CN107291991B - Early defect early warning method for wind turbine generator based on dynamic network sign - Google Patents

Early defect early warning method for wind turbine generator based on dynamic network sign Download PDF

Info

Publication number
CN107291991B
CN107291991B CN201710379072.5A CN201710379072A CN107291991B CN 107291991 B CN107291991 B CN 107291991B CN 201710379072 A CN201710379072 A CN 201710379072A CN 107291991 B CN107291991 B CN 107291991B
Authority
CN
China
Prior art keywords
wind turbine
group
turbine generator
dnm
candidate
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710379072.5A
Other languages
Chinese (zh)
Other versions
CN107291991A (en
Inventor
方瑞明
吴敏玲
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.)
Huaqiao University
Original Assignee
Huaqiao University
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 Huaqiao University filed Critical Huaqiao University
Priority to CN201710379072.5A priority Critical patent/CN107291991B/en
Publication of CN107291991A publication Critical patent/CN107291991A/en
Application granted granted Critical
Publication of CN107291991B publication Critical patent/CN107291991B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation

Abstract

The invention discloses a wind turbine generator early defect early warning method based on dynamic network signs, which is characterized in that real-time running data of a wind turbine generator is processed by combining a support vector regression and probability distribution embedding scheme according to the influence characteristics of a topological graph of the wind turbine generator on the state change of the wind turbine generator, then the dynamic network sign method is adopted to detect the dynamic change of the quantization values of the dynamic network signs in different periods before the state of the wind turbine generator is changed, the complex cross-coupling correlation between each subsystem and each element of the wind turbine generator and the critical state characteristics of the wind turbine generator are comprehensively considered in the state evaluation process of the wind turbine generator, and the early warning is carried out on the time point and the position of the wind turbine generator which. The method reduces the influence of large noise on the unit fault prediction precision, can accurately find the critical turning time point of the unit potential fault and position and track the position where the fault possibly occurs in advance, and finally realizes the comprehensive grasp of the trend of unit development and change.

