CN112036042A - Power equipment abnormality detection method and system based on variational modal decomposition - Google Patents

Power equipment abnormality detection method and system based on variational modal decomposition Download PDF

Info

Publication number
CN112036042A
CN112036042A CN202010908120.7A CN202010908120A CN112036042A CN 112036042 A CN112036042 A CN 112036042A CN 202010908120 A CN202010908120 A CN 202010908120A CN 112036042 A CN112036042 A CN 112036042A
Authority
CN
China
Prior art keywords
tested
power equipment
detected
data
entropy
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010908120.7A
Other languages
Chinese (zh)
Other versions
CN112036042B (en
Inventor
王航
彭敏俊
王晓昆
夏庚磊
夏虹
刘永阔
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.)
Harbin Engineering University
Original Assignee
Harbin Engineering 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 Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202010908120.7A priority Critical patent/CN112036042B/en
Publication of CN112036042A publication Critical patent/CN112036042A/en
Application granted granted Critical
Publication of CN112036042B publication Critical patent/CN112036042B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00Details relating to CAD techniques
    • G06F2111/04Constraint-based CAD
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00Details relating to CAD techniques
    • G06F2111/08Probabilistic or stochastic CAD

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a method and a system for detecting the abnormality of power equipment based on variational modal decomposition. The power equipment abnormality detection method based on variational modal decomposition comprises the following steps: acquiring operation data of power equipment to be tested; the operation data is non-stationary random data; carrying out variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested; performing phase space reconstruction on each intrinsic mode function to be measured to obtain a plurality of reconstruction characteristic parameters to be measured; calculating the permutation entropy and the sample entropy of each to-be-detected reconstruction characteristic parameter, and combining the permutation entropy of the to-be-detected reconstruction characteristic parameter and the sample entropy of the to-be-detected reconstruction characteristic parameter to obtain to-be-detected characteristics; and inputting the characteristics to be tested into the trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be tested. The invention can quickly detect the non-stable and non-linear early-stage tiny abnormity in time, and improve the detection accuracy and the detection efficiency.

