A MMC submodule fault monitoring method based on adaptive clustering algorithm

Abstract

The invention discloses a MMC submodule fault monitoring method based on an adaptive clustering algorithm, and relates to the technical field of multi-level power electronic converters. The method comprises the following specific steps: setting a sampling time, extracting the actual capacitor voltage value of each submodule, establishing a fault characteristic quantity data prediction model based on a cubic exponential smoothing method, and obtaining a fault characteristic coefficient CVV[SM] of an MMC system; obtaining a neighborhood radius parameter EPS and a neighborhood density threshold parameter MinPts according to the fault characteristic coefficient CVV[SM] by using a K nearest neighbor algorithm and a mathematical expectation method, and then calculating a neighborhood density parameter Den to form an adaptively determined DBSCAN clustering algorithm; analyzing CVV[SM] by using an adaptively determined DBSCAN algorithm, and if a certain CVV[SM] shows abnormality for three consecutive sampling periods, the corresponding submodule is a faulty submodule. The invention improves the efficiency of fault diagnosis on the basis of reducing costs.

Description

Technical Field

[0001] The invention belongs to the technical field of multi-level power electronic converters, and in particular relates to a MMC submodule fault monitoring method based on an adaptive clustering algorithm. Background Art

[0002] With the continuous development of science and technology, modular multilevel converter (MMC) has shown extensive research value in renewable energy grid connection, electrified railway supply and motor drive due to its modularity and scalability. However, due to the huge number of submodules (SM) in MMC and the fact that each SM contains multiple semiconductor devices, operational reliability is one of the core issues of MMC in medium voltage high power system applications.

[0003] SM faults include short circuit faults (SCF) and open circuit faults (OCF), both of which can distort the voltage and current and destroy the stable operation of the MMC system, while the latter is more difficult to detect. Therefore, a fast and effective open circuit fault location method is of great significance to the safe operation of the MMC system.

[0004] At present, the methods for detecting and locating open-circuit faults of MMC submodules can be roughly divided into three categories: methods based on additional sensors; methods based on mathematical models; methods based on artificial intelligence; however, the above methods increase the manpower and material costs of detecting and locating faulty submodules due to the addition of additional equipment or the need for precise mathematical models. Some methods require manual setting of thresholds, which reduces the robustness of the methods. Therefore, a MMC submodule fault monitoring method based on an adaptive clustering algorithm is proposed. Summary of the invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide an MMC submodule fault monitoring method based on an adaptive clustering algorithm, which solves the technical problem that the methods in the prior art require the construction of complex mathematical models, resulting in increased manpower and material costs for detecting and locating faulty submodules.

[0006] The purpose of the present invention can be achieved by the following technical solution: A method for monitoring MMC submodule faults based on an adaptive clustering algorithm, the method comprising the following steps:

[0007] Set the sampling time, extract the actual capacitor voltage value of each submodule, establish the fault characteristic data prediction model based on the cubic exponential smoothing method, and obtain the MMC system fault characteristic coefficient CVV[SM];

[0008] The neighborhood radius parameter EPS and neighborhood density threshold parameter MinPts are obtained according to the fault characteristic coefficient CVV[SM] using the K nearest neighbor algorithm and the mathematical expectation method, and then the neighborhood density parameter Den is calculated to form an adaptively determined DBSCAN algorithm.

[0009] The adaptively determined DBSCAN algorithm is used to analyze the CVV[SM]. If a certain CVV[SM] shows abnormality for three consecutive sampling cycles, the corresponding submodule is a faulty submodule.

[0010] Furthermore, the process of obtaining the MMC system fault characteristic coefficient CVV[SM] includes:

[0011] Based on the periodic variation trend of the capacitor voltage and the triple exponential smoothing method, the single exponential prediction value u of the capacitor voltage in the kth sampling node is obtained. cp (1) (k), quadratic exponential prediction value u cp (2) (k), cubic exponential prediction value u cp (3) (k):

[0012]

[0013] In the formula, u cp (1) (k-1) is the single exponential prediction value in the k-1th sampling node, u c (k-1) is the actual value in the k-1th sampling node, u cp (2) (k-1) is the quadratic exponential prediction value in the k-1th sampling node, u cp (3) (k-1) is the cubic exponential prediction value in the k-1th sampling node, and α is the smoothing factor coefficient;