Description

Early defect early warning method for wind turbine generator based on dynamic network sign
Technical Field
The invention relates to the technical field of power generation system security defense, in particular to a wind turbine generator early defect early warning method based on dynamic network signs.
Background
Through the development trend of global new energy in recent years, it can be seen that wind energy has become an important role in energy development, and wind power has also been greatly developed. However, many technical problems still exist in the construction process of the wind power industry in China, and the most prominent of the problems is high failure rate, long failure downtime, large failure influence and expensive maintenance cost of the wind power generator set. According to statistics, the operation and maintenance cost of the wind power industry reaches 10% -15%, which is the biggest bottleneck of the development of the wind power industry, so that the wind power industry is lack of competitiveness in the aspects of safety performance and economic benefit compared with the traditional energy. Therefore, in order to improve the accuracy of state evaluation and prediction of the wind generating set, latent faults of the wind generating set are found in advance, and research and search of a proper wind generating set fault early warning method have important significance for reasonable arrangement of state maintenance of the wind generating set.
The theoretical basis of the wind turbine generator fault early warning research is fault diagnosis. Fault diagnostics are generally classified into mathematical model-based methods and artificial intelligence-based methods. Most of the existing researches aim at different characteristics of different subsystems of a wind turbine generator and different characteristics of monitoring projects, such as temperature measurement, speed, vibration quantity, frequency spectrum, torque and the like of the generator, and the fault diagnosis method is respectively applied to carry out fault early warning on specific elements of the generator, so that the early abnormal operation state of the generator is monitored. Although the wind turbine generator fault early warning method only aiming at a certain specific element or part in the system has clear concept and simple implementation process, the strong interactive coupling between each subsystem and the part of the wind turbine generator is ignored, the development and change trend of the wind turbine generator cannot be comprehensively grasped from the whole level, and the potential fault of the wind turbine generator cannot be judged in advance from the macroscopic system perspective.
With the deepening of the fault early warning research of the wind turbine generator, in recent years, the research angle of the related researches starts to be increased from a specific element in the wind turbine generator to the whole wind turbine generator, and the strong interactive coupling between each subsystem and each component of the wind turbine generator is further considered in the fault early warning method. However, although the existing small part of research can be started aiming at the overall behavior of the fault of the wind turbine, the critical characteristic of the fault of the wind turbine before the state transition is not considered, the evolution process and the characteristic of the fault of the wind turbine cannot be comprehensively grasped, only approximate time early warning can be made on the potential fault of the wind turbine, and the early warning of fixed time and fixed point cannot be achieved; in addition, the large noise interference of the wind turbine generator is not effectively processed in the research process, and the precision of the wind turbine generator fault early warning is influenced. Therefore, how to reduce the influence of the large noise on the fault early warning precision of the unit and consider the strong cross coupling between each subsystem and each element of the unit and the critical characteristic before state transition and the problem needing to be further solved in the fault early warning method of the air inlet electric unit.
Disclosure of Invention
The invention provides a wind turbine generator early-stage defect early-warning method based on a dynamic network marker, which overcomes the defects of the wind turbine generator early-stage defect early-warning method based on the dynamic network marker in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a wind turbine generator early defect early warning method based on dynamic network signs comprises the following steps:
s1: determining multivariable with strong cross-coupling which influences the state change of the wind turbine generator according to a topological graph of the wind turbine generator, selecting a reference data group and an observation data group of each variable by combining support vector regression, performing noise reduction processing on data of the reference data group and the observation data group by using a probability distribution embedding scheme, and obtaining a new reference data sequence and an observation data sequence after processing;
s2: dividing the reference data sequence and the observation data sequence subjected to noise reduction processing in the step S1 into n sections, performing deviation comparison on the reference data sequence and the observation data sequence data in each time period, and selecting variables with significant changes in each time period for correlation clustering to obtain a plurality of candidate DNM groups which significantly influence the state change of the wind turbine generator in each time period;
s3: the mean standard deviation Sd' and the mean pearson correlation coefficient PCC of the variables in each candidate DNM group for each time period obtained in step S2 are calculated respectivelyin' and average Pearson coefficient PCC of variables in each candidate DNM group and the other candidate DNM groupsout' determination of mean standard deviation Sd ' of respective candidate DNM groups, mean Pearson's correlation coefficient PCCin' and mean Pearson coefficient PCCoutWhether three critical characteristics of the wind turbine before state transition are met simultaneously or not is judged, and if yes, the candidate DNM group is used as a dominant DNM group which obviously influences the state change of the wind turbine;
s4: and (4) respectively calculating the quantitative values I of the dynamic network signs of the leading DNM group in each time period in the step S3, and early warning the time point and the position of the unit which possibly have faults by detecting the dynamic changes of the quantitative values of the dynamic network signs in different time periods before the state of the unit is changed.