Description

Power equipment abnormality detection method and system based on variational modal decomposition
Technical Field
The invention relates to the field of power equipment detection, in particular to a power equipment abnormity detection method and system based on variational modal decomposition.
Background
Nuclear power plants are complex in construction, and have a radioactive hazard, which places extremely high demands on safety. Meanwhile, the nuclear power system works continuously for a long time, and faults are easy to occur, if equipment fails and cannot be detected and found in time, serious radioactive release consequences can be caused, and public safety and environmental conditions are damaged. At present, most of the abnormal detection technologies for nuclear power systems and key equipment adopt traditional threshold analysis and manual experience for judgment. However, these conventional techniques cannot completely meet the reliability requirements of complex systems and key devices, and with the continuous development of artificial intelligence techniques and application experiences in other fields, the abnormal states of the operating parameters are detected in time by using efficient and accurate statistical analysis techniques, so that major harmful faults or even serious accidents can be avoided, the operation guarantee capability of the nuclear power system and the key devices can be effectively improved, the potential safety hazards are reduced, and autonomous guarantee is realized.
The state monitoring technology is characterized in that the meanings of the state monitoring technology and the fault diagnosis are mutually distinguished and related, the state monitoring technology is used for carrying out signal noise reduction and data processing on collected data, then displaying abnormal parameters and giving an alarm, the running states of a system and equipment are judged according to the abnormal parameters, and data and information are provided for fault analysis.
Internationally, researchers in the american atton national laboratory were the earliest to apply advanced information processing techniques to nuclear power plant condition monitoring. The northwest national laboratory of the pacific ocean developed a nuclear power plant diagnostic and monitoring system that could provide operational support to operators. The American West House company develops a rule-based expert system method, and can use data acquired by a nuclear power plant safety parameter display system as a nuclear power plant state monitoring, diagnosing and forecasting system of input data; bechel corporation, USA, developed a monitoring system that provided analysis of operating characteristics of a nuclear power plant. A great deal of research and development work has been carried out on the condition monitoring systems of nuclear power plants in japan, and a plurality of condition monitoring prototype systems have been successfully developed. The pentano system was developed by Fantoni, a hallen project in euro and his multinational research team and applied to nuclear power plant sensor monitoring.
In China, some basic research works are also carried out on the aspects of monitoring the states of nuclear power systems and equipment, and for example, a random analysis method in a time domain or a frequency domain is used for carrying out related research on the vibration monitoring of a reactor internals of a nuclear power plant. The frequency research of state monitoring on nuclear power plant equipment is carried out by the scenic nations of the Bay nuclear power plant, which plays a good role in improving the reliability of the equipment and prolonging the service life of the equipment. The yellow aspiration force of the Qinshan nuclear power plant proposes a means of taking key equipment as a monitoring object and monitoring the states of various types of equipment of perfecting an electromechanical instrument in a targeted manner.
When the equipment state is monitored, sensors such as acceleration, speed and acoustic measurement are adopted, signals of the sensors have non-stationary and high-frequency characteristics, and therefore the signals need to be processed and denoised firstly. Traditional feature extraction methods such as short-time fourier transform, power spectrum analysis, etc. have been proven to be unsuitable for continuous non-stationary random signals. In recent years, wavelet-wavelet packet transformation, empirical mode decomposition and integrated empirical mode decomposition are proposed in sequence, but the empirical mode decomposition easily causes a mode aliasing phenomenon, and the wavelet decomposition may bring harmonic waves which are not available in the original physical quantity, so that errors are increased.
With the increase of the characteristic parameters, the monitoring data contains a large amount of redundant information, which affects the accurate judgment of the system running state. The emergence and development of data dimension reduction technology provides a theoretical basis for data-driven anomaly detection. Data dimension reduction and feature extraction methods are mainly divided into a multivariate statistical method, a kernel method and a neural network method. The multivariate statistical method mainly comprises the following steps: principal component analysis methods, independent principal component analysis, and the like. The above methods are all linear processing methods, and therefore, nonlinear information in data cannot be acquired in many cases. The more mature of the nuclear methods is the nuclear principal component analysis method. The artificial neural network method can meet the nonlinear mapping condition of high-dimensional and low-dimensional information space conversion, but the problem of 'over-learning' may occur.
At present, few intelligent state monitoring researches for nuclear power typical equipment are provided, and the state monitoring technology can be expected to have wide application prospects by combining the development levels of the current domestic and foreign power systems and equipment state monitoring technologies, and the development trend of the state monitoring technology mainly has the following aspects: i. the development is from simple and single monitoring to precise and comprehensive monitoring; the reliability and the safety are gradually improved in performance; the development is towards high flexibility and high intelligence on the technical level; the system structure is developed to be standardized, generalized and modularized; v. new methods and technologies such as intelligent methods and information fusion are continuously introduced. The existing power equipment abnormality detection method cannot timely and quickly detect non-stable and non-linear early-stage small abnormality, and the accuracy and the efficiency are to be improved.
Disclosure of Invention
Based on the above, it is necessary to provide a method and a system for detecting abnormality of power equipment based on variational modal decomposition, and to detect non-stationary and non-linear early minute abnormality in a timely and rapid manner, so as to improve the accuracy and efficiency of detection.
In order to achieve the purpose, the invention provides the following scheme:
a power equipment abnormality detection method based on variational modal decomposition comprises the following steps:
acquiring operation data of power equipment to be tested; the operating data is non-stationary random data;
carrying out variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested;
performing phase space reconstruction on each intrinsic mode function to be measured to obtain a plurality of reconstruction characteristic parameters to be measured;
calculating the permutation entropy and the sample entropy of each to-be-detected reconstruction characteristic parameter, and combining the permutation entropy of the to-be-detected reconstruction characteristic parameter and the sample entropy of the to-be-detected reconstruction characteristic parameter to obtain to-be-detected characteristics;
inputting the characteristics to be tested into a trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be tested; the trained convolution self-encoder is formed by taking the running data of the power equipment with known running state as input and taking the corresponding running state as output training; the operating state includes a normal state and a fault state.
Optionally, the performing variational modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic mode functions to be tested specifically includes:
carrying out variation modal decomposition on the operation data of the power equipment to be tested, and constructing a constrained variation model;
introducing a penalty parameter and a penalty factor into the constrained variation model, and determining the optimal solution of the constrained variation model according to a Lagrangian function and a Fourier transform method; the optimal solution comprises an optimal mode function to be tested and a central frequency corresponding to the optimal mode function to be tested;
and determining the modal component to be tested according to the optimal solution, screening the modal component to be tested, and determining the intrinsic modal function to be tested.
Optionally, the constrained variational model is
Figure BDA0002662266390000031
Figure BDA0002662266390000032
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure BDA0002662266390000033
in order to make the derivation of the symbol,
Figure BDA0002662266390000034
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
Optionally, the calculating the permutation entropy and the sample entropy of each to-be-detected reconstructed feature parameter, and combining the permutation entropy of the to-be-detected reconstructed feature parameter and the sample entropy of the to-be-detected reconstructed feature parameter to obtain the to-be-detected feature specifically includes:
arranging all elements corresponding to each vector in the to-be-detected reconstruction characteristic parameters in an ascending order to obtain a plurality of time sequences;
calculating the occurrence probability of each time sequence;
obtaining the permutation entropy of the characteristic parameters to be reconstructed based on the time sequence;
calculating the distance between the maximum element values corresponding to any two vectors in the reconstruction characteristic parameters to be detected;
counting the number of distances greater than the similar tolerance;
determining the sample entropy of the reconstruction characteristic parameter to be detected based on the number;
and combining the permutation entropy of the to-be-detected reconstruction characteristic parameters and the sample entropy of the to-be-detected reconstruction characteristic parameters to obtain to-be-detected characteristics.
Optionally, the method for determining the trained convolutional self-encoder includes:
acquiring running data of power equipment with a known running state and a corresponding running state to obtain training data;
constructing a convolutional self-encoder model; the convolutional self-encoder model comprises a convolutional encoding layer and a convolutional decoding layer; the convolution coding layer comprises a plurality of convolution layers and a plurality of pooling layers; the convolution decoding layer comprises a plurality of layers of up-sampling layers and a plurality of layers of deconvolution layers;
taking the running data in the training data as input, taking the running state in the training data as output, taking a cross entropy function as a loss function, and training the convolution self-encoder model by adopting a random gradient descent method to obtain an initial convolution self-encoder;
and adjusting the initial convolution self-encoder by adopting an N-fold cross verification method to obtain the trained convolution self-encoder.
The invention also provides a power equipment abnormality detection system based on variational modal decomposition, which comprises:
the data acquisition module is used for acquiring the operating data of the power equipment to be tested; the operating data is non-stationary random data;
the variation modal decomposition module is used for performing variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested;
the spatial reconstruction module is used for carrying out phase space reconstruction on each intrinsic mode function to be detected to obtain a plurality of reconstruction characteristic parameters to be detected;
the information entropy calculation module is used for calculating the permutation entropy and the sample entropy of each to-be-detected reconstruction characteristic parameter, and combining the permutation entropy of the to-be-detected reconstruction characteristic parameter and the sample entropy of the to-be-detected reconstruction characteristic parameter to obtain to-be-detected characteristics;