[0014] Based on the cubic exponential smoothing method, the capacitor voltage u in the (k+T)th sampling node is obtained cp Predicted value of (k+T):

[0015] u cp (k+T)=A k +B k ·T+C k ·T 2

[0016] in

[0017] A k =3u cp (1) (k)-3u cp (2)(k)+u cp (3) (k)

[0018]

[0019]

[0020] Then the capacitor voltage u in the k+1th sampling node is obtained cp Prediction value of (k+1):

[0021] u cp (k+1)=β (1) ·u cp (1) (k)+β (2) ·u cp (2) (k)+β (3) ·u cp (3) (k)

[0022] in

[0023]

[0024] In the formula, β (1) is the single exponential prediction coefficient, β (2) is the quadratic exponential prediction coefficient, β (3) is the cubic exponential prediction coefficient;

[0025] Set the initial value u cp (1) (1),u cp (2) (1) and u cp (3) (1):

[0026]

[0027] Based on the genetic algorithm and the periodic fluctuation stability of the capacitor voltage value during the normal operation of the MMC system, the predicted value u cp (k) and the actual value u c (k) minimum sum of squared errors MSE(α k ) corresponds to α k , :

[0028]

[0029] Where N T Refers to the number of sampling cycles involved in the iteration;

[0030] Then the fault characteristic coefficient CVV[SM] of the i-th submodule of the MMC system is obtained:

[0031] CVV[SM i ]=u ci (k)-u cpi (k).

[0032] Furthermore, the EPS acquisition process includes:

[0033] Calculate the distance distribution matrix of the CVV[SM] values ​​of N submodules:

[0034] D N×N ={Dist(i,j)|1≤i≤N,1≤j≤N}

[0035] in

[0036] Dist(i,j)=|CVV[SM i ]-CVV[SM j ]|

[0037] Where N represents the number of submodules in a bridge arm, Dist(i,j) represents CVV[SM i ] and CVV[SM j ];

[0038] Distance distribution matrix D N×N This can be obtained by sorting the elements in each row in ascending order:

[0039] D N×N ={D1,D2,...,D K ,...,D N-1 ,D N}

[0040] The elements in the Kth column constitute the K nearest neighbor distance column vector D of all Euclidean distances between two CVV[SM] K ;

[0041] By adding D K Take the arithmetic mean of the elements in to get the vector D K The Kth average nearest neighbor distance D AVK ; and get the parameter list D of the neighborhood radius EPS EPS :

[0042] D Eps ={D AVK |1≤K≤N}.

[0043] Further, based on the obtained EPS parameter list D EPS , when K is a fixed value and EPS = D AVK When CVV[SM i] data point as the center, with D AVK Draw a circle with the radius CVV[SM j ] and CVV[SM i ] is included in the set N EPS (i), N EPS The expression of (i) is:

[0044] N Eps (i)={Dist(i,j)|Dist(i,j)≤D AVK &1≤j≤N}

[0045] P i =||N Eps (i)||,i=1,2,...,N-1,N

[0046] Among them, P i Represents the set N EPS The number of data in (i),

[0047] Calculate the MinPts value when K is a fixed value:

[0048]

[0049] When K = 1, 2, 3, ..., N, generate the parameter list D of the neighborhood density threshold parameter MinPts MinPts :

[0050] D MinPts ={MinPts(K)|1≤K≤N}.

[0051] Further, based on the obtained EPS parameter list D EPS and MinPts parameter list D MinPts , we get EPS = D AVK The average number of MinPts(K) distributed in each unit area in the circle with radius, that is, the density threshold parameter Den:

[0052]

[0053] Further, the parameter list D of the density threshold parameter Den is obtained by the density threshold parameter Den. Den :

[0054] D Den ={Den(K)|1≤K≤N}.

[0055] Further, the steps of analyzing CVV[SM] by using the adaptive DBSCAN algorithm are as follows:

[0056] According to different K, where K = 1, 2, ..., N, select the corresponding Den;

[0057] Since different K results in different numbers of CVV[SM]s in the abnormal cluster set N, when the number of CVV[SM]s in the set N keeps its value stable, the maximum K value is selected as the K value accordingly, and the density threshold is DenK corresponding to the K value, and the cluster set N of CVV[SM] in the core cluster set C and abnormal CVV[SM] is output, where the K value and DenK correspond to K;

[0058] The values ​​in the cluster set N are judged as outliers, and when they last for 3 detection cycles Δt, the outliers are detected and processed. If no fault occurs, CVV[SM] will be captured in the next period for fault location.