Wherein, the topological structure diagram of the wind turbine generator in the step S1 is represented by a plurality of monitoring items provided by a SCADA system installed on the wind turbine generator;
the step S1 specifically includes the following steps:
s101: selecting a plurality of main SCADA continuous quantity monitoring projects in a plurality of monitoring projects provided by an SCADA system as a multivariate of research;
s102: based on the historical non-fault data of the multiple SCADA monitoring items selected in the step S101, a support vector machine with regression prediction capability is applied, a single variable is used as the output quantity of a support vector regression model, and other variables are used as the modes of the input quantity of the support vector regression model, and the support vector regression model of each monitoring item is established;
s103: based on the monitoring operation data collected under the condition of safe operation of the wind turbine generator, respectively using the model established in the step S102 to predict the state change trend of each monitoring project, and obtaining prediction data as the reference data group; taking the actually measured operation data of the monitoring items related to the operation state of the wind turbine generator as the observation data group;
s104: based on the probability distribution embedding scheme, selecting a proper window interval k, and converting the reference data group and the observation data group with large noise obtained in the step S103 in a multi-time unfolding mode to obtain a new reference data sequence and an observation data sequence with corresponding probability distribution with smaller noise in a high-dimensional space.
When the reference data set and the observation data set with large noise are converted in the multi-time unfolding mode in step S104, the original dimensions of the reference data set and the observation data set are expanded to the second order, i.e., the first order m1(t) and the second order m2(t), where m1(t) ═ x (t-k) + x (t-k +1) + x (t-k +2) +. + x + (t))]/k,t,k∈N*,t<k，m2(t)＝[(x(t-k)-m1(t))2+(x(t-k+1)-m1(t))2+...+(x(t)-m1(t))2]/k,t,k∈N*,t<Where x (t) is reference data or observation data at a certain time, and k is a window interval.
Wherein, the step S2 specifically includes the following steps:
s201: using t test method to determine the already-classified pointsThe method comprises the specific processes of setting a significance level p to be 0.05, respectively calculating a check value of each monitoring item in each time period, and recording the check value as pi(vi) Where i ═ 1, 2.. n, n denotes the number of a certain period, v denotesiA number indicating a monitoring item;
s202: combining the error discovery rate to judge p of each monitoring item in each time intervali(vi) Whether the value satisfies pi(vi)<(vi× 0.05.05, wherein, control (i) is the data interception length of the ith time interval, if satisfied, it is preliminarily classified as one of the monitoring items that the time interval has significant influence on the state change of the wind turbine;
s203: and (4) performing result correction on the monitoring items with remarkable changes preliminarily obtained in each time period by combining two-time digital variation analysis to obtain the final monitoring items with remarkable changes in each time period, and performing correlation clustering to obtain a plurality of candidate DNM groups.
Wherein, in the step S3, the mean standard deviation Sd' and the mean pearson correlation coefficient PCC of the candidate DNM groupin' and mean Pearson coefficient PCCoutThe calculation formula of' is respectively:
Figure GDA0002425133320000041
in the formula, xi(i ═ 1,2, …, N) is an intra-taggroup variable;
Figure GDA0002425133320000042
representing variables x within the groupiAverage value of (d);
Figure GDA0002425133320000043
in the formula riRepresenting a correlation coefficient between an ith variable in the candidate DNM group and a reference variable in the reference data sequence;
Figure GDA0002425133320000044
in the formula RiThe correlation coefficient between the ith variable in the designated candidate DNM group and the reference variable of the reference data sequence in the other candidate DNM group is expressed.
Wherein a correlation coefficient r between the ith variable in the candidate DNM group and a reference variable in a reference data sequenceiThe calculation formula of (2) is as follows:
Figure GDA0002425133320000045
in the formula, xijA jth sample value representing an ith variable; y isjA jth sample value representing a reference variable of a reference data sequence;
Figure GDA0002425133320000046
a sample mean representing the ith variable;
Figure GDA0002425133320000047
a sample mean value representing a reference variable of a reference data sequence;
a correlation coefficient R between the ith variable in the designated candidate DNM group and a reference variable of a reference data sequence in another candidate DNM groupiThe calculation formula of (2) is as follows:
Figure GDA0002425133320000048
in the formula, xijA jth sample value representing an ith variable; y isjA jth sample value representing a reference variable of a reference data sequence within the other candidate DNM group;
Figure GDA0002425133320000049
a sample mean representing the ith variable;
Figure GDA00024251333200000410
sample means of reference variables representing reference data sequences within other candidate DNM groups.
The three critical characteristics in step S3 specifically include that before the state of the wind turbine generator is changed, the average standard deviation Sd' of the variables in the dominant DNM group is significantly increased, and the average pearson correlation coefficient PCC of each pair of variables in the dominant DNM groupin' average Pearson coefficient PCC that will increase and dominate variables within DNM group and variables within non-dominated DNM groupout' will decrease.
In step S4, the calculation formula of the quantized value I of the dynamic network indicator is:
Figure GDA00024251333200000411
∈ (0,1) is a small positive constant used to avoid denominator being zero.
