MMC sub-module IGBT open-circuit fault diagnosis method based on improved density clustering algorithm

Abstract

The application relates to an MMC sub-module IGBT open-circuit fault diagnosis method based on an improved density clustering algorithm and belongs to the field of power electronic devices. The method comprises the following steps: extracting the capacitor voltage of each sub-module in a modular multilevel converter, and obtaining the capacitor voltage spectrum after discrete processing and Fourier transformation; extracting a feature vector containing a frequency mean value, a root mean square frequency, a standard deviation frequency and a center frequency according to the capacitor voltage spectrum; adopting an improved density clustering algorithm based on a GOOSE optimization algorithm to classify faults for the feature vector corresponding to each sub-module; determining the sub-module corresponding to the output result of the improved density clustering algorithm existing a noise point as an open-circuit fault, and judging the fault category according to the device position in the sub-module. The improved DBSCAN algorithm based on the GOOSE optimization algorithm is used to judge the fault sub-module, can diagnose multiple fault points, has high diagnosis precision and strong real-time performance.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of power electronic devices and relates to an MMC sub-module IGBT open-circuit fault diagnosis method based on an improved density clustering algorithm. BACKGROUND

[0002] A modular multilevel converter (MMC) is a new type of topological structure of a voltage source converter, and has been widely applied in high-voltage and high-power fields due to its high modularity, high output level, low switching frequency, and good output voltage waveform. The MMC is formed by cascading a large number of sub-modules, and an insulated gate bipolar transistor (IGBT) and a diode are used as switching devices in each sub-module. However, the fault rate of the MMC increases due to the large number of sub-modules, and the fault detection is difficult. Therefore, the fault diagnosis technology for the MMC is particularly important.

[0003] The faults of an IGBT device mainly include open-circuit faults and short-circuit faults. The short-circuit fault causes a large current to flow through the bridge arm in a short time, which greatly damages the entire MMC system. However, in actual applications, the IGBT device is usually integrated with a short-circuit protection function, which can quickly exclude the short-circuit fault. The open-circuit fault has less significant characteristics than the short-circuit fault, and the device may be in an open-circuit fault state for a long time and cannot be detected, which causes problems such as overvoltage and overcurrent in the system and greatly damages the entire MMC system. At present, for the open-circuit fault diagnosis of the sub-module, some scholars use an artificial neural network method to train the algorithm with a large amount of data to achieve the diagnosis effect. However, the algorithm has high complexity and slow diagnosis speed. Some scholars use the influence of the fault on the bridge arm voltage value to diagnose the fault. The method is simple and fast in diagnosis, but can only diagnose single sub-module faults. Therefore, it is of great significance to realize the rapid diagnosis of multiple sub-module faults for the safety and stability of the MMC system. SUMMARY

[0004] Therefore, the purpose of the present application is to provide an MMC sub-module IGBT open-circuit fault diagnosis method based on an improved density clustering algorithm, which combines device fault feature extraction and fault classification methods to realize the open-circuit fault diagnosis of the MMC sub-module.

[0005] To achieve the above purpose, the present application provides the following technical solutions.

[0006] An MMC sub-module IGBT open-circuit fault diagnosis method based on an improved density clustering algorithm includes the following steps:

[0007] S1, the capacitor voltage of each sub-module in the modular multilevel converter is extracted, and the capacitor voltage spectrum is obtained after discrete processing and Fourier transform.

[0008] S2, extracting a feature vector including frequency mean, root mean square frequency, standard deviation frequency and center frequency according to the capacitor voltage spectrum;

[0009] S3. The feature vector corresponding to each submodule is used to classify the fault using the density clustering algorithm improved based on GOOSE;

[0010] S4. Determine the submodule corresponding to the output result of the improved density clustering algorithm with noise points as an open circuit fault, and determine the fault type according to the device position in the submodule.