[0059] Furthermore, after selecting the corresponding Den, the following operations are performed according to the clustering algorithm:

[0060] Randomly select CVV[SM i ], where i = 1, 2, .., N, and CVV[SM i ] data point as the center, with EPS = D AVK is the radius, check the number of CVV[SM]s in the circle, denoted as Q i ;

[0061] If Q i ≥MinPts(K), CVV[SM i ] is put into the core data set C and recorded as normal data;

[0062] If Q i <MinPts(K), CVV[SM i ] is put into abnormal cluster N and recorded as abnormal data;

[0063] Randomly select CVV[SM] and repeat until every CVV[SM] is selected;

[0064] Let K=K+1, and then repeat the above operation.

[0065] Beneficial effects of the present invention:

[0066] During the use of the present invention, the capacitor voltages of each submodule required in the monitoring process are all parameters that must be obtained by the controller during the stable operation of the modular multi-level converter, so there is no need to increase additional hardware costs. At the same time, the method is easy to implement in the existing MMC system and has strong practicality; there is no need to build an accurate mathematical model and manually set thresholds, and only the fault diagnosis research is carried out based on the mutual relationship of data features, which has strong adaptability and can improve the robustness of the method; there is no need to change the operating state of the system, such as introducing circulating current, and the method is generalizable and suitable for semiconductor device fault detection and positioning under any topology and any working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0068] Figure 1 is a flow chart of the present invention;

[0069] Figure 2 It is a topological structure diagram of the three-phase MMC and submodules of the present invention. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] like Figure 1-2 As shown, a method for monitoring MMC submodule faults based on an adaptive clustering algorithm comprises the following steps: setting a sampling time, extracting the actual capacitance voltage value of each submodule, establishing a fault feature quantity data prediction model based on a cubic exponential smoothing method, and obtaining an MMC system fault feature coefficient CVV[SM]; using a K nearest neighbor algorithm and a mathematical expectation method to obtain a neighborhood radius parameter EPS and a neighborhood density threshold parameter MinPts according to the fault feature coefficient CVV[SM], and then calculating a neighborhood density parameter Den to form an adaptively determined DBSCAN algorithm; using an adaptively determined DBSCAN algorithm to analyze CVV[SM], if a certain CVV[SM] shows an abnormality for three consecutive sampling periods, the corresponding submodule is a faulty submodule.

[0072] The specific steps of the present invention are as follows:

[0073] Set the sampling time, extract the actual capacitor voltage value of each submodule, establish the fault characteristic data prediction model based on the cubic exponential smoothing method, and obtain the MMC system fault characteristic coefficient CVV[SM];

[0074] The neighborhood radius parameter EPS and neighborhood density threshold parameter MinPts are obtained according to the fault characteristic coefficient CVV[SM] using the K nearest neighbor algorithm and the mathematical expectation method, and then the neighborhood density parameter Den is calculated to form an adaptively determined DBSCAN algorithm.

[0075] The adaptive DBSCAN algorithm is used to analyze CVV[SM]. If a CVV[SM] shows abnormality for three consecutive sampling cycles, the corresponding submodule is a faulty submodule.

[0076] The MMC topology consists of six bridge arms, such as Figure 2 As shown, each bridge arm includes N identical submodules (SM) and a bridge arm inductor L s The submodule adopts a half-bridge structure. Each submodule consists of two power switches T1 and T2, two diodes D1 and D2, and a DC capacitor. The capacitor voltage balancing method is: According to the comparison between the bridge arm reference voltage and the carrier, the number of submodules required in a bridge arm is n. on , sort the submodule capacitor voltages in ascending order, and when the bridge arm current is positive, the n capacitors with the lowest capacitor voltage are put into operation. on When the bridge arm current is negative, the capacitor with the highest voltage is n. on submodules.