the anomaly detection module is used for inputting the characteristics to be detected into the trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be detected; the trained convolution self-encoder is formed by taking the running data of the power equipment with known running state as input and taking the corresponding running state as output training; the operating state includes a normal state and a fault state.
Optionally, the variational modal decomposition module specifically includes:
the variational model construction unit is used for carrying out variational modal decomposition on the operation data of the power equipment to be tested and constructing a constrained variational model;
the solving unit is used for introducing penalty parameters and penalty factors into the constrained variation model and determining the optimal solution of the constrained variation model according to a Lagrangian function and a Fourier transform method; the optimal solution comprises an optimal mode function to be tested and a central frequency corresponding to the optimal mode function to be tested;
and the screening unit is used for determining the modal component to be tested according to the optimal solution, screening the modal component to be tested and determining the intrinsic modal function to be tested.
Optionally, the constrained variation model in the variation model building unit is
Figure BDA0002662266390000051
Figure BDA0002662266390000052
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure BDA0002662266390000053
in order to make the derivation of the symbol,
Figure BDA0002662266390000054
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
Optionally, the information entropy calculating module specifically includes:
the sequencing unit is used for sequencing all elements corresponding to each vector in the to-be-tested reconstruction characteristic parameters in an ascending order to obtain a plurality of time sequences;
a probability calculation unit for calculating the occurrence probability of each of the time series;
the permutation entropy calculation unit is used for obtaining the permutation entropy of the characteristic parameters to be reconstructed based on the time sequence;
the distance calculation unit is used for calculating the distance between the maximum element values corresponding to any two vectors in the to-be-detected reconstruction characteristic parameters;
the statistical unit is used for counting the number of the distances greater than the similar tolerance;
the sample entropy calculation unit is used for determining the sample entropy of the reconstruction characteristic parameter to be measured based on the number;
and the merging unit is used for merging the permutation entropy of the to-be-detected reconstruction characteristic parameters and the sample entropy of the to-be-detected reconstruction characteristic parameters to obtain the to-be-detected characteristics.
Optionally, the power equipment abnormality detection system based on the variational modal decomposition further includes a convolution self-encoder determining module;
the convolutional auto-encoder determining module includes:
the training data acquisition unit is used for acquiring the running data of the power equipment with known running state and the corresponding running state to obtain training data;
the model building unit is used for building a convolutional self-encoder model; the convolutional self-encoder model comprises a convolutional encoding layer and a convolutional decoding layer; the convolution coding layer comprises a plurality of convolution layers and a plurality of pooling layers; the convolution decoding layer comprises a plurality of layers of up-sampling layers and a plurality of layers of deconvolution layers;
the training unit is used for training the convolution self-encoder model by taking the running data in the training data as input, taking the running state in the training data as output and taking a cross entropy function as a loss function and adopting a random gradient descent method to obtain an initial convolution self-encoder;
and the verification unit is used for adjusting the initial convolution self-encoder by adopting an N-fold cross verification method to obtain the trained convolution self-encoder.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a method and a system for detecting the abnormality of power equipment based on variational modal decomposition, wherein the variational modal decomposition is adopted to process non-stable nonlinear data, so that the problems of modal aliasing and end effect can be effectively solved; potential features are further deeply mined from each order of intrinsic mode functions obtained by variable mode decomposition, algorithm defects of a single information entropy algorithm in feature extraction can be avoided through calculation of permutation entropy and sample entropy, and therefore time sequence dynamic information hidden in the intrinsic mode functions is guaranteed to be extracted as comprehensively as possible; the method is characterized in that unsupervised feature extraction and data dimension reduction are carried out by adopting a trained convolution self-encoder, and the abnormal detection problem is abstracted into a 'two-classification' problem. The invention can quickly detect the non-stable and non-linear early-stage tiny abnormity in time, and improve the detection accuracy and the detection efficiency.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described 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 to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a method for detecting an anomaly in a power plant based on variational modal decomposition according to an embodiment of the present invention;
FIG. 2 is a diagram of a specific implementation process of a method for detecting an anomaly of a power plant based on variational modal decomposition according to an embodiment of the present invention;
FIG. 3 is a layout diagram of an acceleration sensor on a feedwater pump provided in accordance with an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a power equipment abnormality detection system based on variational modal decomposition according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the 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.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a flowchart of a power equipment abnormality detection method based on variational modal decomposition according to an embodiment of the present invention.
Referring to fig. 1, the method for detecting an abnormality of a power plant based on a variational modal decomposition of the embodiment includes:
step 101: acquiring operation data of power equipment to be tested; the operating data is non-stationary random data.
Step 102: and carrying out variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested.
The step 102 specifically includes:
carrying out variation modal decomposition on the operation data of the power equipment to be tested, and constructing a constrained variation model
Figure BDA0002662266390000081
Figure BDA0002662266390000082
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure BDA0002662266390000083
in order to make the derivation of the symbol,
Figure BDA0002662266390000084
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
Introducing a penalty parameter alpha and a penalty factor lambda into the constrained variation model, and determining the optimal solution of the constrained variation model according to a Lagrangian function and a Fourier transform method; the optimal solution comprises an optimal mode function to be tested and a center frequency corresponding to the optimal mode function to be tested. The optimal solution is as follows:
Figure BDA0002662266390000085
Figure BDA0002662266390000086
Figure BDA0002662266390000087
is the optimal modal function to be measured; mu.si(w) is the ith basis function, i is the serial number of the intrinsic mode to be measured; λ (w) is the time step of the dual rise; w is the frequency; w is akIs the center frequency of each basis function; mu.sk(w) is the kth basis function;
Figure BDA0002662266390000088
and the center frequency corresponding to the optimal mode function to be measured.
And determining the modal component to be tested according to the optimal solution, screening the modal component to be tested, and determining the intrinsic modal function to be tested.
Step 103: and performing phase space reconstruction on each intrinsic mode function to be measured to obtain a plurality of reconstruction characteristic parameters to be measured.
Step 104: and calculating the permutation entropy and the sample entropy of each to-be-detected reconstruction characteristic parameter, and combining the permutation entropy of the to-be-detected reconstruction characteristic parameter and the sample entropy of the to-be-detected reconstruction characteristic parameter to obtain to-be-detected characteristics.
The step 104 specifically includes:
arranging all elements corresponding to each vector in the to-be-detected reconstruction characteristic parameters in an ascending order to obtain a plurality of time sequences; calculating the occurrence probability of each time sequence; obtaining the permutation entropy of the characteristic parameters to be reconstructed based on the time sequence; calculating the distance between the maximum element values corresponding to any two vectors in the reconstruction characteristic parameters to be detected; counting the number of distances greater than the similar tolerance; determining the sample entropy of the reconstruction characteristic parameter to be detected based on the number; and combining the permutation entropy of the to-be-detected reconstruction characteristic parameters and the sample entropy of the to-be-detected reconstruction characteristic parameters to obtain to-be-detected characteristics.
The permutation entropy
Figure BDA0002662266390000091
Wherein, PiIs the probability of occurrence of the ith time series, I1, 2. I is the total number of time series, and m is the embedding dimension.
The sample entropy
Figure BDA0002662266390000092
Wherein m is an embedding dimension, r is a similar tolerance, and N is the total number (total time sequence length) of vectors in the reconstructed characteristic parameter to be detected; b ism(r) is the average of the ratios of the number of distances greater than the similarity tolerance to the total number of vectors when the embedding dimension is m; b ism+1(r) is the average of the ratios of the number of distances greater than the similarity tolerance to the total number of vectors for the embedding dimension m + 1. When the N is limited, the N is,
Figure BDA0002662266390000093
step 105: and inputting the characteristics to be tested into the trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be tested.
The trained convolution self-encoder is formed by taking the running data of the power equipment with known running state as input and taking the corresponding running state as output training; the operating state includes a normal state and a fault state. The method for determining the trained convolutional self-encoder comprises the following steps:
and acquiring the running data of the power equipment with known running state and the corresponding running state to obtain training data.
Constructing a convolutional self-encoder model; the convolutional self-encoder model comprises a convolutional encoding layer and a convolutional decoding layer; the convolution coding layer comprises a plurality of convolution layers and a plurality of pooling layers; the convolution decoding layer comprises a plurality of layers of up-sampling layers and a plurality of layers of deconvolution layers.
And training the convolution self-encoder model by using a random gradient descent method by using the running data in the training data as input, the running state in the training data as output and the cross entropy function as a loss function to obtain an initial convolution self-encoder.
And adjusting the initial convolution self-encoder by adopting an N-fold cross verification method to obtain the trained convolution self-encoder.
The following gives a concrete implementation procedure in practical application. As shown in fig. 2, includes:
