CN111563685B - Power generation equipment state early warning method based on auto-associative kernel regression algorithm - Google Patents

Power generation equipment state early warning method based on auto-associative kernel regression algorithm Download PDF

Info

Publication number
CN111563685B
CN111563685B CN202010385882.3A CN202010385882A CN111563685B CN 111563685 B CN111563685 B CN 111563685B CN 202010385882 A CN202010385882 A CN 202010385882A CN 111563685 B CN111563685 B CN 111563685B
Authority
CN
China
Prior art keywords
state
parameter
formula
vector
value
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
CN202010385882.3A
Other languages
Chinese (zh)
Other versions
CN111563685A (en
Inventor
李刚
仇晨光
曹帅
王亚欧
于国强
郑建勇
陈波
杨振
耿察民
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.)
Southeast University
State Grid Jiangsu Electric Power Co Ltd
Jiangsu Fangtian Power Technology Co Ltd
Original Assignee
Southeast University
State Grid Jiangsu Electric Power Co Ltd
Jiangsu Fangtian Power Technology 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 Southeast University, State Grid Jiangsu Electric Power Co Ltd, Jiangsu Fangtian Power Technology Co Ltd filed Critical Southeast University
Priority to CN202010385882.3A priority Critical patent/CN111563685B/en
Publication of CN111563685A publication Critical patent/CN111563685A/en
Application granted granted Critical
Publication of CN111563685B publication Critical patent/CN111563685B/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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas 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
    • 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
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • 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

Abstract

The invention provides a power generation equipment state early warning method based on an auto-associative kernel regression algorithm, which applies a clustering idea to distance division and provides a variable interval state matrix extraction method based on clustering.

