing data more than one
Data bring evaluation model into, obtain N group probability distribution, are then merged according to the weight of each missing data, to obtain to the end
State probability confidential interval.
The present invention has the advantages that:
(1) various for existing distribution net equipment evaluation criterion species, many data are difficult to obtain (or cost performance is relatively low)
Situation, this method are combined using the probability of malfunction of correlation analysis part of appliance, it is determined that the state being had a great influence to equipment becomes
Amount, with existing State Grid Corporation of China company standard Q/GDW 645-2011《Distribution net equipment assessment guidelines》Compare, eliminate one
Be difficult to measure or routine inspection in be not related to, while be the variable relatively small to equipment state influence degree (as being grounded
Cross-sectional area size), to reduce measurement workload.
(2) it is more coarse for data source, the situation of more error information is included, this method is handled using singular value,
To the processing of normal data step, and data smoothing, the methods of data interpolating is filled a vacancy, eliminate because accidental measurement error etc. is made
Into deviation and wrong data, reach the purpose of control data quality.
(3) the probability fusion theory of this method solves different evidence data source contradictions that may be present, will be from data
The state variable that storehouse extracts is handled, and by the integration technology in subsequent step, obtained result is not singly certain variable
Evaluation result, a but more comprehensive conclusion, pass through evaluation model and realize assessment to distribution network equipment state.
(4) this method constantly enters new Data Collection in historical data base, constantly with new data with existing, by new number
According to statistics, obtain new evaluation index, former out-of-date index caused by replace because renewal of the equipment is regenerated etc., improve existing
Index, realize the dynamic evaluation to equipment.
Brief description of the drawings
Fig. 1 is that the extraction flow chart that correlation analysis determines important state variables is carried out in the present invention;
Fig. 2 is D-S theories composition algorithm in the present invention;
Fig. 3 is the schematic flow sheet of the determination and amendment of dynamic parameter in the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the present invention, the technical scheme in the present invention is clearly and completely described.
The present invention provides a kind of distribution net equipment health degree dynamic diagnosis method of meter and trust evaluation, including following step
Suddenly:
(1) the primitive character collection of monitoring power distribution amount is obtained
Data acquisition and monitor control system based on power distribution network, collect the monitoring variable that existing power distribution network can provide, shape
Into primitive character collection database.Primitive character set owner will include:Historical operational information, fault message, bad condition information, example
Row Test Information etc..
(2) important state variables are determined by carrying out correlation analysis to the information of faulty equipment and non-faulting equipment
Fault source tracing method is used to faulty equipment, the every terms of information of faulty equipment is determined in primitive character collection database,
By carrying out correlation analysis with the information of non-faulting equipment, information in database is screened, obtains same fail result phase
The larger characteristic information of closing property, and determine that it is important state variables.Data quality control is carried out first, and then analysis is obtained
Important state variables input state evaluation model, can so reject non-influence amount, reduce range of value, lift the accurate of evaluation
Degree.Process is as shown in Figure 1.
(3) data of the important state variables to being determined in step (2) carry out quality of data pretreatment
Quality of data pretreatment specifically include 3 δ rules singular value is handled, to the processing of normal data step,
Data smoothing processing and data interpolation processing:
3 δ rules are to singular value processing:The normal state point of difference between the same continuous measurement point of data series two is obtained first
Cloth parameter, i.e. its sample average and sample standard deviation, and verify whether its distribution meets normal distribution law, then with this sample
Average and sample standard deviation go to filter legacy data, when 3 times bigger than sample standard deviation of the error between data and sample average with
When upper, it can be determined that can be replaced at this for singular value point, its value caused by interference by the average of 4 points before and after former data.
To the processing of normal data step:Data step phenomenon in measured data be present.The physical quantity measured produces larger
Transition, and near new numerical value continue longer time, then again revert to normal level.
The identification of step signal and retention theory are as follows:The change between the adjacent data of signal two is calculated first, by 3 δ methods
Then determine all transition, the time width between more adjacent two transition.If the time width between the two transition reaches
To preset value (i.e. the duration is longer), then it is considered as the normal transition of system and is preserved;If time width is very narrow, not up to
Preset value (i.e. the duration is extremely short), then it is considered as singular value and is handled by 3 δ rules.
Data smoothing processing:The weighted average of each 2 points totally five point datas is taken before and after step point, for step point edge
The data of position can use the method for reducing smooth region, retain legacy data as far as possible and accomplish corresponding smooth, formula is such as
Under:
Step both ends end points xi=xi;
Step both ends time end points
Step intermediate point
Data interpolating:All data are divided into two classes by interpolation method, one kind be relevance between different measurement points compared with
Strong data, such as temperature data.Within time of measuring and spatial dimension, can mutually it be converted between each thermometric point data, can
To be filled a vacancy with based on the fitting of the time and space with interpolation.
Another kind of data are the poor data of the horizontal comparativity between different measurement points, such as oil dissolved gas gas phase color
Compose analysis, shelf depreciation, leakage current of an arrester, capacitive apparatus leakage current and dielectric loss etc..This kind of data cannot
Its numerical value lost is estimated with homogeneous data, also cannot calculate its numerical value lost with its historical data, therefore at this
The simple method linearly estimated is taken in method, that is, think data during loss by linear change, its two end data by
Given data determines.This method is relatively more conservative, but data statistics is changed less.