[0011] Furthermore, in step S1, the capacitor voltage v of each submodule SM(i) is extracted from the modular multilevel converter MMC. ci , the capacitor voltage v ci Perform discrete processing to obtain discrete capacitor voltage U c (n), for discrete capacitor voltage U c (n) After fast Fourier transform, the capacitor voltage spectrum c(k) is obtained, k = 1, 2, ..., M, M is the number of spectrum lines, and f is used c (k) represents the corresponding frequency value.

[0012] Further, in step S2, the characteristic vector F is extracted from the capacitor voltage spectrum c(k). C :

[0013] F C =(F c1 ,F c2 ,F c3 ,F c4 )

[0014] Among them, F c1 ,F c2 ,F c3 ,F c4 is the extracted feature quantity, and its corresponding calculation methods are:

[0015]

[0016]

[0017]

[0018]

[0019] In the above formula, F c1 ,F c2 ,F c3 ,F c4 They represent the frequency mean, root mean square frequency, standard deviation frequency, and center frequency respectively.

[0020] Furthermore, in step S3, the density clustering algorithm based on the GOOSE optimization algorithm is used to cluster the feature vector F C Classify each sample in, which includes the following steps:

[0021] S31, the feature vector F corresponding to each submodule C As data to be classified;

[0022] S32, using GOOSE optimization algorithm to calculate the neighborhood radius ε R and density threshold ρ minp Perform optimization to obtain the optimal parameters;

[0023] S33, the feature vector F corresponding to the submodule C Select a sample point and determine the optimal area radius ε R and density threshold ρ minp Determine each sample point as a core point, boundary point or noise point, and then complete the classification, where

[0024] For a sample point t, t∈F C , then first calculate the neighborhood radius ε of the sample point t R The number of samples N R (t), and determine the neighborhood radius ε R The number of samples N R (t)≥ρ minp If it is satisfied, the sample point t is the core point; if not, it is further judged whether the sample point t falls within the neighborhood radius ε of another core point. R If it falls within, the sample t is a boundary point, otherwise the sample point t is a noise point.

[0025] Further, in step S32, the neighborhood radius ε is optimized using the GOOSE optimization algorithm. R and density threshold ρ minp The process is:

[0026] S321, first initialize the population position X gi , that is, randomly select a group (ε R ,ρ minp ), and specify the search space range and the maximum number of iterations Z m , if the initial position of the population exceeds the search space range, it will be reinitialized;

[0027] S322, evaluate the fitness G of each goose's position ai And compare to get the best fitness G abest and the corresponding optimal position X abestAnd save it, where the development and exploration stages are allocated by introducing a random variable rd, where 0≤rd≤1. If rd≥0.5, it enters the development stage, otherwise it enters the exploration stage, and two other random variables po and co are introduced to select the formula, and their values ​​are between 0 and 1, where co should also be less than the specified value ≥, otherwise α is assigned to co.

[0028] Furthermore, during the development phase, the weight of the stone under the goose's feet in the current iteration is calculated using the following formula:

[0029] W s =rand([w1,w2],1,1)

[0030] Where w1 and w2 are the estimated ranges of the stone weight, and then the time T required for the stone to fall to the ground is calculated. e1 The time it takes for the sound of the stone hitting the ground to propagate to other individuals is T e2 :

[0031] T e1 =rand(1,D)

[0032] T e2 =rand(1,D)

[0033] Where D is the dimension of each iteration, then the total time T for the sound to propagate to each individual in a single iteration is e and average time T av for:

[0034]

[0035]

[0036] The distance S that sound travels a and the distance S between the two bodies g As follows, where the distance between individuals is half the distance that sound travels:

[0037] S a =υ a ·T e2

[0038]

[0039] where υ a is the speed of sound in air.

[0040] If the variables po and W s If the values ​​are greater than the specified values ​​a and b respectively, the falling speed of the stone υ is calculated by the following formula s And update the goose's position:

[0041]

[0042]

[0043] where g is the acceleration due to gravity.