Description

Power generation equipment state early warning method based on auto-associative kernel regression algorithm
Technical Field
The invention relates to the technical field of state monitoring of generator set equipment, in particular to a power generation equipment state early warning method based on an auto-associative kernel regression algorithm.
Background
With the continuous deepening of the innovation of the power market system, power generation enterprises face increasingly intensified industrial competition, the maintenance of the safe, economic and environment-friendly long-term stable operation of a unit is increasingly important, how to reasonably arrange the unit maintenance period and improve the utilization rate of each main and auxiliary device becomes a key link. The traditional condition inspection inevitably has the phenomena of blind planning such as 'insufficiency' and 'excess', and the like, so that the condition inspection, a new inspection mode, becomes the focus of attention of researchers in the industry. The key of the state maintenance lies in the accurate acquisition of the equipment state and early warning.
The development of information science technology provides a new idea for the state monitoring and early warning of equipment, and the real-time modeling and knowledge discovery technology for big data is matched with massive historical/real-time data in an information system of a power generation enterprise, wherein the intelligent early warning and diagnosis technology of the equipment based on data modeling and mining is developed rapidly, related technical reports are continuously emerged, and the research results are found: establishing a hidden relation among all parameters of the primary fan by using an association mining method so as to monitor the state of the fan; analyzing a subsequence mode of time by using a transaction search method in data mining, and judging a time sequence abnormal state; establishing a state early warning model of a power generation unit powder preparation system based on a nonlinear state estimation principle, and acquiring a historical state matrix through clustering; a nonparametric estimation model of the power station fan is established by utilizing a multivariate state estimation method, and an improved measure of deviation and a sliding window is introduced to set an early warning line; the nonlinear state estimation method is applied to temperature monitoring of the wind turbine gearbox, and the statistical characteristic of residual errors is analyzed; other technical applications related to the early warning of the equipment state also include gaussian models, clustering methods, correlation analysis methods, auto-associative kernel regression (AAKR), and the like.
However, most of the existing DCS control systems only provide a single-point alarm and related protection function, and only can take emergency measures or provide post analysis after an accident occurs, and thus, the existing DCS control systems cannot effectively monitor the state of equipment, and even cannot timely provide an early warning signal in the early stage of small degradation.
Disclosure of Invention
Technical problem to be solved
The invention provides a power generation equipment state early warning method based on an auto-associative kernel regression algorithm, which aims to solve the practical problems mentioned in the background technology.
(II) technical scheme
In order to achieve the purpose, the invention provides the following technical scheme: a power generation equipment state early warning method based on an auto-associative kernel regression algorithm specifically comprises the following steps:
step 1: constructing a variable-interval historical matrix based on clustering;
step 2: optimizing parameters in a training process through a cross validation method to obtain adjustable parameter bandwidth, and standardizing a state vector based on an AAKR algorithm;
and step 3: and constructing equipment state health degree indexes by using parameter estimation residuals through an AAKR model to track, monitor and early warn the equipment state.
Further, the step 1 specifically includes the following steps:
step 1.1: determining a main parameter by using artificial experience, equally dividing the main parameter by n, determining a vector nearest to a spacing point, and clustering data by taking n as a target class number, wherein n is a positive integer;
step 1.2: grouping the class center and the equidistant division results into a set;
step 1.3: sequencing the vector set to determine a boundary vector;
step 1.4: adding vectors according to the sequence by taking the boundary as a starting point;
step 1.5: judging the main parameter interval of the vector, if the main parameter interval is larger than n/2, adding the vector, and if the main parameter interval is smaller than n/2, deleting the vector;
step 1.6: a final historical storage state matrix is determined.
Further, the step 2 specifically includes:
assume that the observation vector of the current state is
Figure BDA0002483900100000021
The stored historical state matrix is
Figure BDA0002483900100000022
The method is an n multiplied by m data matrix, wherein the number of rows represents the number of stored historical state vectors, the number of columns represents the number of parameters contained in the state vectors, and the number is expressed by a formula as follows:
Figure BDA0002483900100000023
the AAKR algorithm is implemented by combining observation vectors
Figure BDA0002483900100000024
Mapping to a state matrix
Figure BDA0002483900100000025
The expected value of the decision, derived by a state vector combination:
Figure BDA0002483900100000026
wherein the content of the first and second substances,