step 1: fault experimental and test data of typical nuclear power equipment are collected and stored. As shown in fig. 3, the water supply pump in the nuclear power system is taken as an example for explanation in the present embodiment, a non-stationary random signal during the operation of the water supply pump is acquired and obtained through the speed sensor arranged in the figure, and then the data is stored in the computer through the data acquisition board card. As shown in fig. 3, the measuring point 1 and the measuring point 2 are located at the bearing close to the coupling end to monitor the vibration signal of the bearing; the measuring point 3 and the measuring point 4 are arranged at positions close to the bearing and used for monitoring the vibration of the impeller; measuring point 5 and measuring point 7 are used for monitoring the vibration of the pump body, and measuring point 6 is used for monitoring the vibration of the base.
Step 2: and (3) managing the data under different working conditions and different fault states in the step (1) in a classified manner through a computer, and adding a state label to all the data in the step (1) to indicate whether the data are normal or fault, so that the subsequent abnormal detection and state monitoring are facilitated.
And step 3: and (3) carrying out variation modal decomposition on the non-stationary parameter obtained by measurement in the step (1). In the variation modal decomposition, a signal is assumed to be formed by overlapping a plurality of modal functions, and each modal function can be regarded as an AM frequency modulation signal with different center frequencies. Then, determining the frequency center and the bandwidth of each modal function by iteratively searching the extreme value of the constructed variation model to realize the effective separation of each component, and constructing the constrained variation model as
Figure BDA0002662266390000101
Figure BDA0002662266390000102
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure BDA0002662266390000111
in order to make the derivation of the symbol,
Figure BDA0002662266390000112
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
And 4, step 4: and (3) further introducing a penalty parameter alpha and a penalty factor lambda in the formula of the step (3) and establishing a Lagrange function to solve the optimal solution of the constrained variation model. Formula by Fourier transform
Figure BDA0002662266390000113
Conversion from time domain to frequency domain, initialization
Figure BDA0002662266390000114
λ1And n is 0.
And 5: u. ofkAnd ωkThe iterative update is performed by the following formulas, respectively.
Figure BDA0002662266390000115
Figure BDA0002662266390000116
Step 6: according to the formula
Figure BDA0002662266390000117
And updating the lambda.
And 7: and repeating the step 5 and the step 6 until an iteration termination condition is met:
Figure BDA0002662266390000118
wherein, for discrimination accuracy, > 0.
And 8: and outputting a calculation result of the variation modal decomposition to obtain K modal components.
And step 9: analyzing and comparing the values of the central frequencies of the K modal components, mainly observing whether the central frequency values are close to each other, if the numerical values between the two central frequencies are in the same order of magnitude, trying to reduce the number of K, and then repeating the calculation of the steps 3-8 until all the central frequencies are not in the same order of magnitude, thereby avoiding the problem of poor characteristic extraction effect caused by modal aliasing and finally obtaining the intrinsic modal function.
Step 10: and (3) repeating the steps 3-9 respectively for all other non-stationary random signals obtained by measurement in the step 1 to perform variation modal decomposition, and respectively obtaining respective K modal components so as to obtain respective intrinsic modal functions.
Step 11: after the intrinsic mode functions of all the measuring points are obtained through variation mode decomposition screening, in order to further mine the characteristic information and perform abnormality detection, the arrangement entropy and the sample entropy of the time sequence are respectively calculated based on the information entropy principle.
Step 12: the first calculation to be performed using permutation entropy allows detection of abrupt changes in the random signal, which is simple and fast since it is mainly a comparison of adjacent values without regard to the size of the values. The permutation entropy has proven to be very effective in the presence of dynamic or observation noise.
Step 13: performing phase space reconstruction on all the order eigenmode functions obtained in step 10, assuming that each eigenmode function can be represented as a time sequence { X (i) }, i ═ 1,2,. and N }, performing phase space reconstruction on the eigenmode functions to obtain a plurality of vectors X (1), X (2), …, and X (N- (m-1) τ) of the reconstruction characteristic parameters to be measured; where x (i) { x (i), x (i + τ),. -, x (i + (m-1) τ) }, i ═ 1, 2. -, N- (m-1) τ, m is the embedding dimension, τ is the time delay factor. In the specific implementation, m is 30, and the time delay factor is 1.
Step 14, rearranging the elements in the vector X (i) in each to-be-detected reconstruction characteristic parameter obtained in the step 13 according to an ascending order: thus, any vector x (i) can obtain a symbol sequence z (i) ═ l1,l2,...,lm]Wherein I is 1,2, I ≦ m! . m different symbols l1,l2,...,lm]Total m! Different arrangements correspond to a total of m! A different symbol sequence, calculating the probability of each permutation in the time sequence, and expressing as Pi(i=1,2,...,I)。
Step 15: according to the formula
Figure BDA0002662266390000121
And obtaining the arrangement entropy value of each coarse graining intrinsic mode function sequence.
Step 16: and extracting features in the intrinsic mode functions of all orders by adopting sample entropy. The sample entropy is an improvement based on approximate entropy but is higher in precision, and is a strict natural logarithm of conditional probability, so that the calculation of the sample entropy is independent of the length of data per se and is more closely consistent with a known random part. Based on the phase space reconstructed data in step 13, X (i)1),X(i2) The maximum difference of the corresponding elements is defined as the distance between the two, i.e.:
Figure BDA0002662266390000122
wherein k is 1,2, …, m, and i ≠ j, X (i)1) Is the ith1The characteristic parameter to be reconstructed corresponding to the intrinsic mode function X (i)2) Is the ith2And the characteristic parameters to be reconstructed correspond to the intrinsic mode functions.
And step 17: based on step 16, a similarity margin r is selected, and a statistical distance d [ X (i)1),X(i2)]< r number, and noted BiAnd calculating the ratio of the sum to the total number of the vectors N-m +1, and recording the ratio as
Figure BDA0002662266390000131
Then calculate N-m +1
Figure BDA0002662266390000132
Average value of (1), denoted as Bm(r), where N is the total timing length.
Step 18: adding 1 to the embedding dimension m to form a group of m + 1-dimensional vectors, and repeating the step 17 to obtain Bm+1(r) of (A). Thus, the sample entropy of this sequence can be expressed as
Figure BDA0002662266390000133
Thus, when N is limited, the estimate of the sequence sample entropy is:
Figure BDA0002662266390000134
step 19: after the permutation entropy and the sample entropy time sequence parameters of the intrinsic mode functions of each order are obtained, the permutation entropy and the sample entropy time sequence parameters are combined and combined to form comprehensive characteristics, the Number of the characteristics can be represented as Number multiplied by K multiplied by 2, wherein the Number is the Number of sensors, and K is the Number of the intrinsic mode functions. Because the feature dimension is too large, if a statistical threshold is set for each parameter, the parameter is too large, which may adversely affect subsequent anomaly detection and failure analysis. Therefore, the embodiment adopts the convolution noise reduction self-encoder to deeply extract the intrinsic characteristics of the data, realizes the effects of dimension reduction and characteristic extraction, and then converts the anomaly detection problem into the 'two-classification' problem so as to realize the anomaly detection of the nuclear power key equipment.
Step 20: converting the comprehensive characteristics (two-dimensional input data of dimension N × D) obtained in step 19 into three-dimensional stacked data blocks of dimension (N-num _ steps +1) × (num _ steps × D), wherein N is the total time sequence length, D is the dimension of the characteristic parameter, num _ steps is the time sequence number of elements in the training input, and the total data input length is (N-num _ steps +1) because of the overlapping of data in each sliding process. Therefore, the input data at each moment are not isolated characteristic parameters of a certain moment but are combined in a period of time, and the subsequent calculation by adopting the convolution noise reduction self-coding is facilitated.
Step 21: and carrying out unsupervised nonlinear feature extraction according to the convolutional denoising self-coding principle. And constructing multilayer convolution and pooling in the encoding process and multilayer deconvolution and upsampling in the decoding process through a Tensorflow framework to form a deep abstract feature with invariable time sequence. The resulting high-level data feature may be denoted as CijWhere i represents the time series data length and j represents the dimension of the characteristic parameter.
Step 22: the use of dropout operations for each convolution and deconvolution process on the basis of step 21 makes the self-encoding network more "robust".
Step 23: all the activation functions involved in the convolutional self-encoder are adjusted to be Leaky ReLU, so that dead nodes can be avoided on the basis of the ReLU activation functions, and the nonlinear characteristics in data can be reflected.
Step 24: training a convolutional self-encoder model; in the training process of the convolution self-encoder model, in order to improve the training speed and efficiency, the normal data and the abnormal data obtained in the step 2 are divided into two types and used as training and testing data of the convolution noise reduction self-encoder.
Step 25: defining a loss function and optimizing parameters; the invention uses a cross entropy function as a loss function. In order to optimize the weights and biases in the convolutional self-encoder and the long-term memory network, an SGD optimization algorithm is adopted to solve the network in the training process, so that the loss function value is as small as possible, and finally, the network structure parameters which best meet the abnormal detection of the nuclear power key equipment are obtained. In the calculation process of each back propagation, the learning rate of the first 5 iterations is set to be 0.001, the learning rate is not attenuated, and the attenuation rate of the learning rate of each subsequent iteration is set to be 0.99. With the increase of the number of training rounds and the reduction of the training error, the convolutional self-encoder model can continuously approximate to the actual characteristic.
Step 26: and adopting N-fold cross validation, using part of data as training data, and using other data as validation data to test the accuracy of the model. And (5) carrying out repeated iteration and optimization on the hyper-parameters in the step (25) to obtain a convolutional self-encoder model with the highest accuracy.
Step 27: in the process of detecting the abnormality of the actual nuclear power equipment, the abnormal non-stationary data is subjected to variation modal decomposition according to the modes of the steps 1, 3-10 to obtain intrinsic modal functions of each order of each measuring point.
Step 28: and obtaining real-time permutation entropy characteristic parameters for the data obtained in the step 27 according to a mode of 12-15.
Step 29: and obtaining real-time sample entropy characteristic parameters for the data obtained in the step 27 according to the mode of the steps 16-18.
Step 30: inputting the data into the trained convolutional noise reduction self-coding network according to the data format required by the steps 19-26, and if the calculation result shows that the equipment is in normal operation, continuously monitoring; if the classification result is abnormal operation, the operation parameters are abnormal, the system has faults, and the fault diagnosis needs to be carried out in time.
The method adopts the confusion matrix and the fault diagnosis accuracy as indexes to evaluate the accuracy and the effectiveness of the method. The related results can be referred by operation and decision-making personnel, and related measures can be taken in time, so that the safety is ensured, and the economy can be improved.