Compared with the prior art, the invention has the following beneficial effects: according to the invention, starting from the integrity of the wind turbine generator, firstly, a support vector regression model and a probability distribution embedding scheme are combined, after environmental noise of unit operation data is properly processed, then, the complex interactive coupling correlation between each subsystem and elements of the unit and critical state characteristics of the unit are comprehensively considered into the unit state evaluation process by adopting dynamic network signs, the influence of large noise on the unit fault prediction precision is reduced, the critical transition time point of the unit potential fault can be accurately found in advance, the position where the fault possibly occurs can be positioned and tracked, the trend of unit development change can be comprehensively grasped, and the fault before the unit accident occurs can be early warned.
The invention is further explained in detail with the accompanying drawings and the embodiments; however, the early-stage defect early-warning method for the wind turbine generator based on the dynamic network marker is not limited to the embodiment.
Drawings
FIG. 1 is a block flow diagram of the present invention;
FIG. 2 is a topological diagram of a type GE1.5S L E wind turbine of the present invention;
FIG. 3 is a block diagram of the support vector regression model building process of the present invention;
FIG. 4 is a graph of the mean standard deviation Sd' for each candidate DNM group for each time period of the present invention;
FIG. 5 shows the average Pearson correlation coefficient PCC for each candidate DNM group for each time interval of the present inventionin' value;
FIG. 6 shows the average Pearson coefficient PCC of the candidate DNM groups for each time interval according to the present inventionout' value;
FIG. 7 is a quantized value I of the dominant candidate DNM group for each time interval of the present invention;
FIG. 8 is a diagram of the dominant DNM groups and intra-group variable numbers selected at various time periods in accordance with the present invention;
FIG. 9 is a graph of the dynamic network flag change for each monitoring item of FIG. 8 during time period 1 according to the present invention;
FIG. 10 is a graph of the dynamic network flag change for each monitored item of FIG. 8 during time period 2 according to the present invention;
FIG. 11 is a graph of the dynamic network flag change for each monitoring item of FIG. 8 at time period 3 according to the present invention;
FIG. 12 is a graph of the dynamic network flag change for each monitoring item of FIG. 8 at time period 4 in accordance with the present invention;
FIG. 13 is a graph of the dynamic network flag change for each monitoring item of FIG. 8 during time period 5 in accordance with the present invention;
FIG. 14 is a graph of the dynamic network flag change for each monitoring item of FIG. 8 during time period 6 in accordance with the present invention;
FIG. 15 is a graph of the dynamic network flag change for each monitoring item of FIG. 8 during time period 7 in accordance with the present invention;
fig. 16 is a dynamic network flag change diagram of each monitoring item in fig. 8 during time period 8 according to the present invention.
Detailed Description
The method comprises the following steps of firstly selecting a kernel function, carrying out initial parameter processing on the selected kernel function, synchronously learning observation data groups of 33 related monitoring items in 40 minutes before halt as sample data, and sometimes, directly carrying out prediction comparison on the sample data without learning the data groups, carrying out logarithmic learning processing on the selected kernel function, carrying out logarithmic screening processing on the observation data groups, carrying out normalization processing on the observation data groups, and carrying out statistics on the basis of the initial parameter, wherein the statistics on the initial parameter and the initial parameter, and the statistics on the initial parameter, wherein in the embodiment, referring to fig. 1, the present embodiment, taking a wind turbine generator type GE1.5S L E and a single-machine rated capacity of 1.5MW as an example, a topological graph of the unit is shown in fig. 2, MAT L AB as a working platform to simulate the invention, selecting 33 related monitoring items of 33 related data regression model, selecting a prediction value as an observation data group of a model in 40 minutes before the one of a fault caused by the fault as an observation data group as a model, and selecting SVR model, and using a flow as a flow shown in the case, and a flow shown in the example, and a model, and a latent model, and a method for directly carrying out a latent model for carrying out a prediction algorithm for carrying out a prediction comparison and a prediction algorithm for carrying out a prediction for.
33 SCADA continuous quantity monitoring items selected from table 1
Figure GDA0002425133320000061
33 SCADA continuous quantity monitoring items selected in continuation table 1
Figure GDA0002425133320000071
Selecting a proper window interval k for the selected reference data group and the selected observation data group based on a probability distribution embedding scheme, and converting in a multi-time unfolding mode to obtain a new reference data sequence and an observation data sequence with corresponding probability distribution of smaller noise in a high-dimensional space; specifically, the dimensionality of the original reference data set and the original observation data set is expanded to the second order, and the specific calculation formula is as follows:
m1(t)＝[x(t-k)+x(t-k+1)+x(t-k+2)+...+x(t)]/k,t,k∈N*,t<k
m2(t)＝[(x(t-k)-m1(t))2+(x(t-k+1)-m1(t))2+...+(x(t)-m1(t))2]/k,t,k∈N*,t<k
where x (t) is reference data or observation data at a certain time, and k is a window interval.
And after the selected reference data group and the selected observation data group are subjected to the noise reduction treatment, dividing the selected reference data group and the selected observation data group into 8 time periods respectively, wherein each time period comprises 5 sampling points. Then, performing deviation comparative analysis on the reference data group and the observation data group in the divided 8 time periods by adopting a t-test method, specifically, setting a significance level p to be 0.05, respectively calculating test values of monitoring items in each time period, and recording the test values as pi(vi) Where i ═ 1, 2.. n, n denotes the number of the time period, v denotesiA number indicating a monitoring item; then, combining the error discovery rate to judge p of each monitoring item in each time intervali(vi) Whether the value satisfies, pi(vi)<(viAnd (i) × 0.05.05, wherein the control (i) is the data interception length of the ith time interval, if the data interception length is satisfied, the data interception length is preliminarily classified into one of monitoring items which obviously affect the state change of the wind turbine generator in the time interval, and finally, the monitoring items which are obtained preliminarily in each time interval and have obvious changes are subjected to result correction by combining double-number change analysis to obtain the final monitoring items which are obtained in each time interval and subjected to correlation clustering to obtain a plurality of candidate DNM groups.