Data are eliminated the false and retained the true by these methods, discarded the dross and selected the essential, i.e., smoothing processing have been carried out to noise, to unusual
Value is handled, and the data for making to be input in evaluation model are more regular.
(4) important state variables after step (3) processing are given a mark
For different equipment state variables, scoring method is different, for not quantifiable state variable (as outside
See) use experience scoring;For can quantify but Different Individual due to manufacture etc. the different state variable of reason (as insulate
Resistance) use functional arrangement scoring;For that can quantify and only be used statistics by the state variable (such as connector temperature) of influence on system operation
Distribution map scoring.
(5) the marking result to important state variables is inputted into evaluation model, and draws evaluation result, the evaluation model
Establishment step and internal computation flow it is as follows:
1. establish equipment state Comment gathers:It is good, and it is typically, suspicious, it is abnormal, dangerous;
2. the membership function of Comment gathers is established according to Fuzzy Criteria:
Kilter membership function:
Normal condition membership function:
Suspicious state membership function:
Abnormality membership function:
Precarious position membership function:
3. bring membership function (five kinds of states more than i.e. into after each important state variables are determined into score by scoring functions
Corresponding function), probability assignment function of the variable to equipment state is obtained, it is as follows:
Wherein a%+b%+c%+d%+e%=1.
Such as when certain variable marking 88, bring membership function into, according to the valid interval of function, can obtain good general
Rate 80%, normal probability 20%, other state probabilities are 0%.
4. the evaluation result of different important state variables is merged to eliminate conflict
Step 3. in equipment end-state corresponding to multiple variables often exist conflict with it is inconsistent, to the general of each state
Rate is also not quite similar, and is merged using the evaluation result of different variables to weaken or even eliminate these conflicts.
The evaluation result (state probability) of obtained every kind of variable, it is the variable by step 2. middle membership function
For basic probability assignment (BPA) the function m of each state:2u→ [0,1], wherein:
M (A) represents the basic probability assignment (such as m (good)=a%) to comment state A.U is identification framework, i.e., all
The set of comment.
Followed by probability conflict composition rule:
Wherein m (ASynthesis) represent compound prbability assignment function of two variables to state A.Specific to this method, m (ASynthesis) i.e.
For synthesis of the different variables for comment state A (all comment states form identification framework U) appointment probability, i, j represent i-th
Or j variable, mi(X)、mj(Y) respectively represent i-th, j-th of variable in identification framework U each comment state it is substantially general
Rate assignment function.
After all variables are synthesized to comment shape probability of state assignment function, you can counted and institute on evidence
The equipment comment confidential interval of (all variables) confidence level.Process is as shown in Figure 2.
Specific example is as follows, by taking certain transformer as an example:
Each variable score and the corresponding comment shape probability of state obtained by membership function are as follows:
M in composite formulai(X)、mj(Y) it is:mInsulaion resistance(good/normal/suspicious ...), mTemperature(it is good/normally/can
Doubt ...) or its dependent variable for comment shape probability of state assignment function.
Corresponding comment state probability is as follows after each variable synthesis:
Comment state |
Well |
Normally |
It is suspicious |
It is abnormal |
It is dangerous |
Compound prbability |
98.2% |
1.8% |
0% |
0% |
0% |
(6) evaluation procedure under non-full information
When evaluating certain equipment, if certain variable data missing that should be in input step (5), it will influence equipment and comment
The degree of accuracy of valency and confidence level, therefore using following non-full information evaluation method, replace lacking using big data statistical result
Variable, make evaluation result relatively more accurate:
1. the statistical distribution of the data is brought into marking formula, the score distribution situation F (X of the data are thus obtained
Point);
2. score distribution function is divided according to membership function institute by stages, and respectively to obtaining in each section
Distribution function F (X points) is divided to be integrated;
3. the integral result of integration segment is to characterize the shape probability of state corresponding to the integration segment;
4. then together bring the probability and other information source into fusion formula, you can obtain final equipment state confidence
Section;
It is 5. such as when missing data species is N, missing data in N is successively and existing when missing data more than one
Data bring evaluation model into, obtain N group probability distribution, are then merged according to the weight of each missing data, to obtain to the end
State probability confidential interval.
(7) determination and amendment of dynamic parameter
The important state variables feature of input is once updated at interval of cycle time T, and produces new statistics ginseng
Number:Variance, average etc., statistical parameter old in former scoring functions is replaced with new statistical parameter, realizes the self-renewing of algorithm,
Dynamically scoring functions in evaluation algorithms are modified, and record the relevant parameter after changing, to reach evaluation model with setting
The purpose of self-recision for update.Process is as shown in Figure 3.
The related data statistics of equipment monthly re-starts once, reaches the purpose of dynamic evaluation.It is described above, it is only this
The embodiment of invention, but protection scope of the present invention is not limited thereto, any technology people for belonging to the art
Member the invention discloses technical scope in, the change or replacement that can readily occur in should all be covered in protection scope of the present invention
Within.Therefore, protection scope of the present invention should be defined by scope of the claims.