[0044] If the variables po and W s are less than the specified values ​​a and b respectively, the falling speed of the stone υ is calculated by the following formula s And update the goose's position:

[0045]

[0046]

[0047] Where, X a(it+1) Indicates the updated position of the goose.

[0048] Furthermore, in the search phase, the sound propagation time T in each iteration is compared. e And find the shortest time M e , and introduce a new variable β, the formula is as follows:

[0049]

[0050] Where lp is the current iteration number, M x is the total number of iterations, and the position of the goose is updated according to the following formula:

[0051] X a(it+1) =rand(1,D)·(M e β)+X abest

[0052] Repeat the above steps until the maximum number of iterations is reached or the preset conditions are met, and the algorithm outputs the optimal position X abest is the optimal parameter.

[0053] Furthermore, in step S4, if there are noise points in the results output by the improved density clustering algorithm, it is determined that the corresponding submodule has an open circuit fault, and according to the different device locations of the open circuit fault, it is divided into device Tu fault, device Tl fault, and simultaneous fault of devices Tu and Tl.

[0054] The beneficial effects of the present invention are:

[0055] The proposed MMC submodule open circuit fault diagnosis method cleverly combines the advantages of the GOOSE optimization algorithm and DBSCAN to achieve efficient and accurate identification of MMC submodule open circuit faults. The introduction of the GOOSE optimization algorithm significantly improves the speed and accuracy of data processing, enabling the fault diagnosis process to rapidly respond to changes in system status and meet application scenarios with high real-time requirements. The application of the DBSCAN algorithm, with its powerful clustering analysis capabilities, effectively distinguishes the data characteristics of normal and faulty submodules. Even in complex situations where multiple fault points exist simultaneously, it can accurately locate them one by one, greatly improving diagnostic accuracy.

[0056] This invention eliminates the need for additional hardware sensor installation, avoiding the associated cost increases and increased system complexity. Furthermore, it is not limited by variations in specific system parameters, exhibiting excellent adaptability and robustness. This means that regardless of the scale or structure of the power system, the method maintains stable diagnostic performance. This innovative approach, independent of precise mathematical models, significantly broadens the application scope of fault diagnosis technology and reduces its implementation difficulty, benefiting a wide range of power electronics devices, from large power grids to distributed energy systems.

[0057] In summary, the MMC submodule open circuit fault diagnosis method proposed in the present invention has high precision, strong real-time performance, low cost and wide applicability.

[0058] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0060] Figure 1 Schematic diagram of the overall process of the open circuit fault diagnosis method of the present invention;

[0061] Figure 2 This is a schematic diagram of the fault classification process of the improved density clustering algorithm of the present invention;

[0062] Figure 3 4 is a fault diagnosis simulation waveform diagram of an open circuit of device Tu in SM1 of the MMC simulation model under the embodiment;

[0063] Figure 44 is a fault diagnosis simulation waveform diagram of an open circuit between device Tu in SM1 and device T1 in SM2 of the MMC simulation model under the embodiment;

[0064] Figure 5 4 is a fault diagnosis simulation waveform diagram of the device Tu and the device T1 in SM1 of the MMC simulation model under the embodiment being open at the same time. DETAILED DESCRIPTION

[0065] The following describes the embodiments of the present invention by means of specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0066] Among them, the accompanying drawings are only for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting the present invention. In order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the dimensions of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions may be omitted in the accompanying drawings.

[0067] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "back", etc. indicating directions or positional relationships, they are based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operate in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0068] See also Figures 1 to 5 , which is a MMC sub-module IGBT open circuit fault diagnosis method based on improved density clustering algorithm.