Inputting the processed data of each time interval into the flow of fig. 1 for dynamic network sign calculation, wherein the average standard deviation Sd' and the average pearson correlation coefficient PCC of the dynamic network sign in each time intervalin' and mean Pearson coefficient PCCout', respectively adopting a calculation formula:
Figure GDA0002425133320000072
in the formula, xi(i ═ 1,2, …, N) is an intra-taggroup variable;
Figure GDA0002425133320000073
representing variables x within the groupiAverage value of (d);
Figure GDA0002425133320000074
in the formula riIndicates the distance between the ith variable in the candidate DNM group and the reference variable of the reference data sequenceThe correlation coefficient is calculated by the formula
Figure GDA0002425133320000081
In the formula, xijA jth sample value representing an ith variable; y isjA jth sample value representing a reference variable of a reference data sequence;
Figure GDA0002425133320000082
a sample mean representing the ith variable;
Figure GDA0002425133320000083
a sample mean value representing a reference variable of a reference data sequence;
Figure GDA0002425133320000084
in the formula RiRepresenting the correlation coefficient between the ith variable in the designated candidate DNM group and the reference variable of the reference data sequence in the other candidate DNM group, and the calculation formula is
Figure GDA0002425133320000085
In the formula, xijA jth sample value representing an ith variable; y isjA jth sample value representing a reference variable of a reference data sequence within the other candidate DNM group;
Figure GDA0002425133320000086
a sample mean representing the ith variable;
Figure GDA0002425133320000087
a sample mean value representing a reference variable of a reference data sequence within the other candidate DNM group; the calculation results are shown in fig. 4-6, so as to obtain the dominant DNM group selected in each time period and the variable number in the group as shown in fig. 8, then the quantized value I of the dominant dynamic network flag is calculated,
Figure GDA0002425133320000088
is a small normal number for avoiding denominator being zero, where the specific calculation result is 0.001 as shown in fig. 7, and the dynamic network flag change corresponding to each time intervalThe chemistry is shown in fig. 9-16.
Specifically, this unscheduled shutdown of the unit is due to the SCADA system giving a "total fault" alarm at a negative output power value at 1 month, 22 months, 2:41:00, 2012. The maintenance personnel confirm that the failure point is the failure of the function of the input template of the transmission control system controller, and the maintenance and the replacement of elements cause the shutdown for about 9 hours.
As can be seen from fig. 4 to 7, compared with the change of each index of the reference data set of the dominant DNM group in each time period, the change of each index of the observation data set of the dominant DNM group in each time period is obvious to a different extent, which shows that the selected dominant DNM group in each time period can gradually represent the transition of the critical state of the unit. The quantized value calculated by the dynamic network marker starts to rapidly rise after the time interval 3, the quantized value reaches the peak at the time interval 4, the most obvious critical transition early warning signal appears in the leading DNM group in the time interval, and the fact that the wind turbine generator can perform early warning 20 minutes before the fault appears through calculation of the quantized value of the dynamic network marker shows that certain effect is achieved on preventing the fault of the wind turbine generator and the spread of cluster cascading faults.
The black filled circles in fig. 9-16 are study-independent items, and the circles with the patterns inside the circles being small black dots or diagonal stripes are dynamic network marker points, where the appearance of circles with diagonal stripe patterns indicates that the system enters a critical transition state. As can be seen from fig. 8 and fig. 9 to 16, in the dynamic network marker dynamic marker locating process in each period, the dynamic markers mainly appear in the transmission control system in a concentrated manner, and when the most obvious critical transition early warning signal appears in period 4, the dynamic locating of the dominant DNM group is the nacelle cabinet temperature and the impeller rotation speed. The engine room cabinet is used as an input pivot of the control system, and has the main functions of collecting signals such as temperature, pressure, rotating speed, environmental parameters and the like measured by various sensors in the engine room, transmitting the information to the tower bottom cabinet through a field bus, and executing signals output by the tower bottom cabinet, such as control of equipment such as variable pitch, yaw, cable release, motors, oil pumps, fans and the like, wherein if the temperature of the engine room cabinet is abnormal, the input function of the controller is influenced. As can be known from the main topological graph (fig. 2) of the wind turbine, the change of the nacelle directly affects the rotating speed of the impeller, so that the fault location is that the nacelle temperature and the rotating speed of the impeller of the transmission control system meet the actual conditions and the actual maintenance results.
Therefore, the wind turbine generator fault early warning method based on the dynamic network marker can comprehensively master the evolution process of the running state of the wind turbine generator, reduce the influence of large noise on the fault early warning precision of the wind turbine generator, detect the critical state of the running change of the wind turbine generator at fixed time and fixed point, and finally play a role in sending out an early warning signal before the wind turbine generator has an accident.
The above embodiment is only used to further illustrate the early-stage defect warning method for the wind turbine generator based on the dynamic network marker, but the present invention is not limited to the embodiment, and any simple modification, equivalent change and modification made according to the technical essence of the present invention to the above embodiment fall within the protection scope of the technical solution of the present invention.

