Method for judging intermittent fault of inverter system through distribution characteristics

Abstract

The invention discloses a method for judging an intermittent fault of an inverter system through distribution characteristics, and the method comprises the steps: collecting and processing three-phase voltage signals of the inverter system in normal operation at equal intervals, and obtaining the characteristic data of each sampling moment; then completing feature extraction of the feature data by using a plurality of basic detectors after offline training; then, the KL divergence of sequence distribution is calculated to serve as a difference factor to complete long-term sequence distribution feature extraction, and a sliding window singular value serves as a single-point feature of a monitored object; and finally, fusing the two types of feature factors as a monitoring index so as to complete real-time online detection of the intermittent fault of the inverter system.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and more specifically, relates to a method for identifying intermittent faults of inverter systems by using distribution characteristics. Background Art

[0002] With the development of science and technology, the importance of power systems in modern society has become increasingly significant. As one of the key components of the DC-AC conversion process in the power grid, the inverter system has been widely used in photovoltaic energy, wind power generation, aerospace and other fields. Once the inverter system fails, it may cause the shutdown of the entire power system, seriously affecting social production and causing a large amount of property losses. In addition, as the core component of the power system, the inverter system's failure will be transferred and spread with the grid topology, thereby affecting the stability and service life of other components in the system. Therefore, timely and effective fault detection technology can effectively protect the safety and stability of the power system, and can minimize downtime and maintenance costs.

[0003] In power electronic systems, intermittent faults are a common type of fault. In contrast to permanent faults, intermittent faults often disappear on their own without human intervention and reappear randomly during the operation of the system. Intermittent faults often have small fault amplitudes and short durations. Therefore, for intermittent faults, the detection method needs to have both good real-time and sensitivity. Existing intermittent fault detection methods can be divided into analytical model methods and data-driven methods. The analytical model method requires mathematical modeling of the actual process and achieves fault detection by comparing the actual data with the model output. Existing analytical models often calculate residuals by constructing observers. However, the analytical model method relies heavily on very accurate modeling and parameter estimation of the actual system, which is costly and has a long development cycle. In addition, with the iterative update of technology, the constructed model needs to change its structure and adjust its parameters again, which will bring greater overhead in practical applications. At present, data-driven methods have received widespread attention, which replaces the elaborate and cumbersome model construction with rich historical data training. Principal component analysis (PCA) is a classic data-driven method, but its direct application in the field of intermittent fault detection has not achieved good detection results. In order to extract the dynamic characteristics of the process, the dynamic principal component analysis method (Dynamic PCA, DPCA) was also proposed. In addition, the idea of ​​sliding window was introduced to extract the sequence information of the process, and the moving average principal component analysis method (Moving average PCA, MAPCA) was proposed. It is noted that different methods are often more sensitive to different data features, and integrating different detectors as a new detection model often results in better fault detection performance. However, most of the existing integration frameworks only utilize single-point features of the data or simply perform linear transformation through sliding window technical solutions, which often makes it difficult to effectively extract the sequence distribution characteristics of the process. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for distinguishing intermittent faults of an inverter system by distribution characteristics, by measuring the difference in sequence distribution information of multiple variables in a sliding window and designing an intermittent fault detection index based on singular value decomposition, thereby completing the intermittent fault detection of the inverter system.

[0005] To achieve the above-mentioned object of the invention, the present invention provides a method for identifying intermittent faults of an inverter system by using distribution characteristics, characterized in that it comprises the following steps:

[0006] (1) Collect and process the three-phase voltage signals of the inverter system under normal operation at equal intervals;

[0007] (2) Build a weak detector and train it offline;

[0008] (3) Setting control limits for detecting intermittent faults in the inverter system through sliding window analysis and processing;

[0009] (4) Collect the characteristic data of the inverter system in real time, calculate the fault monitoring index at the current moment, and then perform real-time online detection of intermittent faults of the inverter system according to the set control limits.

[0010] The object of the invention of the present invention is achieved in this way:

[0011] The invention discloses a method for distinguishing intermittent faults of an inverter system by using distribution characteristics. The three-phase voltage signals of the inverter system under normal operation are collected at equal intervals and processed to obtain characteristic data at each sampling moment. Then, a plurality of basic detectors after offline training are used to complete feature extraction of the characteristic data. Subsequently, the KL divergence of the sequence distribution is calculated as a difference factor to complete the extraction of long-term sequence distribution characteristics, and a sliding window singular value is used as a single point feature of a monitoring object. Finally, the two types of characteristic factors are fused as monitoring indicators to complete real-time online detection of intermittent faults of the inverter system.