And 3-10, performing signal processing on the measurement data of the high-frequency, non-stable and non-linear vibration sensor and the like on the nuclear power equipment through the variation modal decomposition. The method is mainly characterized in that the traditional time domain analysis and frequency domain analysis methods cannot completely mine the characteristic information contained in the dynamic time sequence data. Therefore, a time-frequency domain analysis method is necessary to process non-stationary random data and comprehensively extract fault characteristics. Wavelet transform and wavelet packet transform are the most mature signal processing algorithms in theory, but have the problems that proper base wavelets are difficult to determine and the number of decomposition layers is difficult to set reasonably; empirical mode decomposition and its variants can adaptively extract the intrinsic mode function of the original data but has the problem of mode aliasing and the theoretical basis is not perfect. Based on the above problems, the present embodiment adopts variational mode decomposition to process non-stationary nonlinear data, and determines the number of mode components by combining the aliasing of the center frequency, so as to effectively solve the problems of mode aliasing and endpoint effect.
And 11, potential features are further deeply mined by reasonably designing the intrinsic mode functions of each order obtained by the variation mode decomposition, and algorithm defects of a single information entropy algorithm in feature extraction can be avoided by calculating the permutation entropy and the sample entropy, so that the time sequence dynamic information hidden in the intrinsic mode functions can be comprehensively extracted as far as possible.
And 12-15, processing by a phase space reconstruction technology according to the principle of the arrangement entropy, representing the time series dynamics mutation of the intrinsic mode function obtained after the variation mode decomposition processing, performing coarse graining processing on the time series, and then calculating the arrangement entropy value under the set scale. In order to ensure the real-time performance of feature extraction, the embodiment adopts the maximum overlapping moving window method for selection, and coarse graining processing and continuous permutation entropy calculation can be performed on the time series in the moving window.
And 16-18, obtaining the sample entropy of each order of intrinsic mode function, wherein the sample entropy is one of the information entropies, and the information entropy is the uncertainty proposed by Shannon to represent the information. The physical meaning of the sample entropy is similar to the approximate entropy. The lower the sample entropy, the higher the sequence self-similarity, the larger the sample entropy, the more complex the sequence. The sample entropy is an improved method based on approximate entropy and has higher precision, not only has the excellent characteristics of the approximate entropy, but also does not need to be matched with a self template. Compared with approximate entropy, the sample entropy is strict natural logarithm of conditional probability, so that the calculation of the sample entropy is independent of the length of data per se and has closer consistency with known random parts.
Step 20 converts the original dimension data obtained from the preamble into a three-dimensional data set, which can be consistent with the dimension of the input data of the convolutional self-coding network, and can also further characterize the permutation entropy and the sample entropy into a segment of time series data, instead of giving a judgment only by the value of one permutation entropy or sample entropy, so that more attention is paid to a segment of time series, and more data characteristics after an abnormality occurs can be reflected.
21-26, adopting a convolution noise reduction self-encoder to perform unsupervised feature extraction and data dimension reduction, and abstracting an abnormal detection problem into a 'second classification' problem; then, the dropout operation is adopted, so that overfitting of the self-coding network can be prevented; by introducing the Leaky ReLU activation function, dead nodes can be avoided on the basis of the ReLU activation function, and the sparse model can better fit nonlinear features in data. In the calculation process of each back propagation, the learning rate of the first 5 iterations is set to be 0.001, the learning rate is not attenuated, and the attenuation rate of the learning rate of each subsequent iteration is set to be 0.99. Through the change of the learning rate, the most appropriate weight and bias can be found more accurately in the back propagation calculation process, and finally the accuracy of the model is improved.
In addition, the method can also carry out the feature extraction of non-stationary signals and the signal noise reduction by methods such as short-time Fourier transform, wavelet transform, empirical mode decomposition and the like; wavelet transform and wavelet packet transform are the most mature signal processing algorithms in theory, but have the problems that proper base wavelets are difficult to determine and the decomposition layer number is difficult to set reasonably; empirical mode decomposition and its variants can adaptively extract the intrinsic mode function of the original data but has the problem of mode aliasing and the theoretical basis is not perfect. Based on the above problems, the present embodiment adopts the variational mode decomposition to process the non-stationary nonlinear data, and determines the number of the mode components by combining the indexes such as the aliasing property of the center frequency, so as to effectively solve the problems of the mode aliasing and the endpoint effect. Although the variational modal decomposition achieves the noise reduction effect of original data and restores an intrinsic modal function, potential features still need to be deeply mined on the basis, and the permutation entropy is an average entropy parameter used for measuring the complexity of a one-dimensional time sequence, and the dynamic sudden change of a time signal is detected by comparing the sizes of adjacent data. Meanwhile, in order to avoid the problem of insufficient feature extraction of the permutation entropy features under certain special conditions, the sample entropy is adopted to further extract the features of the data subjected to the variation modal decomposition, and then the sample entropy and the permutation entropy under the same phase space are spliced to form a mixed feature, so that the expressive force of the data features is improved. And finally, a convolution noise reduction self-encoder is adopted to carry out data dimension reduction and anomaly detection, and compared with methods such as kernel principal component analysis and independent component analysis, the method can provide an anomaly detection result of the nuclear power equipment with relatively higher accuracy.
The invention also provides a power equipment abnormality detection system based on the variational modal decomposition, and fig. 4 is a schematic structural diagram of the power equipment abnormality detection system based on the variational modal decomposition provided by the embodiment of the invention. Referring to fig. 4, the power equipment abnormality detection system based on the variational modal decomposition of the embodiment includes:
the data acquisition module 201 is used for acquiring the operation data of the power equipment to be tested; the operating data is non-stationary random data.
And the variation modal decomposition module 202 is configured to perform variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested.
And the spatial reconstruction module 203 is configured to perform phase-space reconstruction on each intrinsic mode function to be measured to obtain a plurality of reconstruction characteristic parameters to be measured.
The information entropy calculation module 204 is configured to calculate an arrangement entropy and a sample entropy of each to-be-detected reconstructed feature parameter, and combine the arrangement entropy of the to-be-detected reconstructed feature parameter and the sample entropy of the to-be-detected reconstructed feature parameter to obtain the to-be-detected feature.
The anomaly detection module 205 is configured to input the feature to be detected into a trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be detected; the trained convolution self-encoder is formed by taking the running data of the power equipment with known running state as input and taking the corresponding running state as output training; the operating state includes a normal state and a fault state.
As an optional implementation manner, the variational modal decomposition module 202 specifically includes:
and the variation model construction unit is used for carrying out variation modal decomposition on the operation data of the power equipment to be tested and constructing a constrained variation model.
The solving unit is used for introducing penalty parameters and penalty factors into the constrained variation model and determining the optimal solution of the constrained variation model according to a Lagrangian function and a Fourier transform method; the optimal solution comprises an optimal mode function to be tested and a center frequency corresponding to the optimal mode function to be tested.
And the screening unit is used for determining the modal component to be tested according to the optimal solution, screening the modal component to be tested and determining the intrinsic modal function to be tested.
As an alternative embodiment, the constrained variation model in the variation model building unit is
Figure BDA0002662266390000171
Figure BDA0002662266390000172
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure BDA0002662266390000173
in order to make the derivation of the symbol,
Figure BDA0002662266390000174
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
As an optional implementation manner, the information entropy calculation module 204 specifically includes:
and the sequencing unit is used for sequencing all elements corresponding to each vector in the to-be-detected reconstruction characteristic parameters in an ascending order to obtain a plurality of time sequences.
And the probability calculation unit is used for calculating the occurrence probability of each time sequence.
And the permutation entropy calculation unit is used for obtaining the permutation entropy of the characteristic parameters to be reconstructed based on the time sequence.
And the distance calculation unit is used for calculating the distance between the maximum element values corresponding to any two vectors in the to-be-detected reconstruction characteristic parameters.
And the statistical unit is used for counting the number of the distances greater than the similarity tolerance.
And the sample entropy calculation unit is used for determining the sample entropy of the reconstruction characteristic parameter to be detected based on the number.
And the merging unit is used for merging the permutation entropy of the to-be-detected reconstruction characteristic parameters and the sample entropy of the to-be-detected reconstruction characteristic parameters to obtain the to-be-detected characteristics.
As an optional implementation mode, the power equipment abnormality detection system based on the variational modal decomposition further comprises a convolution self-encoder determination module. The convolutional auto-encoder determining module includes:
and the training data acquisition unit is used for acquiring the running data of the power equipment with known running state and the corresponding running state to obtain the training data.
The model building unit is used for building a convolutional self-encoder model; the convolutional self-encoder model comprises a convolutional encoding layer and a convolutional decoding layer; the convolution coding layer comprises a plurality of convolution layers and a plurality of pooling layers; the convolution decoding layer comprises a plurality of layers of up-sampling layers and a plurality of layers of deconvolution layers.
And the training unit is used for training the convolution self-encoder model by taking the running data in the training data as input, taking the running state in the training data as output and taking the cross entropy function as a loss function and adopting a random gradient descent method to obtain the initial convolution self-encoder.
And the verification unit is used for adjusting the initial convolution self-encoder by adopting an N-fold cross verification method to obtain the trained convolution self-encoder.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (10)