Claims (8)

1. A wind turbine generator early defect early warning method based on dynamic network signs is characterized by comprising the following steps:
s1: determining multivariable with strong cross-coupling which influences the state change of the wind turbine generator according to a topological graph of the wind turbine generator, selecting a reference data group and an observation data group of each variable by combining support vector regression, performing noise reduction processing on data of the reference data group and the observation data group by using a probability distribution embedding scheme, and obtaining a new reference data sequence and an observation data sequence after processing;
s2: dividing the reference data sequence and the observation data sequence subjected to noise reduction processing in the step S1 into n sections, respectively performing deviation comparison on the reference data sequence and the observation data sequence data in each time period, and then respectively selecting variables with obvious changes in each time period for correlation clustering to obtain a plurality of candidate dynamic network mark groups which obviously influence the state change of the wind turbine generator in each time period, wherein the dynamic network mark groups are marked as DNM groups;
s3: the average standard deviation Sd' of the variables within each candidate DNM group for each period obtained in step S2 is calculated,Mean Pearson correlation coefficient PCCin' and average Pearson coefficient PCC of variables in each candidate DNM group and the other candidate DNM groupsout' determination of mean standard deviation Sd ' of respective candidate DNM groups, mean Pearson's correlation coefficient PCCin' and mean Pearson coefficient PCCoutWhether three critical characteristics of the wind turbine before state transition are met simultaneously or not is judged, and if yes, the candidate DNM group is used as a dominant DNM group which obviously influences the state change of the wind turbine;
s4: and (4) respectively calculating the quantitative values I of the dynamic network signs of the leading DNM group in each time period in the step S3, and early warning the time point and the position of the unit which possibly have faults by detecting the dynamic changes of the quantitative values of the dynamic network signs in different time periods before the state of the unit is changed.
2. The early wind turbine generator early defect warning method based on the dynamic network marker as claimed in claim 1, wherein: in the step S1, the topological structure diagram of the wind turbine generator is represented by a plurality of monitoring items provided by an SCADA system installed on the wind turbine generator;
the step S1 specifically includes the following steps:
s101: selecting a plurality of SCADA continuous quantity monitoring projects in a plurality of monitoring projects provided by an SCADA system as a multivariate of research;
s102: based on the historical non-fault data of the multiple SCADA monitoring items selected in the step S101, a support vector machine with regression prediction capability is applied, a single variable is used as the output quantity of a support vector regression model, and other variables are used as the modes of the input quantity of the support vector regression model, and the support vector regression model of each monitoring item is established;
s103: based on the monitoring operation data collected under the condition of safe operation of the wind turbine generator, respectively using the model established in the step S102 to predict the state change trend of each monitoring project, and obtaining prediction data as the reference data group; taking the actually measured operation data of the monitoring items related to the operation state of the wind turbine generator as the observation data group;
s104: based on the probability distribution embedding scheme, selecting a window interval k, and converting the reference data group and the observation data group with large noise obtained in step S103 in a multi-time unfolding manner to obtain a new reference data sequence and an observation data sequence with corresponding probability distribution with smaller noise in a high-dimensional space.
3. The early wind turbine generator early defect warning method based on the dynamic network marker as claimed in claim 2, wherein: when the reference data group and the observation data group with large noise are converted in the multi-time expansion mode in the step S104, the original dimensions of the reference data group and the observation data group are expanded to the second order, namely, the first order m1(t) and the second order m2(t), wherein m1(t) ═ x (t-k) + x (t-k +1) + x (t-k +2) +. + x + x (t)]/k,t,k∈N*,t<k，m2(t)＝[(x(t-k)-m1(t))2+(x(t-k+1)-m1(t))2+...+(x(t)-m1(t))2]/k,t,k∈N*,t<Where x (t) is reference data or observation data at a certain time, and k is a window interval.
4. The early wind turbine generator early defect warning method based on the dynamic network marker as claimed in claim 1, wherein: the step S2 specifically includes the following steps:
s201: and performing deviation comparative analysis on the data in n divided time periods by adopting a t test method, wherein the specific process is that a significance level p is set to be 0.05, the test values of all monitoring items in all time periods are respectively calculated and recorded as pi(vi) Where i ═ 1, 2.. n, n denotes the number of a certain period, v denotesiA number indicating a monitoring item;
s202: combining the error discovery rate to judge p of each monitoring item in each time intervali(vi) Whether the value satisfies pi(vi)<(vi× 0.05.05, wherein, control (i) is the data interception length of the ith time interval, if satisfied, it is preliminarily classified as one of the monitoring items that the time interval has significant influence on the state change of the wind turbine;
s203: and (4) performing result correction on the monitoring items with remarkable changes preliminarily obtained in each time period by combining two-time digital variation analysis to obtain the final monitoring items with remarkable changes in each time period, and performing correlation clustering to obtain a plurality of candidate DNM groups.
5. The early wind turbine generator early defect warning method based on the dynamic network marker as claimed in claim 1, wherein: the mean standard deviation Sd' and the mean pearson correlation coefficient PCC of the candidate DNM group in step S3in' and mean Pearson coefficient PCCoutThe calculation formula of' is respectively:
Figure FDA0002425133310000031
in the formula, xi(i ═ 1,2, …, N) is an intra-taggroup variable;
Figure FDA0002425133310000032
representing variables x within the groupiAverage value of (d);
Figure FDA0002425133310000033
in the formula riRepresenting a correlation coefficient between an ith variable in the candidate DNM group and a reference variable in the reference data sequence;
Figure FDA0002425133310000034
in the formula RiAnd representing a correlation coefficient between the ith variable in the designated candidate DNM group and a reference variable of a reference data sequence in other candidate DNM groups, wherein N is the total number of variables in the candidate dynamic network marker group.
6. The early wind turbine generator early defect warning method based on the dynamic network marker as claimed in claim 5, wherein: a correlation coefficient r between the ith variable in the candidate DNM group and a reference variable in a reference data sequenceiThe calculation formula of (2) is as follows:
Figure FDA0002425133310000035
in the formula, xijTo representThe jth sample value of the ith variable; y isjA jth sample value representing a reference variable of a reference data sequence;
Figure FDA0002425133310000036
a sample mean representing the ith variable;
Figure FDA0002425133310000037
a sample mean value representing a reference variable of a reference data sequence;
a correlation coefficient R between the ith variable in the designated candidate DNM group and a reference variable of a reference data sequence in another candidate DNM groupiThe calculation formula of (2) is as follows:
Figure FDA0002425133310000038
in the formula, xijA jth sample value representing an ith variable; y isjA jth sample value representing a reference variable of a reference data sequence within the other candidate DNM group;
Figure FDA0002425133310000039
a sample mean representing the ith variable;
Figure FDA00024251333100000310
sample means of reference variables representing reference data sequences within other candidate DNM groups.
7. The early wind turbine generator early defect warning method based on the dynamic network marker as claimed in claim 1, wherein: the three critical characteristics in step S3 specifically include that before the state of the wind turbine generator is changed, the average standard deviation Sd' of the variables in the dominant DNM group is increased, and the average pearson correlation coefficient PCC of each pair of variables in the dominant DNM groupin' average Pearson coefficient PCC that will increase and dominate variables within DNM group and variables within non-dominated DNM groupout' will decrease.
8. A dynamic based on claim 1The early defect early warning method of the wind turbine generator with the network sign is characterized by comprising the following steps: the calculation formula of the quantized value I of the dynamic network indicator in step S4 is as follows:
Figure FDA0002425133310000041
∈ (0,1) is a small positive constant used to avoid denominator being zero.
CN201710379072.5A 2017-05-25 2017-05-25 Early defect early warning method for wind turbine generator based on dynamic network sign Active CN107291991B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710379072.5A CN107291991B (en) 2017-05-25 2017-05-25 Early defect early warning method for wind turbine generator based on dynamic network sign