[0069] Example

[0070] This embodiment is based on the construction of a three-phase 6-level MMC simulation model in Simulink, and proposes a specific implementation method of an MMC submodule IGBT open circuit fault diagnosis method based on an improved density clustering algorithm, such as Figure 1 As shown, it includes the following steps:

[0071] S1. Extracting the capacitor voltage of each submodule in the modular multilevel converter and obtaining the capacitor voltage spectrum after discrete processing and Fourier transform.

[0072] S2, extracting a feature vector including frequency mean, root mean square frequency, standard deviation frequency and center frequency according to the capacitor voltage spectrum;

[0073] S3. The feature vector corresponding to each submodule is used to classify the fault using the density clustering algorithm improved based on GOOSE;

[0074] S4. Determine the submodule corresponding to the output result of the improved density clustering algorithm with noise points as an open circuit fault, and determine the fault type according to the device position in the submodule.

[0075] In step S1, the capacitor voltage v of each submodule SM(i) is extracted from the actual modular multilevel converter MMC. ci In this embodiment, the capacitor voltage is extracted from the constructed MMC simulation model. ci Perform discrete processing to obtain discrete capacitor voltage U c (n), for discrete capacitor voltage U c (n) After performing fast Fourier transform (FFT), the capacitor voltage spectrum c(k) is obtained, where k = 1, 2, ..., M, where M is the number of spectrum lines, and f c (k) is the corresponding frequency value.

[0076] In step S2, the frequency mean F is selected from the capacitor voltage spectrum c(k). c1 , RMS frequency F c2 , standard deviation frequency F c3 and the center of gravity frequency F c4 is the feature quantity, and all the feature quantities are combined into a feature vector F C , where the corresponding feature quantity is calculated as follows:

[0077]

[0078]

[0079]

[0080]

[0081] Then the eigenvector F C Expressed as:

[0082] F C =(F c1 ,F c2 ,Fc3 ,F c4 )

[0083] In step S3, the feature vector F C As the data to be classified by the algorithm, the improved density clustering algorithm DBSCAN based on the GOOSE optimization algorithm is used to classify faults; in the traditional density clustering algorithm, there is a sample t, t∈F C And in a certain neighborhood radius ε R It contains a number of samples. If the number of samples N R (t)≥ρ minp When , the sample t is called the core point, ρ minp is the density threshold set. If the number of samples in the neighborhood of sample t is less than the density threshold, but it is in the neighborhood of a core point, it is called a boundary point. A point that is neither a core point nor a boundary point is a noise point. In this invention, the neighborhood radius ε is optimized by the GOOSE algorithm. R and density threshold ρ minp To perform optimization processing. Specifically, Figure 2 As shown, the improved density clustering algorithm DBSCAN based on the GOOSE optimization algorithm of the present invention performs fault classification, including the following steps:

[0084] S31, the feature vector F corresponding to each submodule C As data to be classified;

[0085] S32, using GOOSE optimization algorithm to calculate the neighborhood radius ε R and density threshold ρ minp Perform optimization to obtain the optimal parameters;

[0086] S33, the feature vector F corresponding to the submodule C Select a sample point and determine the optimal area radius ε R and density threshold ρ minp Determine each sample point as a core point, boundary point or noise point to complete the classification.

[0087] In step S32, the neighborhood radius ε is optimized using the GOOSE optimization algorithm. R and density threshold ρ minp The process is:

[0088] S321, first initialize the population position X gi , that is, randomly select a group (ε R ,ρ minp ), and specify the search space range and the maximum number of iterations Z m , if the initial position of the population exceeds the search space range, it will be reinitialized;

[0089] S322, evaluate the fitness G of each goose's position ai And compare to get the best fitness G abest and the corresponding optimal position X abest And save it, where the development and exploration stages are allocated by introducing a random variable rd, where 0≤rd≤1. If rd≥0.5, it enters the development stage, otherwise it enters the exploration stage, and two other random variables po and co are introduced to select the formula, and their values ​​are between 0 and 1, where co should also be less than a certain specified value α, otherwise α is assigned to co.