1. A power equipment abnormality detection method based on variational modal decomposition is characterized by comprising the following steps:
acquiring operation data of power equipment to be tested; the operating data is non-stationary random data;
carrying out variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested;
performing phase space reconstruction on each intrinsic mode function to be measured to obtain a plurality of reconstruction characteristic parameters to be measured;
calculating the permutation entropy and the sample entropy of each to-be-detected reconstruction characteristic parameter, and combining the permutation entropy of the to-be-detected reconstruction characteristic parameter and the sample entropy of the to-be-detected reconstruction characteristic parameter to obtain to-be-detected characteristics;
inputting the characteristics to be tested into a trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be tested; the trained convolution self-encoder is formed by taking the running data of the power equipment with known running state as input and taking the corresponding running state as output training; the operating state includes a normal state and a fault state.
2. The method according to claim 1, wherein the power equipment abnormality detection method based on the variational modal decomposition is characterized in that the variational modal decomposition is performed on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested, and specifically comprises:
carrying out variation modal decomposition on the operation data of the power equipment to be tested, and constructing a constrained variation model;
introducing a penalty parameter and a penalty factor into the constrained variation model, and determining the optimal solution of the constrained variation model according to a Lagrangian function and a Fourier transform method; the optimal solution comprises an optimal mode function to be tested and a central frequency corresponding to the optimal mode function to be tested;
and determining the modal component to be tested according to the optimal solution, screening the modal component to be tested, and determining the intrinsic modal function to be tested.
3. The method for detecting the abnormality of the power equipment based on the variational modal decomposition according to claim 2, wherein the constrained variational model is
Figure FDA0002662266380000011
Figure FDA0002662266380000012
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure FDA0002662266380000021
in order to make the derivation of the symbol,
Figure FDA0002662266380000022
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
4. The method according to claim 1, wherein the step of calculating the permutation entropy and the sample entropy of each to-be-detected reconstructed feature parameter and combining the permutation entropy of the to-be-detected reconstructed feature parameter and the sample entropy of the to-be-detected reconstructed feature parameter to obtain the to-be-detected feature specifically comprises the steps of:
arranging all elements corresponding to each vector in the to-be-detected reconstruction characteristic parameters in an ascending order to obtain a plurality of time sequences;
calculating the occurrence probability of each time sequence;
obtaining the permutation entropy of the characteristic parameters to be reconstructed based on the time sequence;
calculating the distance between the maximum element values corresponding to any two vectors in the reconstruction characteristic parameters to be detected;
counting the number of distances greater than the similar tolerance;
determining the sample entropy of the reconstruction characteristic parameter to be detected based on the number;
and combining the permutation entropy of the to-be-detected reconstruction characteristic parameters and the sample entropy of the to-be-detected reconstruction characteristic parameters to obtain to-be-detected characteristics.
5. The method for detecting the abnormality of the power equipment based on the variational modal decomposition according to claim 1, wherein the trained convolutional self-encoder is determined by the following method:
acquiring running data of power equipment with a known running state and a corresponding running state to obtain training data;
constructing a convolutional self-encoder model; the convolutional self-encoder model comprises a convolutional encoding layer and a convolutional decoding layer; the convolution coding layer comprises a plurality of convolution layers and a plurality of pooling layers; the convolution decoding layer comprises a plurality of layers of up-sampling layers and a plurality of layers of deconvolution layers;
taking the running data in the training data as input, taking the running state in the training data as output, taking a cross entropy function as a loss function, and training the convolution self-encoder model by adopting a random gradient descent method to obtain an initial convolution self-encoder;
and adjusting the initial convolution self-encoder by adopting an N-fold cross verification method to obtain the trained convolution self-encoder.
6. A power equipment anomaly detection system based on variational modal decomposition, comprising:
the data acquisition module is used for acquiring the operating data of the power equipment to be tested; the operating data is non-stationary random data;
the variation modal decomposition module is used for performing variation modal decomposition on the operation data of the power equipment to be tested to obtain a plurality of intrinsic modal functions to be tested;
the spatial reconstruction module is used for carrying out phase space reconstruction on each intrinsic mode function to be detected to obtain a plurality of reconstruction characteristic parameters to be detected;
the information entropy calculation module is used for calculating the permutation entropy and the sample entropy of each to-be-detected reconstruction characteristic parameter, and combining the permutation entropy of the to-be-detected reconstruction characteristic parameter and the sample entropy of the to-be-detected reconstruction characteristic parameter to obtain to-be-detected characteristics;
the anomaly detection module is used for inputting the characteristics to be detected into the trained convolution self-encoder to obtain a fault diagnosis result of the power equipment to be detected; the trained convolution self-encoder is formed by taking the running data of the power equipment with known running state as input and taking the corresponding running state as output training; the operating state includes a normal state and a fault state.
7. The system for detecting the abnormality of the power equipment based on the variational modal decomposition according to claim 6, wherein the variational modal decomposition module specifically comprises:
the variational model construction unit is used for carrying out variational modal decomposition on the operation data of the power equipment to be tested and constructing a constrained variational model;
the solving unit is used for introducing penalty parameters and penalty factors into the constrained variation model and determining the optimal solution of the constrained variation model according to a Lagrangian function and a Fourier transform method; the optimal solution comprises an optimal mode function to be tested and a central frequency corresponding to the optimal mode function to be tested;
and the screening unit is used for determining the modal component to be tested according to the optimal solution, screening the modal component to be tested and determining the intrinsic modal function to be tested.
8. The system for detecting abnormality of power equipment based on variational modal decomposition according to claim 7, wherein the constrained variational model in the variational model building unit is
Figure FDA0002662266380000031
Figure FDA0002662266380000032
Wherein u iskIs the kth mode function to be measured, k is the serial number of the mode function to be measured, k is 1,2,3, omegakIs the center frequency corresponding to the mode function to be measured,
Figure FDA0002662266380000041
in order to make the derivation of the symbol,
Figure FDA0002662266380000042
is a derivative function, t is time, j is an imaginary number, f (w) is operation data of the power equipment to be tested, uk(t) is the kth modal component at time t.
9. The system for detecting abnormality of power equipment based on variational modal decomposition according to claim 6, wherein the information entropy calculation module specifically comprises:
the sequencing unit is used for sequencing all elements corresponding to each vector in the to-be-tested reconstruction characteristic parameters in an ascending order to obtain a plurality of time sequences;
a probability calculation unit for calculating the occurrence probability of each of the time series;
the permutation entropy calculation unit is used for obtaining the permutation entropy of the characteristic parameters to be reconstructed based on the time sequence;
the distance calculation unit is used for calculating the distance between the maximum element values corresponding to any two vectors in the to-be-detected reconstruction characteristic parameters;
the statistical unit is used for counting the number of the distances greater than the similar tolerance;
the sample entropy calculation unit is used for determining the sample entropy of the reconstruction characteristic parameter to be measured based on the number;
and the merging unit is used for merging the permutation entropy of the to-be-detected reconstruction characteristic parameters and the sample entropy of the to-be-detected reconstruction characteristic parameters to obtain the to-be-detected characteristics.
10. The power equipment anomaly detection system based on variational modal decomposition according to claim 6, characterized by further comprising a convolutional auto-encoder determination module;
the convolutional auto-encoder determining module includes:
the training data acquisition unit is used for acquiring the running data of the power equipment with known running state and the corresponding running state to obtain training data;
the model building unit is used for building a convolutional self-encoder model; the convolutional self-encoder model comprises a convolutional encoding layer and a convolutional decoding layer; the convolution coding layer comprises a plurality of convolution layers and a plurality of pooling layers; the convolution decoding layer comprises a plurality of layers of up-sampling layers and a plurality of layers of deconvolution layers;
the training unit is used for training the convolution self-encoder model by taking the running data in the training data as input, taking the running state in the training data as output and taking a cross entropy function as a loss function and adopting a random gradient descent method to obtain an initial convolution self-encoder;
and the verification unit is used for adjusting the initial convolution self-encoder by adopting an N-fold cross verification method to obtain the trained convolution self-encoder.
CN202010908120.7A 2020-09-02 2020-09-02 Power equipment abnormality detection method and system based on variational modal decomposition Active CN112036042B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010908120.7A CN112036042B (en) 2020-09-02 2020-09-02 Power equipment abnormality detection method and system based on variational modal decomposition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010908120.7A CN112036042B (en) 2020-09-02 2020-09-02 Power equipment abnormality detection method and system based on variational modal decomposition