[0077] A method for monitoring faults of MMC submodules based on an adaptive clustering algorithm comprises: setting a sampling time, extracting actual capacitance voltage values ​​of each submodule, establishing a fault feature quantity data prediction model based on a cubic exponential smoothing method, and obtaining a fault feature coefficient CVV[SM] of an MMC system; obtaining a neighborhood radius parameter EPS and a neighborhood density threshold parameter MinPts according to the fault feature coefficient CVV[SM] by using a K-nearest neighbor algorithm and a mathematical expectation method, and then calculating a neighborhood density parameter Den to form an adaptively determined DBSCAN algorithm; analyzing CVV[SM] by using the adaptively determined DBSCAN algorithm, and if a certain CVV[SM] shows abnormality for three consecutive sampling periods, the corresponding submodule is a faulty submodule.

[0078] The specific steps include:

[0079] (1) Based on the periodic variation trend of the capacitor voltage and the triple exponential smoothing method, the single exponential prediction value u of the capacitor voltage in the kth sampling node is obtained cp (1)(k), quadratic exponential predicted value u cp (2) (k), cubic exponential prediction value u cp (3) (k):

[0080]

[0081] In the formula, u cp (1) (k-1) is the single exponential prediction value in the k-1th sampling node, u c (k-1) is the actual value in the k-1th sampling node, u cp (2) (k-1) is the quadratic exponential prediction value in the k-1th sampling node, u cp (3) (k-1) is the cubic exponential prediction value in the k-1th sampling node, and α is the smoothing factor coefficient.

[0082] Get the capacitor voltage u in the (k+T)th sampling node cp Predicted value of (k+T):

[0083] u cp (k+T)=A k +B k ·T+C k ·T 2

[0084] in

[0085] A k =3u cp (1) (k)-3u cp (2) (k)+u cp (3) (k)

[0086]

[0087]

[0088] Then we can get the capacitor voltage u in the (k+1)th sampling node: cp Prediction value of (k+1):

[0089] u cp (k+1)=β (1) ·u cp (1) (k)+β (2) ·u cp (2) (k)+β (3) ·u cp(3) (k)

[0090] in

[0091]

[0092] In the formula, β (1) is the single exponential prediction coefficient, β (2) is the quadratic exponential prediction coefficient, β (3) is the cubic exponential prediction coefficient.

[0093] Set the initial value u cp (1) (1),u cp (2) (1) and u cp (3) (1):

[0094]

[0095] Based on the genetic algorithm and the periodic fluctuation stability of the capacitor voltage value during the normal operation of the MMC system, the predicted value u cp (k) and the actual value u c (k) corresponds to the minimum sum of squared errors MSE(αk) k , :

[0096]

[0097] Where N T Refers to the number of sampling cycles involved in the iteration.

[0098] Get the fault characteristic coefficient CVV[SM] of the i-th submodule of the MMC system:

[0099] CVV[SM i ]=u ci (k)-u cpi (k)

[0100] (2) Figure 2 As shown, according to the obtained fault characteristic coefficient CVV[SM], the distance distribution matrix of the CVV[SM] values ​​of N submodules is calculated:

[0101] D N×N ={Dist(i,j)|1≤i≤N,1≤j≤N}

[0102] in

[0103] Dist(i,j)=|CVV[SM i ]-CVV[SM j ]|

[0104] Where N represents the number of submodules in a bridge arm, Dist(i,j) represents CVV[SM i ] and CVV[SM j ] is the Euclidean distance between ].

[0105] Distance distribution matrix D N×N This can be obtained by sorting the elements in each row in ascending order:

[0106] D N×N ={D1,D2,...,D K ,...,D N-1 ,D N}

[0107] The elements in the Kth column constitute the K nearest neighbor distance column vector D of all Euclidean distances between two CVV[SM] K .

[0108] By adding D K Take the arithmetic mean of the elements in to get the vector D K The Kth average nearest neighbor distance D AVK . And get the parameter list D of the neighborhood radius EPS EPS :

[0109] D Eps ={D AVK |1≤K≤N}

[0110] Based on the obtained EPS parameter list D EPS , when K is a fixed value and EPS = D AVK When CVV[SM i ] data point as the center, with D AVK Draw a circle with the radius CVV[SM j ] and CVV[SM i ] is included in the set N EPS (i). N EPS (i) The specific expression is:

[0111] N Eps (i)={Dist(i,j)|Dist(i,j)≤D AVK &1≤j≤N}

[0112] P i =||N Eps (i)||,i=1,2,...,N-1,N

[0113] Among them, P i Represents the set N EPS The number of data in (i).