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710379072.5A CN107291991B (en) 2017-05-25 2017-05-25 Early defect early warning method for wind turbine generator based on dynamic network sign

Publications (2)

Publication Number Publication Date
CN107291991A CN107291991A (en) 2017-10-24
CN107291991B true CN107291991B (en) 2020-08-07

Family

ID=60094043

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710379072.5A Active CN107291991B (en) 2017-05-25 2017-05-25 Early defect early warning method for wind turbine generator based on dynamic network sign

Country Status (1)

Country Link
CN (1) CN107291991B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109993189A (en) * 2018-01-02 2019-07-09 中国移动通信有限公司研究院 A kind of network failure method for early warning, device and medium
CN108563525B (en) * 2018-04-13 2021-04-20 中国人民解放军陆军装甲兵学院 Intelligent body environment task interference method and system in complex system reliability simulation
CN109657982B (en) * 2018-12-20 2022-02-11 三一重能有限公司 Fault early warning method and device
CN110503133A (en) * 2019-07-26 2019-11-26 东北大学 A kind of centrifugal compressor failure prediction method based on deep learning
CN111579731B (en) * 2020-06-28 2022-07-29 华侨大学 Early-stage defect early warning method for transformer based on combined model
CN112232427B (en) * 2020-10-13 2023-10-03 宁波大学 Wind driven generator fault detection method based on distributed regression model
CN112307619A (en) * 2020-10-29 2021-02-02 北京航天测控技术有限公司 Construction method of early warning model, and equipment fault early warning method and device
CN112861350B (en) * 2021-02-08 2023-03-07 华侨大学 Temperature overheating defect early warning method for stator winding of water-cooled steam turbine generator