[0090] During the development phase, the weight of the stone under the goose's feet in the current iteration is calculated using the following formula:

[0091] W s =rand([w1,w2],1,1)

[0092] Where w1 and w2 are the estimated ranges of the stone weight, and then the time T required for the stone to fall to the ground is calculated. e1 The time it takes for the sound of the stone hitting the ground to propagate to other individuals is T e2 :

[0093] T e1 =rand(1,D)

[0094] T e2 =rand(1,D)

[0095] Where D is the dimension of each iteration, then the total time T for the sound to propagate to each individual in a single iteration is e and average time T av for:

[0096]

[0097]

[0098] The distance S that sound travels a and the distance S between the two bodies g As follows, where the distance between individuals is half the distance that sound travels:

[0099] S a =v a ·T e2

[0100]

[0101] where υ a is the speed of sound in air.

[0102] If the variables po and W s If the values ​​a and b are greater than a certain value respectively, the falling speed of the stone υ can be calculated by the following formula s And update the goose's position:

[0103]

[0104]

[0105] where g is the acceleration due to gravity.

[0106] If the variables po and W s If the values ​​a and b are less than a certain value, the falling speed v of the stone can be calculated by the following formula: s And update the goose's position:

[0107]

[0108]

[0109] In the search phase, the sound propagation time T in each iteration is compared e And find the shortest time M e , and introduce a new variable β, the formula is as follows:

[0110]

[0111] Where lp is the current iteration number, M x is the total number of iterations. Then the position of the goose can be updated according to the following formula:

[0112] X a(it+1) =rand(1,D)·(M e β)+X abest

[0113] Repeat the above steps until the maximum number of iterations is reached or the preset conditions are met, and the algorithm outputs the optimal position X abest is the optimal parameter.

[0114] In step S4, if there are noise points in the results output by the improved density clustering algorithm, it is determined that the corresponding submodule has an open circuit fault. Based on the different locations of the open circuit fault components, it is divided into three categories: failure of the upper switch tube Tu of the submodule, failure of the lower switch tube Tl, and simultaneous failure of components Tu and Tl, which are represented by fault signals 1, 2, and 3 respectively.

[0115] In this embodiment, a three-phase 6-level MMC simulation model is built, and the improved density clustering algorithm is used to analyze the MMC submodules. Set the first submodule Tu of the upper bridge arm to have an open circuit fault, and use the algorithm to perform open circuit fault diagnosis. The fault diagnosis simulation waveform is as follows Figure 3As shown, the fault time is 0.263s, the diagnosis time is 0.267s, and the total time is 4.1ms.

[0116] Set the upper bridge arm SM1 Tu and SM2 Tl to have open circuit faults, perform fault diagnosis, and the simulation waveform is as follows Figure 4 As shown, the fault time of SM1 is 0.263s, the diagnosis time is 0.267s, and the time taken is 4.2ms; the fault time of SM2 is 0.320s, the diagnosis time is 0.324s, and the time taken is 4.6ms.

[0117] Set Tu and Tl of the upper bridge arm SM1 to fail at the same time, perform fault diagnosis, and the simulation waveform is as follows Figure 5 As shown, the fault time is 0.275s, the diagnosis time is 0.279s, and the total time is 4.5ms.