Publications (2)

Publication Number Publication Date
CN112036042A true CN112036042A (en) 2020-12-04
CN112036042B CN112036042B (en) 2022-07-26

Family

ID=73592237

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010908120.7A Active CN112036042B (en) 2020-09-02 2020-09-02 Power equipment abnormality detection method and system based on variational modal decomposition

Country Status (1)

Country Link
CN (1) CN112036042B (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112706901A (en) * 2020-12-31 2021-04-27 华南理工大学 Semi-supervised fault diagnosis method for main propulsion system of semi-submerged ship
CN112837254A (en) * 2021-02-25 2021-05-25 普联技术有限公司 Image fusion method and device, terminal equipment and storage medium
CN113158814A (en) * 2021-03-26 2021-07-23 清华大学 Bearing health state monitoring method based on convolution self-encoder
CN114199570A (en) * 2021-12-13 2022-03-18 浙江工业大学 Wind power gearbox bearing fault identification method based on improved variational modal decomposition
CN114386452A (en) * 2021-12-06 2022-04-22 西安交通大学 Method for detecting faults of sun wheel of nuclear power circulating water pump
CN114444551A (en) * 2022-04-02 2022-05-06 西南交通大学 Aerodynamic load identification method based on wavelet transform and convolution self-encoder
CN117691629A (en) * 2024-02-04 2024-03-12 西安热工研究院有限公司 Frequency modulation method and system for fused salt coupling thermal power generating unit
CN117977636A (en) * 2024-03-28 2024-05-03 西安热工研究院有限公司 Sample entropy-based frequency modulation method and system for fused salt coupling thermal power generating unit