[0114] Calculate the MinPts value when K is a fixed value:

[0115]

[0116] When K = 1, 2, 3, ..., N, generate the parameter list D of the neighborhood density threshold parameter MinPts MinPts :

[0117] D MinPts ={MinPts(K)|1≤K≤N}

[0118] Based on the obtained EPS parameter list D EPS and MinPts parameter list D MinPts , we get EPS = D AVK The average number of MinPts(K) distributed in each unit area in the circle with radius, that is, the density threshold parameter Den:

[0119]

[0120] Then we get the parameter list D of the density threshold parameter Den Den :

[0121] D Den ={Den(K)|1≤K≤N}

[0122] (3) Figure 2 As shown, according to different K (K = 1, 2, ..., N) select the corresponding Den, and according to the clustering algorithm randomly select CVV [SM i ](i=1,2,..,N), and CVV[SM i ] data point as the center, with EPS = D AVK is the radius, check the number of CVV[SM]s in the circle, denoted as Q i If Q i ≥MinPts(K),CVV[SM i ] into the core data set C and recorded as normal data. Otherwise, CVV[SM i ] is put into the abnormal cluster N and recorded as abnormal data. When K is a fixed value, randomly select CVV[SM] and repeat the above operation until each CVV[SM] is selected.

[0123] Different K results in different numbers of CVV[SM]s in the abnormal cluster set N. When the number of CVV[SM]s in the set N keeps its value stable, the maximum K value is selected as the optimal K value accordingly, and the optimal density threshold is DenK corresponding to the K value. The cluster set N of CVV[SM] and abnormal CVV[SM] in the core cluster set C is output, where the K value and DenK correspond to the K value.

[0124] The values ​​in cluster set N can be judged as outliers, and the outliers can be detected and processed when they last for 3 sampling periods. If no fault occurs, CVV[SM] will be captured in the next period for fault location.

[0125] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0126] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for monitoring MMC submodule faults based on an adaptive clustering algorithm, characterized in that: The method comprises the following steps: Set the sampling time, extract the actual capacitor voltage value of each submodule, establish the fault characteristic data prediction model based on the cubic exponential smoothing method, and obtain the MMC system fault characteristic coefficient CVV[SM]; The acquisition process of the MMC system fault characteristic coefficient CVV[SM] includes: Based on the periodic variation trend of the capacitor voltage and the triple exponential smoothing method, the single exponential prediction value u of the capacitor voltage in the kth sampling node is obtained. cp (1) (k), quadratic exponential prediction value u cp (2) (k), cubic exponential prediction value u cp (3) (k): In the formula, u cp (1) (k-1) is the single exponential prediction value in the k-1th sampling node, u c (k-1) is the actual value in the k-1th sampling node, u cp (2) (k-1) is the quadratic exponential prediction value in the k-1th sampling node, u cp (3) (k-1) is the cubic exponential prediction value in the k-1th sampling node, and α is the smoothing factor coefficient; Based on the cubic exponential smoothing method, the capacitor voltage u in the (k+T)th sampling node is obtained cp Predicted value of (k+T): u cp (k+T)=A k +B k ·T+C k ·T 2 in A k =3u cp (1) (k)-3u cp (2) (k)+u cp (3) (k) Then the capacitor voltage u in the k+1th sampling node is obtained cp Prediction value of (k+1): in cp (k+1)=β (1) ·in cp (1) (k)+β (2) ·in cp (2) (k)+β (3) ·in cp (3) (k) in In the formula, β (1) is the single exponential prediction coefficient, β (2) is the quadratic exponential prediction coefficient, β (3) is the cubic exponential prediction coefficient; Set the initial value u cp (1) (1),u cp (2) (1) and u cp (3) (1): Based on the genetic algorithm and the periodic fluctuation stability of the capacitor voltage value during the normal operation of the MMC system, the predicted value u cp (k) and the actual value u c (k) minimum sum of squared errors MSE(α k ) corresponds to α k , : Where N T Refers to the number of sampling cycles involved in the iteration; Then the fault characteristic coefficient CVV[SM] of the i-th submodule of the MMC system is obtained: CVV[SM i ]=u ci (k)-u cpi (k); The neighborhood radius parameter EPS and neighborhood density threshold parameter MinPts are obtained according to the fault characteristic coefficient CVV[SM] using the K nearest neighbor algorithm and the mathematical expectation method, and then the neighborhood density parameter Den is calculated to form an adaptively determined DBSCAN algorithm. The adaptively determined DBSCAN algorithm is used to analyze the CVV[SM]. If a certain CVV[SM] shows abnormality for three consecutive sampling cycles, the corresponding submodule is a faulty submodule; The steps of analyzing CVV[SM] using the adaptively determined DBSCAN algorithm are as follows: According to different K, where K = 1, 2, ..., N, select the corresponding Den; Since different K results in different numbers of CVV[SM]s in the abnormal cluster set N, when the number of CVV[SM]s in the set N keeps its value stable, the maximum K value is selected as the K value accordingly, and the density threshold is DenK corresponding to the K value, and the cluster set N of CVV[SM] in the core cluster set C and abnormal CVV[SM] is output, where the K value and DenK correspond to K; If the value in the cluster set N is judged as an outlier, and the outlier is detected and processed when it lasts for 3 detection cycles Δt, if no fault occurs, CVV[SM] will be captured in the next period for fault location.