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7216061B2 (en) * 2004-08-25 2007-05-08 Siemens Corporate Research, Inc. Apparatus and methods for detecting system faults using hidden process drivers
WO2012050262A1 (en) * 2010-10-15 2012-04-19 한국전력공사 Method and system for monitoring the performance of plant instruments using ffvr and glrt
CN104951851A (en) * 2015-07-08 2015-09-30 华侨大学 Wind turbine state prediction model establishing method based on grey relation-regression SVM (support vector machine)
CN105160444A (en) * 2015-10-22 2015-12-16 广东电网有限责任公司电力调度控制中心 Electrical equipment failure rate determining method and system
CN106096276A (en) * 2016-06-13 2016-11-09 江南大学 A kind of swine flu epidemic situation early warning models based on dynamic network mark

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7216061B2 (en) * 2004-08-25 2007-05-08 Siemens Corporate Research, Inc. Apparatus and methods for detecting system faults using hidden process drivers
WO2012050262A1 (en) * 2010-10-15 2012-04-19 한국전력공사 Method and system for monitoring the performance of plant instruments using ffvr and glrt
CN104951851A (en) * 2015-07-08 2015-09-30 华侨大学 Wind turbine state prediction model establishing method based on grey relation-regression SVM (support vector machine)
CN105160444A (en) * 2015-10-22 2015-12-16 广东电网有限责任公司电力调度控制中心 Electrical equipment failure rate determining method and system
CN106096276A (en) * 2016-06-13 2016-11-09 江南大学 A kind of swine flu epidemic situation early warning models based on dynamic network mark

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Failure Prediction of Wind Turbines Using Improved Gray Relation Analysis Based Support Vector Machine Method;Fang R ET AL;《Journal of Computational & Theoretical Nanoscience》;20160901;5887-5895 *
Forecasting abrupt changes in foreign exchange markets: method using dynamical network marker;Oya S ET AL;《New Journal of Physics》;20151209;1-65 *
基于SCADA和支持向量回归的风电机组状态在线评估方法;梁颖等;《电力系统自动化》;20130725;7-12,31 *
电力系统故障预测方法研究;许奇歆等;《电工文摘》;20160630;6-11 *

Also Published As

Publication number Publication date
CN107291991A (en) 2017-10-24

Similar Documents

Publication Publication Date Title
CN107291991B (en) Early defect early warning method for wind turbine generator based on dynamic network sign
EP3638900B1 (en) Independent monitoring system for a wind turbine
CN107742053B (en) Wind turbine generator set abnormity identification method and device
CN101799674B (en) Method for analyzing service state of numerical control equipment
CN108072524B (en) Wind turbine generator gearbox bearing fault early warning method
CN112784373A (en) Fault early warning method for wind turbine generator gearbox
KR20160073945A (en) System and method for managing wind plant
CN112487910A (en) Fault early warning method and system for nuclear turbine system
CN112799898B (en) Interconnection system fault node positioning method and system based on distributed fault detection
CN110006552B (en) Method for detecting abnormal temperature of equipment
CN110926809A (en) Big data analysis-based wind turbine generator transmission chain fault early warning method
CN112733283A (en) Wind turbine generator component fault prediction method
CN111878320B (en) Monitoring method and system of wind generating set and computer readable storage medium
CN111125895B (en) On-line assessment method for service power characteristics of complete wind driven generator
CN103925155A (en) Self-adaptive detection method for abnormal wind turbine output power
CN106441843B (en) A kind of rotating machinery fault method for waveform identification
CN104074687A (en) Load and performance testing method and device used for megawatt wind generation set
CN113374652A (en) Method for evaluating service life of wind generating set
CN110469461B (en) Fracture estimation method and device for fan toothed belt and readable storage medium
CN114742363A (en) Energy efficiency state evaluation method, system and medium for wind turbine generator
US11339763B2 (en) Method for windmill farm monitoring
CN115929569A (en) Fault diagnosis method for variable pitch system of wind turbine generator
CN115575579A (en) Carbon monitoring method and system based on monitoring source analysis
CN115456041A (en) Equipment fault early warning method and device, computing equipment and storage medium
CN115578084A (en) Wind turbine generator set frequency converter fault early warning method based on deep convolution self-encoder

Legal Events

Date Code Title Description
PB01 Publication
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