[0012] At the same time, the method for distinguishing intermittent faults of an inverter system by distribution characteristics of the present invention also has the following features:

[0013] Beneficial effects:

[0014] (1) The present invention utilizes integrated sliding window technology to fuse the statistics obtained by different basic detectors through sliding window singular values, and calculates the bi-norm of the singular value vector to characterize the single point characteristics of the sample, so as to perform intermittent fault detection;

[0015] (2) The present invention calculates the probability density distribution in the sliding window and the overall probability density distribution respectively, and then measures the difference between the above two density distributions through KL divergence. Finally, it is combined with the sliding window singular value as a state monitoring indicator of the three-phase inverter system, thereby realizing effective process feature extraction and intermittent fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a method for distinguishing intermittent faults of an inverter system by distribution characteristics according to the present invention;

[0017] Figure 2 This is the detection result of the intermittent fault of inductance degradation of the inverter circuit by the present invention. DETAILED DESCRIPTION

[0018] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0019] Example

[0020] Figure 1 The present invention is a flow chart of a method for distinguishing intermittent faults of an inverter system through distribution characteristics.

[0021] In this embodiment, if Figure 1 As shown, the present invention provides a method for identifying intermittent faults of an inverter system by using distribution characteristics, comprising the following steps:

[0022] (1) Data collection and processing;

[0023] (1.1) The three-phase voltage signal of the inverter system under normal operation is collected at equal intervals of 1 second, where the single-phase voltage signal collected at the i-th moment is denoted as U i,Θ , Θ represents any one of the three phases of the inverter system, i=1,2,…,N, N represents the number of acquisitions, in this embodiment, N=4000;

[0024] (1.2) Extract the single-phase voltage signal U i,Θ The characteristic data in i,Θ , including frequency, mean, peak-to-peak value, effective value, margin, and kurtosis;

[0025] (1.3) The feature data extracted from the three single-phase voltage signals are combined into a feature data set E = {e1,…,e i ,…e N},e i represents the characteristic data extracted at the i-th moment. In this embodiment, the three-phase voltage signal includes three single-phase voltage signals, and six types of characteristic data are extracted from each single-phase voltage signal. Therefore, the characteristic data at one moment includes 18 characteristic quantities in total;

[0026] (2) Build a weak detector and train it offline;

[0027] Construct principal component analysis PCA, dynamic principal component analysis DPCA, kernel principal component analysis KPCA, sliding window average principal component analysis MAPCA, exponentially weighted sliding window average principal component analysis EWMAPCA and the most weighted sliding window average T 2 The control chart method OWMA-TCC has six basic detectors, and then the six basic detectors are trained offline until convergence;

[0028] (3) Sliding window analysis;

[0029] (3.1) The feature data e at each moment in the feature data set E i They are respectively input into the six basic detectors after offline training, among which PCA, DPCA, MAPCA, KPCA and EWMAPCA output T 2and Q statistics, and outputs a T through OWMA-TCC 2 Statistics, the 11 statistics extracted at each moment form a feature vector u i ={f1(e i ),f2(e i ),...,f 11 (e i )};

[0030] (3.2) Concatenate the feature vectors extracted at N sampling moments into a feature integration matrix U in the column dimension;

[0031]

[0032] (3.3) Normalize the feature matrix U and first calculate the mean of each column element in U, where the mean of the kth column element of U is denoted as μ k , k=1,2,…,11;

[0033] The mean u of each column k Forming a row matrix

[0034] Then, U is normalized to a matrix with a mean of 0 and a variance of 1 according to the following formula:

[0035]

[0036] The superscript T indicates transposition, 1 ω represents the identity matrix with element 1 and length ω, Σ is the covariance matrix of U, Σ * It means to find the inverse transformation of Σ;

[0037] remember The normalized elements in are Then the matrix It is expressed as:

[0038]

[0039] (3.4) Set the sliding window with a window size of ω×11 and a sliding step of 1, where the width ω~[12,N);

[0040] In this embodiment, assuming that the size of the sliding window is 45×11, in the process of sliding from top to bottom, when the sliding window slides to the N-44th row, the sliding window stops;

[0041] (3.5) Taking the first element f1(e1) of the unnormalized feature matrix U as the starting point, slide the matrix U from top to bottom using the sliding window according to the sliding step, and record the corresponding matrix after each sliding, where the sliding window matrix corresponding to the jth sliding is U j ;