Figure BDA0002483900100000031
for the estimated values determined from the state matrix and the observation vector,
Figure BDA0002483900100000032
representing a mapping function, and assuming that parameters are independent of each other, S is defined as:
Figure BDA0002483900100000033
in formula (3)
Figure BDA0002483900100000034
Representing the statistical variance of the jth parameter in the history storage state vector, as shown in equation (2)
Figure BDA0002483900100000035
And
Figure BDA0002483900100000036
respectively substituting into x (j) to realize data standardization:
Figure BDA0002483900100000037
mu in the formula (4)jIs the average value of the jth parameter in the history storage state matrix, y (j) is
Figure BDA0002483900100000038
And
Figure BDA0002483900100000039
normalized standard values;
the state vector in the history storage matrix is combined, as shown in formula (5):
Figure BDA00024839001000000310
w (k) in the formula (5) is a corresponding weight coefficient representing the current observation vector
Figure BDA00024839001000000311
And the kth state vector xnc(k) A measure of the similarity between the two,
Figure BDA00024839001000000312
for the estimated value of the jth parameter, the similarity measure is calculated by using high-dimensional space mapping in the AAKR algorithm, and the kernel function inner product is introduced:
Figure BDA00024839001000000313
phi in the formula (6) isMapping function of observation vector from RJThe space maps to a high-dimensional Euclidean space H,<>for the inner product operator, the traditional AAKR algorithm uses a gaussian kernel function to perform the calculation:
Figure BDA00024839001000000314
wherein h is the adjustable parameter bandwidth;
according to Mercer's kernel theory, the kernel function shown in equation (8) can be considered as the inner product of an infinite dimension Euclidean space:
Figure BDA00024839001000000315
the right side of the formula (8) is a mathematical expression form of an infinite dimension Euclidean space inner product;
and (3) calculating by adopting a Mahalanobis distance operator:
Figure BDA00024839001000000316
derived from
Figure BDA0002483900100000041
As an estimate in a plant state health indicator model
Figure BDA0002483900100000042
Further, the adjustable parameter bandwidth h in the formula (7) is obtained by optimizing parameters in a training process through a cross-validation method, and specifically includes a maximum value h of a given bandwidth hmaxAnd a minimum value hminH is to bemaxAnd hminThe method is divided into M equal parts, and for each h, the root mean square error under the 4-fold cross validation mode is calculated, wherein the calculation formula is as follows:
Figure BDA0002483900100000043
SSR in formula (10)hI.e. the root mean square error of the 4-fold cross validation,
Figure BDA0002483900100000044
the observed value of the jth parameter of the ith group of samples under the k modeling condition is shown, ts is the number of groups of training samples of each group after division,
Figure BDA0002483900100000045
for the estimated value of the jth variable of the ith group of samples under the condition of the kth modeling, the value formula of h in the change interval is as follows:
Figure BDA0002483900100000046
h in formula (11)mIs the bandwidth value of the mth time, M is the bandwidth in hmaxAnd hminThe number of the divided regions in the range and the bandwidth coefficient h are from the minimum value hminTo a maximum value hmaxIn the process (2), there is an optimum coefficient hoptI.e. SSRhThe point with the minimum value is taken, the error of model estimation under the 4-fold cross validation learning mechanism is the minimum, namely the optimal bandwidth coefficient, and h is taken asoptThe adjustable parameter bandwidth h in equation (5) is taken.
Further, the step 3 specifically includes: constructing equipment state health degree indexes through the standardized parameter estimation residuals in the step 2, and tracking, monitoring and early warning the equipment state, wherein the equipment state health degree indexes are as follows:
Figure BDA0002483900100000047
in the formula (12)
Figure BDA0002483900100000048
Is an observed value of the parameter(s),
Figure BDA0002483900100000049
is an estimated value of the parameter, and s is a health index.
(III) compared with the prior art, has the following beneficial effects
The invention discloses a power generation equipment state early warning method based on an improved auto-associative kernel regression algorithm, which applies a clustering idea to distance division, provides a variable interval state matrix extraction method based on clustering, adopts a cross validation learning mechanism, optimizes model parameters in a training process, obtains an optimal state monitoring model, and can generate an equipment state early warning signal in advance for field personnel to strive for processing time and reduce loss caused by faults.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a method for extracting a variable interval state matrix based on clustering;
FIG. 2 is a diagram of an equipment failure degradation process;
FIG. 3 is a schematic diagram of 4-fold cross-validation training;
FIG. 4 is a bandwidth factor optimization diagram;
FIG. 5 is a primary air fan parameter state estimation diagram;
FIG. 6 is a diagram of coal mill parameter state estimation;
FIG. 7 is a primary fan state warning diagram, wherein a is a primary fan health trend graph and b is a fan current estimation graph;