2. The MMC submodule fault monitoring method based on an adaptive clustering algorithm according to claim 1, characterized in that: The EPS acquisition process includes: Calculate the distance distribution matrix of the CVV[SM] values ​​of N submodules: D N×N ={Dist(i,j)|1≤i≤N,1≤j≤N} in Dist(i,j)=|CVV[SM i ]-CVV[SM j ]| Where N represents the number of submodules in a bridge arm, Dist(i,j) represents CVV[SM i ] and CVV[SM j ]; Distance distribution matrix D N×N This can be obtained by sorting the elements in each row in ascending order: D N×N ={D1,D2,...,D K ,...,D N-1 ,D N } The elements in the Kth column constitute the K nearest neighbor distance column vector D of all Euclidean distances between two CVV[SM] K ; By adding D K Take the arithmetic mean of the elements in to get the vector D K The Kth average nearest neighbor distance D AVK ; and get the parameter list D of the neighborhood radius EPS EPS : D Eps ={D AVK |1≤K≤N}。 3. The MMC submodule fault monitoring method based on the adaptive clustering algorithm according to claim 2, characterized in that: Based on the obtained EPS parameter list D EPS , when K is a fixed value and EPS = D AVK When CVV[SM i ] data point as the center, with D AVK Draw a circle with the radius CVV[SM j ] and CVV[SM i ] is included in the set N EPS (i), N EPS The expression of (i) is: N Eps (i)={Dist(i,j)|Dist(i,j)≤D AVK &1≤j≤N} P i =||N Eps (i)||,i=1,2,...,N-1,N Among them, P i Represents the set N EPS The number of data in (i), Calculate the MinPts value when K is a fixed value: When K = 1, 2, 3, ..., N, generate the parameter list D of the neighborhood density threshold parameter MinPts MinPts : D MinPts ={MinPts(K)|1≤K≤N}。 4. The MMC submodule fault monitoring method based on the adaptive clustering algorithm according to claim 3 is characterized in that: Based on the obtained EPS parameter list D EPS and MinPts parameter list D MinPts , we get EPS = D AVK The average number of MinPts(K) distributed in each unit area in the circle with radius, that is, the density threshold parameter Den:

5. The MMC submodule fault monitoring method based on the adaptive clustering algorithm according to claim 4 is characterized in that: The parameter list D of the density threshold parameter Den is obtained by the density threshold parameter Den Den : D Den ={Den(K)|1≤K≤N}。 6. The MMC submodule fault monitoring method based on an adaptive clustering algorithm according to claim 1, characterized in that: After selecting the corresponding Den, the following operations are performed according to the clustering algorithm: Randomly select CVV[SM i ], where i = 1, 2, .., N, and CVV[SM i ] data point as the center, with EPS = D AVK is the radius, check the number of CVV[SM]s in the circle, denoted as Q i ; If Q i ≥MinPts(K), CVV[SM i ] is put into the core data set C and recorded as normal data; If Q i <MinPts(K), CVV[SM i ] is put into abnormal cluster N and recorded as abnormal data; Randomly select CVV[SM] and repeat until every CVV[SM] is selected; Let K=K+1, and then repeat the above operation.

Images