[0042] (3.6) Normalize each sliding window matrix;

[0043] For any sliding window matrix U j , first calculate U j The mean of each column element in U j The mean of the k-th column elements is denoted by μ j,k , k=1,2,…,11;

[0044] The mean u of each column j,k Forming a row matrix

[0045] Then, according to the following formula, U j Normalized to a matrix with mean 0 and variance 1

[0046]

[0047] Among them, Σ j It's U j The covariance matrix of j ) * Indicates the search for Σ j The inverse transformation of

[0048] remember The normalized elements in are Then the matrix It is expressed as:

[0049]

[0050] (3.7), for each normalized matrix Perform differentiation metric factor calculations;

[0051] (3.7.1), firstly, The kernel density distribution of each column statistic in is estimated, where the kth column is estimated as:

[0052]

[0053] Among them, g k(x) is the estimated probability density distribution of the kth column, x is the variable corresponding to the required estimated probability density, h is a setting constant, K(·) represents the Gaussian kernel function, and in this embodiment, h is set to 1;

[0054] (3.7.2), then the matrix The kernel density distribution of each column statistic in is estimated, where the kth column is estimated as:

[0055]

[0056] Among them, g k (x) is the estimated probability density distribution of the kth column, x is the variable corresponding to the required estimated probability density, h is a setting constant, K(·) represents the Gaussian kernel function, and in this embodiment, h is set to 1;

[0057] (3.7.3), calculation matrix The probability density distribution of each column and the matrix The KL divergence between the probability density distributions of each column, where the KL divergence between the probability density distributions of the kth column is calculated as:

[0058]

[0059] Among them, X represents the value range of variable x;

[0060] (3.7.4) Sum the KL divergence of each column and use it as the difference measurement factor θ j :

[0061]

[0062] (3.8), for the matrix Perform singular value decomposition and obtain the singular value vector [σ j,1 ,...,σ j,k ,...,σ j,11 ], where σ j,k is the kth singular value obtained by decomposition;

[0063] (3.9), calculate the singular value factor d j :

[0064]

[0065] in, is the mean of the kth element of the singular value vector, std k is the standard deviation of the kth element of the singular value vector;

[0066] (3.10), according to the singular value factor d j and the difference measure θj Calculate the fault monitoring index D j :

[0067]

[0068] (3.11), set control limits;

[0069] (3.11.1) Fault monitoring index D j Perform kernel density estimation:

[0070]

[0071] Wherein, g(x) is the estimated probability density, x is the variable corresponding to the required estimated probability density, h is a setting constant, K(·) represents the Gaussian kernel function, and in this embodiment, h is set to 1;

[0072] (3.11.2) Set the control limit Δ, which satisfies:

[0073]

[0074] Wherein, α is the set confidence level. In this embodiment, α is 99%;

[0075] (4) Online status monitoring of inverter system;

[0076] (4.1) According to step (1), the characteristic data e of the inverter system at the N+1 moment is collected. N+1 , e N+1 Input into six basic detectors respectively to extract the feature vector u at the N+1th moment N+1 ={f1(e N+1 ),f2(e N+1 ),…,f 11 (e N+1 )};

[0077] Then add it to the sliding window matrix U j and delete the last row of the sliding window matrix U j The first row of the sliding window matrix U at time N+1 is obtained N+1 ;

[0078] (4.2) Repeat steps (3.5) to (3.10) to calculate the fault monitoring index D at the N+1th moment N+1 ;

[0079] (4.3) According to the set control limit Δ, the state of the inverter system at the N+1th moment is determined. If D N+1 >Δ, it is judged that an intermittent fault occurs in the inverter system at this moment; otherwise, it is considered that the inverter system is operating normally.

[0080] Experimental simulation

[0081] The platform used for verification of the present invention consists of four parts, namely, a DC power supply, an inverter circuit system, an electronic load and a data acquisition module. Among them, the supply voltage of the DC power supply is controlled to 80V, the IGBT in the inverter circuit system adopts the SKM100GAL12T4 module produced by Siemens, the IGBT driver core model is 2SC0435T2F1-17, the adapter board model used is 2BB043, and the mode of the three-way electronic load is set to AC constant resistance mode, with a resistance of 50 ohms. The implanted fault is the filter inductor at the output end of the inverter circuit system, and a resistor is connected in parallel to the inductor to simulate the degraded intermittent fault of the inductor. In this experiment, 4000 normal data were collected to train the model. The test data consists of 4000 samples, of which the parallel resistance values ​​of 1000-2000 and 3300-3600 samples are 500Ω, 2400-3000 samples are 1000Ω, and the rest are normal data. The fault detection rate is calculated as the number of samples exceeding the control limit within the fault duration. Then, the false alarm rate is obtained by using the number of false positive samples outside the fault duration.