FIG. 8 is a coal mill state warning diagram, wherein a is a coal mill health trend graph and b is a coal mill extraction pressure estimation graph.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention relates to a power generation equipment state early warning method based on an improved auto-associative kernel regression algorithm, in particular to an equipment state monitoring and early warning method based on an AAKR algorithm, which applies a clustering idea to distance division and provides a variable interval state matrix extraction method based on clustering.
The method comprises the following implementation steps:
(1) variable-interval history matrix construction method based on clustering
As shown in fig. 1, it is known from the principle of the AAKR algorithm that the estimation value of the observation vector at the current time is obtained by performing weighted combination on all vectors in the history storage state matrix, and it is obvious that the history storage state matrix has a critical influence on the application of the algorithm. The current method for acquiring the historical storage state matrix mainly comprises the following steps:
the direct application method comprises the following steps: according to the determined related parameters, collecting historical data of a certain period, and directly using the historical data as a historical storage state matrix for application;
second, equidistant sampling method: firstly, collecting a large amount of historical data, selecting a certain main parameter, carrying out equidistant division on the whole data set by taking the parameter as a reference, and selecting an equidistant boundary point set to cooperate as a historical storage state matrix;
thirdly, clustering method: a matrix capable of representing each typical running state of the object is extracted from a large amount of historical data, the matrix is similar to the clustering concept in a mathematical method, and a class center set obtained by clustering is selected as a historical storage state matrix.
The direct application method and the equidistant sampling method are both results of manual selection and are greatly influenced by subjective factors; although the clustering method can avoid the disadvantage of randomness, the advantages of manual experience are lost. Therefore, the clustering idea is applied to distance division, and a variable interval state matrix extraction method based on clustering is provided, and the specific steps are shown in fig. 1. The method utilizes artificial experience to determine main parameters, utilizes a clustering algorithm to dominate a vector extraction process, overcomes the randomness defect of equidistant division, and finally obtains a variable-interval state vector set which is used as a state matrix to realize the calculation of state estimation.
(1.1) State Warning theory
At present, thermal generator sets are all provided with Distributed Control Systems (DCS), and alarm limits, tripping limits and the like are set for important monitoring parameters related to all equipment. When a certain parameter reaches or exceeds the alarm limit DCS system, an alarm signal of the corresponding parameter is given on an operator picture, and when the related parameter of the equipment reaches or exceeds the trip limit DCS system, the corresponding main and auxiliary equipment is tripped out through a protection logic instruction. For thermal power generating units, in case of sudden load drop or trip event, which will cause great economic loss, the operation manager wants to give early warning signal in the early stage of equipment failure or system abnormality, so as to strive for more time for relevant personnel to deal with, and prevent or delay parameter alarm or forced tripping of equipment.
Theoretically, the equipment failure generally deteriorates after a certain period of time from early stage to alarm trip, and the deterioration process is shown in fig. 2. The abscissa in the figure represents time and the ordinate represents the state of the device, described by various parameters, such as temperature, pressure, flow, current, etc.
As shown in fig. 2, a black curve in the graph represents a monitoring parameter of the equipment and represents an operation state trend of the equipment, two dotted lines represent a normal change interval of the operation of the parameter, once the normal change interval is exceeded, the operation state of the equipment begins to be slightly degraded, and the development reaches an alarm limit after a period of time (represented by a dotted line), the system gives an alarm signal, if the development trend cannot be effectively controlled, the state parameter finally changes to a protection fixed value (represented by a solid line), and the protection logic is triggered to cause the equipment to trip. As shown in fig. 1, generally, in order to prevent the interference of parameter fluctuation on normal operation, the difference between the alarm fixed value setting and the protection fixed value of the parameter in the DCS system is not too large, the time for intervention and rescue after an operator finds an alarm signal on an operation interface is very limited, and the effect is generally difficult to ensure.
The part of the state parameter trend curve in the ellipse in fig. 2 is an interval from the beginning of the equipment generating tiny deterioration to the parameter alarming, the parameter usually has a larger difference from exceeding the normal interval to the alarming limit, and the equipment also has a certain time from the beginning of the deterioration to the protection tripping. If the degradation information of the equipment operation state can be monitored in the interval by a technical method, sufficient processing time is provided for field personnel, the probability of tripping the equipment is reduced, and the equipment state monitoring in the oval range is generally called early state warning.