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105758644A (en) * 2016-05-16 2016-07-13 上海电力学院 Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy
CN110111015A (en) * 2019-05-13 2019-08-09 中南大学 A kind of power quality analysis method based on the multiple dimensioned arrangement entropy of variation mode decomposition
CN110659621A (en) * 2019-09-27 2020-01-07 山东科技大学 Combined noise reduction method based on variational modal decomposition and permutation entropy
CN111397896A (en) * 2020-03-08 2020-07-10 华中科技大学 Fault diagnosis method and system for rotary machine and storage medium
CN111520267A (en) * 2019-12-30 2020-08-11 哈尔滨工程大学 Common rail fuel injector fault diagnosis method based on FOA-VMD and HDE

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105758644A (en) * 2016-05-16 2016-07-13 上海电力学院 Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy
CN110111015A (en) * 2019-05-13 2019-08-09 中南大学 A kind of power quality analysis method based on the multiple dimensioned arrangement entropy of variation mode decomposition
CN110659621A (en) * 2019-09-27 2020-01-07 山东科技大学 Combined noise reduction method based on variational modal decomposition and permutation entropy
CN111520267A (en) * 2019-12-30 2020-08-11 哈尔滨工程大学 Common rail fuel injector fault diagnosis method based on FOA-VMD and HDE
CN111397896A (en) * 2020-03-08 2020-07-10 华中科技大学 Fault diagnosis method and system for rotary machine and storage medium

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
GUIJI TANG等: "Rolling bearing fault diagnosis based on variational mode decomposition and permutation entropy", 《2016 13TH INTERNATIONAL CONFERENCE ON UBIQUITOUS ROBOTS AND AMBIENT INTELLIGENCE (URAI)》, 27 October 2016 (2016-10-27), pages 626 - 630 *
ZHIJIAN WANG 等: "Research on Fault Diagnosis of Gearbox with", 《SENSORS》, 28 October 2018 (2018-10-28), pages 1 - 15 *
彭敏俊等: "压水堆核电站堆芯物理/热工水力耦合特性研究", 《原子能科学技术》, vol. 48, no. 12, 20 December 2014 (2014-12-20), pages 2298 - 2303 *
李亚兰等: "基于VMD和特征融合的辐射源信号识别", 《系统工程与电子技术》, vol. 42, no. 7, 7 January 2020 (2020-01-07), pages 1449 - 1502 *
赵小强等: "使用改进残差神经网络的滚动轴承变工况故障诊断方法", 《西安交通大学学报》, vol. 54, no. 09, 20 May 2020 (2020-05-20), pages 24 - 29 *
陈东宁等: "基于变分模态分解和多尺度排列熵的故障诊断", 《计算机集成制造系统》, vol. 23, no. 12, 23 May 2017 (2017-05-23), pages 2604 - 2611 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112706901B (en) * 2020-12-31 2022-04-22 华南理工大学 Semi-supervised fault diagnosis method for main propulsion system of semi-submerged ship
CN112706901A (en) * 2020-12-31 2021-04-27 华南理工大学 Semi-supervised fault diagnosis method for main propulsion system of semi-submerged ship
CN112837254A (en) * 2021-02-25 2021-05-25 普联技术有限公司 Image fusion method and device, terminal equipment and storage medium
CN112837254B (en) * 2021-02-25 2024-06-11 普联技术有限公司 Image fusion method and device, terminal equipment and storage medium
CN113158814A (en) * 2021-03-26 2021-07-23 清华大学 Bearing health state monitoring method based on convolution self-encoder
CN113158814B (en) * 2021-03-26 2022-06-03 清华大学 Bearing health state monitoring method based on convolution self-encoder
CN114386452B (en) * 2021-12-06 2024-04-05 西安交通大学 Nuclear power circulating water pump sun gear fault detection method
CN114386452A (en) * 2021-12-06 2022-04-22 西安交通大学 Method for detecting faults of sun wheel of nuclear power circulating water pump
CN114199570A (en) * 2021-12-13 2022-03-18 浙江工业大学 Wind power gearbox bearing fault identification method based on improved variational modal decomposition
CN114199570B (en) * 2021-12-13 2024-02-02 浙江工业大学 Wind power gear box bearing fault identification method based on improved variation modal decomposition
CN114444551B (en) * 2022-04-02 2022-06-10 西南交通大学 Aerodynamic load identification method based on wavelet transform and convolution self-encoder
CN114444551A (en) * 2022-04-02 2022-05-06 西南交通大学 Aerodynamic load identification method based on wavelet transform and convolution self-encoder
CN117691629A (en) * 2024-02-04 2024-03-12 西安热工研究院有限公司 Frequency modulation method and system for fused salt coupling thermal power generating unit
CN117691629B (en) * 2024-02-04 2024-04-30 西安热工研究院有限公司 Frequency modulation method and system for fused salt coupling thermal power generating unit
CN117977636A (en) * 2024-03-28 2024-05-03 西安热工研究院有限公司 Sample entropy-based frequency modulation method and system for fused salt coupling thermal power generating unit
CN117977636B (en) * 2024-03-28 2024-06-11 西安热工研究院有限公司 Sample entropy-based frequency modulation method and system for fused salt coupling thermal power generating unit

Also Published As

Publication number Publication date
CN112036042B (en) 2022-07-26

Similar Documents

Publication Publication Date Title
CN112036042B (en) Power equipment abnormality detection method and system based on variational modal decomposition
Li et al. Enhanced generative adversarial networks for fault diagnosis of rotating machinery with imbalanced data
Zou et al. Bearing fault diagnosis based on combined multi-scale weighted entropy morphological filtering and bi-LSTM
CN109974782B (en) Equipment fault early warning method and system based on big data sensitive characteristic optimization selection
Xia et al. Multi-stage fault diagnosis framework for rolling bearing based on OHF Elman AdaBoost-Bagging algorithm
CN103245907B (en) A kind of analog-circuit fault diagnosis method
CN105300692B (en) A kind of bearing failure diagnosis and Forecasting Methodology based on expanded Kalman filtration algorithm
CN102435910B (en) Power electronic circuit health monitoring method based on support vector classification
CN113434970B (en) Health index curve extraction and service life prediction method for mechanical equipment
CN115758212B (en) Mechanical equipment fault diagnosis method based on parallel network and transfer learning
Liu et al. A two-stage approach for predicting the remaining useful life of tools using bidirectional long short-term memory
Wang et al. Prediction model of natural gas pipeline crack evolution based on optimized DCNN-LSTM
CN101900789A (en) Tolerance analog circuit fault diagnosing method based on wavelet transform and fractal dimension
Guo et al. A hybrid prognosis scheme for rolling bearings based on a novel health indicator and nonlinear Wiener process
CN114623050A (en) Method, system and medium for detecting abnormal operation data of wind turbine generator
Son et al. Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge
Zhu et al. Prediction of bearing remaining useful life based on DACN-ConvLSTM model
CN111881594A (en) Non-stationary signal state monitoring method and system for nuclear power equipment
CN111122162A (en) Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy
Zhao et al. A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition
CN112507479A (en) Oil drilling machine health state assessment method based on manifold learning and softmax
CN114819315A (en) Bearing degradation trend prediction method based on multi-parameter fusion health factor and time convolution neural network
CN117609836A (en) Electromagnetic sensitivity prediction and health management method for integrated module
Liu et al. An improved empirical mode decomposition method for vibration signal
Lan et al. Performance degradation prediction model of rolling bearing based on self-checking long short-term memory network

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