[0082] In this experiment, principal component analysis (PCA), dynamic PCA (DPCA), kernel PCA (KPCA), moving average PCA (MAPCA), exponentially weighted MAPCA (EWMAPCA), optimal weighted sliding window average T 2 Control chart (Optimal weighted moving average T 2 control chart, OWMA-TCC) and the FEDD of the present invention verify their performance in the inverter system. The CPV of the PCA-based method is 80%, and the delay of DPCA is 1. The kernel function of KPCA is the radial basis function with a parameter of 0.000016. The moving window length of MAPCA is 7 and the forgetting factor is 0.7. The window width of OWMA-TCC is set to 8, and the window width of FEDD is set to 15. The significance level of each detector is set to 1%, and the control limits are calculated by KDE.

[0083] Table 1 lists the fault detection rates of the verified methods for intermittent faults of inductance degradation in inverter systems. Among them, the detection effects of PCA, DPCA and KPCA are not satisfactory. For PCA, the false alarm rate obtained by monitoring the T2 statistic is 0.29%, and the fault detection rate is only 1.32%, which fails to achieve effective fault detection. In addition, the final fault detection rate of PCA's Q statistic is only 0.84% ​​under the condition of a false alarm rate of 0.38%, which is not an effective detection indicator. Similarly, the T2 and Q statistics of DPCA and KPCA cannot achieve effective fault detection, and the fault detection rates of the above methods do not exceed 2%. Due to the introduction of sliding window technology, the fault detection rates of the three methods of MAPCA, EWMAPCA and OWMA-TCC have been greatly improved. As the intermittent fault detection model with the best effect, OWMA-TCC can detect 84.37% of intermittent faults. The integrated detection method proposed in the present invention has achieved a fault detection rate of 93.68% at a false alarm rate of 2.71%, which is 9.31% higher than the fault detection rate of the best existing method. The above results demonstrate the superiority and effectiveness of the proposed method.

[0084] Table 1 shows the fault detection rate (%) of the verification method for the intermittent fault of inductance degradation in the inverter system;

[0085]

[0086] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A method for identifying intermittent faults in an inverter system by using distribution characteristics, characterized in that: The following steps are involved: (1) Collect and process the three-phase voltage signals of the inverter system under normal operation at equal intervals; (2) Build a weak detector and train it offline; (3) Setting control limits for detecting intermittent faults in the inverter system through sliding window analysis and processing; (4) Collect the characteristic data of the inverter system in real time, calculate the fault monitoring index at the current moment, and then perform real-time online detection of intermittent faults of the inverter system according to the set control limits.

2. A method for identifying intermittent faults of an inverter system by using distribution characteristics according to claim 1, characterized in that: The method for collecting and processing three-phase voltage signals in step (1) is as follows: (2.1) The three-phase voltage signals of the inverter system under normal operation are collected at equal intervals, where the single-phase voltage signal collected at the i-th moment is denoted as U i,Θ , Θ represents any phase in the three phases of the inverter system, i=1,2,...,N, N represents the number of acquisitions; (2.2) Extract the single-phase voltage signal U i,Θ The characteristic data in i,Θ , including frequency, mean, peak-to-peak value, effective value, margin, and kurtosis; (2.3) The feature data extracted from the three single-phase voltage signals are combined into a feature data set E = {e1,...,e i ,...e N },e i Represents the feature data extracted at the i-th moment.

3. A method for identifying intermittent faults of an inverter system by using distribution characteristics according to claim 1, characterized in that: The weak detectors include: principal component analysis PCA, dynamic principal component analysis DPCA, kernel principal component analysis KPCA, sliding window average principal component analysis MAPCA, exponentially weighted sliding window average principal component analysis EWMAPCA and the most weighted sliding window average T 2 Control chart method OWMA-TCC six basic detectors.