[0118] In summary, the proposed method can effectively diagnose the open circuit fault of the sub-module IGBT, has a fast judgment speed and can diagnose multiple fault points.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for diagnosing open-circuit faults of MMC submodule IGBTs based on an improved density clustering algorithm, characterized by: It includes the following steps: S1. Extract the capacitor voltage of each submodule in the modular multilevel converter, and obtain the capacitor voltage spectrum after discrete processing and Fourier transform. S2, extracting a feature vector including frequency mean, root mean square frequency, standard deviation frequency and center frequency according to the capacitor voltage spectrum; S3. The feature vector corresponding to each submodule is used to classify the fault using a density clustering algorithm improved based on the GOOSE optimization algorithm; S4. Determine the submodule corresponding to the output result of the improved density clustering algorithm with noise points as an open circuit fault, and determine the fault type based on the device position in the submodule; In step S3, the density clustering algorithm based on the GOOSE optimization algorithm is used to cluster the feature vectors Classify each sample in, which includes the following steps: S31, the feature vector corresponding to each submodule As data to be classified; S32, using GOOSE optimization algorithm to calculate the neighborhood radius and density threshold Perform optimization to obtain the optimal parameters; S33, the feature vector corresponding to the submodule Select a sample point and determine the optimal area radius and density threshold Determine each sample point as a core point, boundary point or noise point, and then complete the classification, where For the sample points , , then first calculate the sample points Neighborhood radius The number of samples within , and determine the neighborhood radius The number of samples within Is it satisfied? If so, then the sample point As the core point; if it does not meet the requirements, further judge the sample point Whether it falls within the neighborhood radius of another core point If it falls within, the sample is a boundary point, otherwise the sample point is a noise point; In step S32, the neighborhood radius is optimized using the GOOSE optimization algorithm. and density threshold The process is: S321, first initialize the population position , that is, randomly select a group , and specify the search space range and maximum number of iterations , if the initial position of the population exceeds the search space range, it will be reinitialized; S322. Evaluate the fitness of each goose's position And compare to get the best fitness and the corresponding optimal position and save, where, by introducing the random variable to allocate between development and exploration phases, where ,like Then it enters the development phase, otherwise it enters the exploration phase, and introduces two other random variables and To select the formula, the value is between 0 and 1, where Should also be less than the specified value Otherwise, Assign to ; During the development phase, the weight of the stone under the goose's feet in the current iteration is calculated using the following formula: in and The estimated range of the stone weight is then used to calculate the time it takes for the stone to fall to the ground. The time it takes for the sound of a stone hitting the ground to reach other individuals : in is the dimension of each iteration, then the total time it takes for the sound to propagate to each individual in a single iteration is and average time for: Distance sound travels and the distance between the two bodies As follows, where the distance between individuals is half the distance that sound travels: in is the speed of sound in air; If the variable and Greater than the specified value and , the falling speed of the stone can be calculated by the following formula And update the goose's position: in is the acceleration due to gravity; If the variable and Less than the specified value and , the falling speed of the stone can be calculated by the following formula And update the goose's position: Where, Indicates the updated position of the goose; During the search phase, the sound travel time is compared in each iteration And find the shortest time , while introducing a new variable , the formula is as follows: in is the current iteration number, is the total number of iterations, and the position of the goose is updated according to the following formula: Repeat the above steps until the maximum number of iterations is reached or the preset conditions are met, and the algorithm outputs the optimal position is the optimal parameter; In step S4, if there are noise points in the results output by the improved density clustering algorithm, it is determined that the corresponding submodule has an open circuit fault, and according to the different device locations of the open circuit fault, it is divided into device Tu fault, device T1 fault, and simultaneous fault of devices Tu and T1.

2. The MMC submodule IGBT open circuit fault diagnosis method based on the improved density clustering algorithm according to claim 1, characterized in that: In step S1, each submodule is extracted from the modular multilevel converter MMC. Capacitor voltage , the capacitor voltage Perform discrete processing to obtain discrete capacitor voltage , for discrete capacitor voltages After fast Fourier transform, the capacitor voltage spectrum is obtained , , M is the number of spectrum lines, using Indicates the corresponding frequency value.

3. The MMC submodule IGBT open circuit fault diagnosis method based on the improved density clustering algorithm according to claim 2, characterized in that: In step S2, the capacitor voltage spectrum Extract feature vectors : in, is the extracted feature quantity, and its corresponding calculation methods are: In the above formula, They represent the frequency mean, root mean square frequency, standard deviation frequency, and center frequency respectively.

Images