(2) AAKR algorithm
The basic idea of the AAKR algorithm is: and estimating and reconstructing the real-time observation state vector by using the normal state sample to obtain an estimation value in the current normal state, and judging the current operation state by comparing the deviation between the observation value and the estimation value. Assume that the observation vector of the current state is
Figure RE-GDA0002570808150000071
The stored historical state matrix is
Figure RE-GDA0002570808150000072
Figure RE-GDA0002570808150000073
The method is an n multiplied by m data matrix, wherein the row number represents the number of stored historical state vectors, the column number represents the number of parameters contained in the state vectors, and the formula is as follows:
Figure BDA0002483900100000073
the AAKR algorithm is implemented by combining observation vectors
Figure BDA0002483900100000074
Mapping to a state matrix
Figure BDA0002483900100000075
The expected value of the decision, derived by a state vector combination:
Figure BDA0002483900100000076
wherein the content of the first and second substances,
Figure BDA0002483900100000077
for the estimated values determined from the state matrix and the observation vector,
Figure BDA0002483900100000078
representing a mapping function, and assuming that parameters are independent of each other, S is defined as:
Figure BDA0002483900100000079
in formula (3)
Figure BDA00024839001000000710
Representing the statistical variance of the jth parameter in the history storage state vector, as shown in equation (2)
Figure BDA00024839001000000711
And
Figure BDA00024839001000000717
respectively substituting into x (j) to realize data standardization:
Figure BDA00024839001000000712
mu in the formula (4)jIs the average value of the jth parameter in the history storage state matrix, y (j) is
Figure BDA00024839001000000713
And
Figure BDA00024839001000000714
normalized standard values.
The state vector in the history storage matrix is combined, as shown in formula (5):
Figure BDA00024839001000000715
w (k) in the formula (5) is a corresponding weight coefficient representing the current observation vector
Figure BDA00024839001000000716
And the kth state vector xnc(k) A measure of the similarity between the two,
Figure BDA0002483900100000081
for the estimated value of the jth parameter, the similarity measure is calculated by using high-dimensional space mapping in the AAKR algorithm, and the kernel function inner product is introduced:
Figure BDA0002483900100000082
phi in equation (6) is the mapping function, taking the observation vector from RJThe space maps to a high-dimensional Euclidean space H,<>for the inner product operator, the traditional AAKR algorithm uses a gaussian kernel function to perform the calculation:
Figure BDA0002483900100000083
wherein h is the adjustable parameter bandwidth;
according to Mercer's kernel theory, the kernel function shown in equation (8) can be considered as the inner product of an infinite dimension Euclidean space:
Figure BDA0002483900100000084
the right side of the formula (8) is a mathematical expression form of an infinite dimension Euclidean space inner product;
and (3) calculating by adopting a Mahalanobis distance operator:
Figure BDA0002483900100000085
it can be seen that the mahalanobis distance between the normalized state vectors is the euclidean distance, which is essentially a special case of the mahalanobis distance. Compared with Euclidean distance, the Mahalanobis distance considers dimension difference among parameters, reduces the influence of the dimension difference on the stability of the algorithm, and obtains
Figure BDA0002483900100000086
As an estimate in a plant state health indicator model
Figure BDA0002483900100000087
(2.1) Cross-validation learning mechanism
The gaussian kernel function shown in formula (7) is used for measuring the distance difference between state vectors, but the function contains an adjustable parameter bandwidth h, the selection of the adjustable parameter bandwidth h has a great influence on the application effect of the algorithm, and the parameters are optimized in the training process by adopting a 4-fold cross validation method, as shown in fig. 3.
The 4-fold cross validation training method is to divide the acquired data into 4 sub-sample sets with small number difference, wherein 3 sub-sets are used for training, 1 sub-set is used for testing, the sub-sets used for testing are exchanged in turn, and the whole cross validation carries out 4 model training tests.
The specific process is as follows: maximum value h of given bandwidth hmaxAnd a minimum value hminH is to bemaxAnd hminDividing the test result into M equal parts, and calculating the root mean square error under the 4-fold cross validation mode for each hThe difference, the calculation formula is:
Figure BDA0002483900100000088
SSR in formula (10)hI.e. the root mean square error of the 4-fold cross validation,
Figure BDA0002483900100000089
the observed value of the jth parameter of the ith group of samples under the k modeling condition is shown, ts is the number of groups of training samples of each group after division,
Figure BDA0002483900100000091
for the estimated value of the jth variable of the ith group of samples under the condition of the kth modeling, the value formula of h in the change interval is as follows:
Figure BDA0002483900100000092
h in formula (11)mIs the bandwidth value of the mth time, M is the bandwidth in hmaxAnd hminThe number of the divided regions in the range. It can be seen that when SSR is usedhThe minimum value corresponds to the optimal bandwidth coefficient h.