4. The method for identifying intermittent faults of an inverter system by using distribution characteristics according to claim 1, characterized in that: The process of setting the control limit in step (3) is as follows: (4.1) The feature data e at each moment in the feature data set E i They are respectively input into the six basic detectors after offline training, among which PCA, DPCA, MAPCA, KPCA and EWMAPCA output T 2 and Q statistics, and outputs a T through OWMA-TCC 2 Statistics, the 11 statistics extracted at each moment form a feature vector u i ={f1(e i ),f2(e i ),...,f 11 (e i )}; (4.2) Concatenate the feature vectors extracted at N sampling moments into a feature integration matrix U in the column dimension; (4.3) Normalize the feature matrix U and first calculate the mean of each column element in U, where the mean of the kth column element of U is denoted as μ k , k=1,2,…,11; The mean u of each column k Forming a row matrix Then, U is normalized to a matrix with a mean of 0 and a variance of 1 according to the following formula: The superscript T indicates transposition, 1 ω represents the identity matrix with element 1 and length ω, Σ is the covariance matrix of U, Σ * It means to find the inverse transformation of Σ; remember The normalized elements in are Then the matrix It is expressed as: (4.4) Set the sliding window with a window size of ω×11 and a sliding step of 1, where the width ω~[12,N); (4.5) Taking the first element f1(e1) of the unnormalized feature matrix U as the starting point, slide the matrix U from top to bottom using the sliding window according to the sliding step, and record the corresponding matrix after each sliding, where the sliding window matrix corresponding to the jth sliding is U j ; (4.6) Normalize each sliding window matrix; For any sliding window matrix U j , first calculate U j The mean of each column element in U j The mean of the k-th column elements is denoted by μ j,k , k=1,2,…,11; The mean u of each column j,k Forming a row matrix Then, U is calculated according to the following formula: j Normalized to a matrix with mean 0 and variance 1 Among them, Σ j It's U j The covariance matrix of j ) * Indicates the search for Σ j The inverse transformation of remember The normalized elements in are Then the matrix It is expressed as: (4.7), for each normalized matrix Perform differentiation metric factor calculations; (4.7.1), firstly, The kernel density distribution of each column statistic in is estimated, where the kth column is estimated as: Among them, g k (x) is the estimated probability density distribution of the kth column, x is the variable corresponding to the required estimated probability density, h is the setting constant, and K(·) represents the Gaussian kernel function; (4.7.2), then the matrix The kernel density distribution of each column statistic in is estimated, where the kth column is estimated as: Among them, g k (x) is the estimated probability density distribution of the kth column, x is the variable corresponding to the required estimated probability density, h is the setting constant, and K(·) represents the Gaussian kernel function; (4.7.3), calculation matrix The probability density distribution of each column and the matrix The KL divergence between the probability density distributions of each column, where the KL divergence between the probability density distributions of the kth column is calculated as: Among them, X represents the value range of variable x; (4.7.4), sum up the KL divergence of each single column and use it as the difference measurement factor θ j : (4.8), for the matrix Perform singular value decomposition and obtain the singular value vector [σ j,1 ,...,σ j,k ,...,σ j,11 ], where σ j,k is the kth singular value obtained by decomposition; (4.9), calculate the singular value factor d j : in, is the mean of the kth element of the singular value vector, std k is the standard deviation of the kth element of the singular value vector; (4.10), according to the singular value factor d j and the difference measure θ j Calculate the fault monitoring index D j : (4.11), set control limits; (4.11.1) Fault monitoring indicator D j Perform kernel density estimation: Where g(x) is the estimated probability density, x is the variable corresponding to the required estimated probability density, h is the setting constant, and K(·) represents the Gaussian kernel function; (4.11.2) Set the control limit Δ, which satisfies: Among them, α is the set confidence level.

5. The method for identifying intermittent faults of an inverter system by using distribution characteristics according to claim 1, characterized in that: The method for real-time online detection of intermittent faults in the inverter system is: (5.1) According to step (1), the characteristic data e of the inverter system at the N+1 moment is collected. N+1 , e N+1 Input into six basic detectors respectively to extract the feature vector u at the N+1th moment N+1 ={f1(e N+1 ),f2(e N+1 ),…,f 11 (e N+1 )}; then add it to the sliding window matrix U j and delete the last row of the sliding window matrix U j The first row of the sliding window matrix U at time N+1 is obtained N+1 ; (5.2), by sliding window matrix U N+1 Perform sliding window analysis to calculate the fault monitoring index D at the N+1th moment N+1 ; (5.3) According to the set control limit Δ, the state of the inverter system at the N+1th moment is determined. If D N+1 >Δ, it is judged that an intermittent fault occurs in the inverter system at this moment; otherwise, it is considered that the inverter system is operating normally.

Images