Taking the primary air fan parameter model shown in table 1 as an example, 800 groups of data under normal operation conditions are collected and divided into 4 sub-sample sets with the same number, each sub-sample set comprises 200 samples, the change interval of h is set to be 0.02-0.24, the interval is 0.02, the bandwidth coefficient is optimized by adopting a 4-fold cross validation training method, and the obtained result is shown in fig. 4.
From fig. 4, it can be seen that the bandwidth factor h is taken from the minimum value hminTo a maximum value hmaxIn the process (2), there is an optimum coefficient hoptNamely, the point with the minimum SSR value in the graph, at this time, the error of model estimation under the 4-fold cross validation learning mechanism is the minimum, namely, the optimal bandwidth coefficient.
(3) Device status early warning application
(3.1) monitoring parameter determination
Coal mills and primary blowers in thermal power generating sets are core equipment which are responsible for providing a boiler combustion fuel task, and the current, temperature, pressure, vibration and the like of equipment operation are important parameters needing to be monitored. A600 MW unit in China is taken as an object, and according to the operation principle of equipment, a relevant monitoring parameter model of primary air fan and coal mill equipment is established and is shown in tables 1 and 2.
TABLE 1 Primary air monitoring parameters
Figure BDA0002483900100000093
Figure BDA0002483900100000101
TABLE 2 coal pulverizer monitoring parameters
Figure BDA0002483900100000102
The primary air fan extraction parameters in table 1 mainly include shaft temperature, shaft vibration, current, pressure and the like, and the mechanical performance of the information such as the shaft temperature and the vibration is monitored in an important point; in table 2, the monitoring parameters of the coal mill mainly include primary air pressure, air powder temperature, cold and hot air valve positions, motor current and the like, and the operation performance of the equipment is mainly monitored.
(3.2) device Condition monitoring
According to the model determined by the cross validation learning mechanism, 1000 groups of data in a continuous time are selected, the sampling period is 30S, and the results of primary air fan and coal mill parameter monitoring are shown in fig. 5 and 6.
Fig. 5 and fig. 6 respectively show the results of the state estimation of the parameters of the primary air fan and the coal pulverizer, from which it can be seen that the state estimation model established by the AAKR algorithm can accurately estimate the relevant parameters of the equipment, and the state monitoring of the equipment is realized by tracking the measured values in real time.
(3.3) device status Pre-warning
All the parameters are standardized through the step 2, then modeling calculation is carried out by using the standardized values, the finally obtained result is an estimated value, the formula (12) is a fitting index of the estimated value and an observed value, and a health index, the AAKR model utilizes parameter estimation residual errors to construct equipment state health indexes, and the calculation formula is as follows:
Figure BDA0002483900100000111
in the formula (12)
Figure BDA0002483900100000112
Is the value of the observed value and is,
Figure BDA0002483900100000113
the estimated value of the parameter after the standard is adopted, s is a health degree index, and when the obviously estimated residual error is smaller than a set early warning value, the current state of the equipment is normal.
By means of a thermal power generating unit mechanism model, simulation of two faults of primary fan performance reduction and coal mill blocking is carried out respectively, the operation state change of equipment is reflected through a health degree index, and the health degree index trends of the primary fan and the coal mill based on an AAKR model are shown in the figures 7 and 8.
The intersection point in fig. 7 and fig. 8 is the time point of triggering the warning signal indicating the state of the equipment, and it can be seen from this that almost at the beginning of the fault, the health index of the AAKR state model already indicates that there is a significant trend change in the operating state of the equipment, and at this time, the measured value of the monitored parameter of the equipment also shows a trend deviation from the estimated value, as shown in fig. 7(b) and fig. 8(b), but at this time, the value of the parameter does not reach the alarm value and the protection value in the control system, and the system alarm is not triggered. Obviously compare with control system, the AAKR model can produce equipment state early warning signal in advance, strives for processing time for field personnel, reduces the loss that the trouble arouses.
A state monitoring and early warning model of the power generation equipment is established based on an AAKR algorithm, the clustering idea is applied to distance division, a variable interval historical storage state matrix extraction method based on clustering is provided, samples are divided into 4 parts with equal number in the model training process, a 4-fold cross validation learning mechanism is adopted to optimize the model parameter bandwidth, and an optimal equipment state monitoring model is obtained. The reliability and effectiveness of the method are verified through relevant field operation data and fault simulation calculation of a primary fan and coal mill equipment of a certain 600MW unit.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not to be considered in all respects as illustrative or restrictive, and the invention is not to be limited to the specific embodiments shown and described. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims (4)

1. A power generation equipment state early warning method based on an auto-associative kernel regression algorithm is characterized by comprising the following steps:
step 1: constructing a variable-interval historical matrix based on clustering;
step 2: optimizing parameters in a training process through a cross validation method to obtain adjustable parameter bandwidth, and performing state vector standardization based on an AAKR algorithm;
and step 3: constructing equipment state health degree indexes by using parameter estimation residuals through an AAKR model to track, monitor and early warn the equipment state;
the step 2 specifically comprises:
assume that the observation vector of the current state is
Figure FDA0002483900090000011
The stored historical state matrix is
Figure FDA0002483900090000012
Figure FDA0002483900090000013
The method is an n multiplied by m data matrix, wherein the row number represents the number of stored historical state vectors, the column number represents the number of parameters contained in the state vectors, and the formula is as follows:
Figure FDA0002483900090000014
the AAKR algorithm is implemented by combining observation vectors
Figure FDA0002483900090000015
Mapping to a state matrix
Figure FDA0002483900090000016
The expected value of the decision, derived by a state vector combination:
Figure FDA0002483900090000017
wherein the content of the first and second substances,
Figure FDA00024839000900000110
for the estimated values determined from the state matrix and the observation vector,
Figure FDA00024839000900000111
representing a mapping function, assuming that the parameters are in independent relation, S is defined as:
Figure FDA0002483900090000018
in formula (3)
Figure FDA00024839000900000112
Representing the statistical variance of the jth parameter in the history storage state vector, as shown in equation (2)
Figure FDA00024839000900000113
And
Figure FDA00024839000900000114
respectively substituting into x (j) to realize data standardization:
Figure FDA0002483900090000019
mu in the formula (4)jIs the average value of the jth parameter in the history storage state matrix, y (j) is
Figure FDA00024839000900000115
And
Figure FDA00024839000900000116
a normalized standard value;
the state vector in the history storage matrix is combined, as shown in formula (5):
Figure FDA0002483900090000021
w (k) in the formula (5) is a corresponding weightCoefficients representing the current observation vector
Figure FDA0002483900090000026
And the kth state vector xnc(k) A measure of the similarity between the two,
Figure FDA0002483900090000027
for the estimated value of the jth parameter, the similarity measure is calculated by using high-dimensional space mapping in the AAKR algorithm, and the kernel function inner product is introduced:
Figure FDA0002483900090000022
phi in equation (6) is the mapping function, taking the observation vector from RJThe space maps to a high-dimensional Euclidean space H,<>for the inner product operator, the traditional AAKR algorithm uses a gaussian kernel function to perform the calculation:
Figure FDA0002483900090000023
wherein h is the adjustable parameter bandwidth;
according to Mercer's kernel theory, the kernel function shown in equation (8) can be considered as the inner product of an infinite dimension Euclidean space:
Figure FDA0002483900090000024
the right side of the formula (8) is a mathematical expression form of an infinite dimension Euclidean space inner product;
and (3) calculating by adopting a Mahalanobis distance operator:
Figure FDA0002483900090000025
derived from
Figure FDA0002483900090000028
As an estimate in a plant state health indicator model
Figure FDA0002483900090000029
2. The power generation equipment state early warning method based on the auto-associative kernel regression algorithm according to claim 1, wherein the step 1 specifically comprises the following steps:
step 1.1: determining a main parameter by using artificial experience, equally dividing the main parameter by n, determining a vector nearest to a spacing point, and clustering data by taking n as a target class number, wherein n is a positive integer;
step 1.2: grouping the class center and the equidistant division results into a set;
step 1.3: sequencing the vector set to determine a boundary vector;
step 1.4: adding vectors according to the sequence by taking the boundary as a starting point;
step 1.5: judging the main parameter interval of the vector, if the main parameter interval is larger than n/2, adding the vector, and if the main parameter interval is smaller than n/2, deleting the vector;
step 1.6: a final historical storage state matrix is determined.
3. The power generation equipment state early warning method based on the auto-associative kernel regression algorithm as claimed in claim 1, wherein the adjustable parameter bandwidth h in the formula (7) is obtained by optimizing parameters in a training process through a cross-validation method, and specifically comprises a maximum value h of a given bandwidth hmaxAnd a minimum value hminH is to bemaxAnd hminThe method is divided into M equal parts, and for each h, the root mean square error under the 4-fold cross validation mode is calculated, wherein the calculation formula is as follows:
Figure FDA0002483900090000031
SSR in formula (10)hI.e. the root mean square error of the 4-fold cross validation,
Figure FDA0002483900090000034
the observed value of the jth parameter of the ith group of samples under the k modeling condition is shown, ts is the number of groups of training samples of each group after division,
Figure FDA0002483900090000035
for the estimated value of the jth variable of the ith group of samples under the condition of the kth modeling, the value formula of h in the change interval is as follows:
Figure FDA0002483900090000032
h in formula (11)mIs the bandwidth value of the mth time, M is the bandwidth in hmaxAnd hminThe number of the intervals divided in the range and the bandwidth coefficient h are from the minimum value hminTo a maximum value hmaxIn the process (2), there is an optimum coefficient hoptI.e. SSRhThe point with the minimum value is taken, the error of model estimation under the 4-fold cross validation learning mechanism is the minimum, namely the optimal bandwidth coefficient, and h is usedoptIs the value of the adjustable parameter bandwidth h in the formula (7).
4. The power generation equipment state early warning method based on the auto-associative kernel regression algorithm according to claim 1, wherein the step 3 specifically comprises: constructing equipment state health degree indexes through the standardized parameter estimation residuals in the step 2, and tracking, monitoring and early warning the equipment state, wherein the equipment state health degree indexes are as follows:
Figure FDA0002483900090000033
in the formula (12)
Figure FDA0002483900090000036
Is an observed value of the parameter(s),
Figure FDA0002483900090000037
is an estimated value of the parameter, and s is a health index.
CN202010385882.3A 2020-05-09 2020-05-09 Power generation equipment state early warning method based on auto-associative kernel regression algorithm Active CN111563685B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010385882.3A CN111563685B (en) 2020-05-09 2020-05-09 Power generation equipment state early warning method based on auto-associative kernel regression algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010385882.3A CN111563685B (en) 2020-05-09 2020-05-09 Power generation equipment state early warning method based on auto-associative kernel regression algorithm

Publications (2)

Publication Number Publication Date
CN111563685A CN111563685A (en) 2020-08-21
CN111563685B true CN111563685B (en) 2022-03-08

Family

ID=72070949

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010385882.3A Active CN111563685B (en) 2020-05-09 2020-05-09 Power generation equipment state early warning method based on auto-associative kernel regression algorithm

Country Status (1)

Country Link
CN (1) CN111563685B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113008583B (en) * 2020-11-10 2022-03-29 大连理工大学 Method and device for monitoring state and automatically alarming abnormality of rotating machine
CN114338070B (en) * 2021-09-03 2023-05-30 中国电子科技集团公司第三十研究所 Shadowsocks (R) identification method based on protocol attribute

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104102773B (en) * 2014-07-05 2017-06-06 山东鲁能软件技术有限公司 A kind of equipment fault early-warning and state monitoring method
CN106368816B (en) * 2016-10-27 2018-09-25 中国船舶工业系统工程研究院 A kind of online method for detecting abnormality of marine low speed diesel engine based on baseline offset

Also Published As

Publication number Publication date
CN111563685A (en) 2020-08-21

Similar Documents

Publication Publication Date Title
CN111237134B (en) Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model
US20220283576A1 (en) Automatic diagnosis method, system and storage medium for equipment
CN111709490B (en) Fan health state assessment method based on GRU neural network
KR101096793B1 (en) data collecting method for detection and on-time warning system of industrial process
CN109543210A (en) A kind of Wind turbines failure prediction system based on machine learning algorithm platform
CN103150473A (en) Method and device for monitoring and diagnosing generating efficiency of wind turbine generator in real-time manner
CN110529202B (en) Steam turbine fault diagnosis detection and early warning method and system based on power plant data
Li et al. A novel wind speed-sensing methodology for wind turbines based on digital twin technology
Siegel et al. An auto-associative residual processing and K-means clustering approach for anemometer health assessment
CN111563685B (en) Power generation equipment state early warning method based on auto-associative kernel regression algorithm
EP3553044A1 (en) System and method of remote object monitoring
CN110209144B (en) Two-layer real-time monitoring and alarm tracing method based on dynamic and static cooperative difference analysis
CN112597696B (en) Boiler four-pipe leakage early warning method based on extreme learning machine principle
CN112861350B (en) Temperature overheating defect early warning method for stator winding of water-cooled steam turbine generator
CN112598144A (en) CNN-LSTM burst fault early warning method based on correlation analysis
Chen et al. Prediction of icing fault of wind turbine blades based on deep learning
Li et al. A condition monitoring approach of multi-turbine based on VAR model at farm level
Wang et al. Wind turbine fault detection and identification through self-attention-based mechanism embedded with a multivariable query pattern
CN112418306A (en) Gas turbine compressor fault early warning method based on LSTM-SVM
CN111192163B (en) Generator reliability medium-short term prediction method based on wind turbine generator operating data
CN116677570A (en) Fault early warning method and system based on cabin temperature monitoring of offshore wind turbine
CN112576454A (en) Wind turbine generator main shaft temperature early warning method and device based on multi-dimensional early warning strategy
CN112308391A (en) Real-time monitoring and anomaly detection method for equipment state based on neural network
CN116151799A (en) BP neural network-based distribution line multi-working-condition fault rate rapid assessment method
Tutiv'en et al. Wind turbine main bearing condition monitoring via convolutional autoencoder neural networks

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