MMC-MG fault power generation module diagnosis method based on voltage deviation signal

Abstract

The MMC-MG fault power generation module diagnosis method based on the voltage deviation signal comprises the following steps: firstly, taking voltage deviation signals output by power generation modules under different fault types as effective basis for GM fault diagnosis; then, establishing a bridge arm power generation module diagnosis model based on a deep extreme learning machine; in order to solve the problem that the accuracy fluctuation is large when a deep extreme learning machine (DELM) is adopted, a seagull algorithm is used for optimizing the weight and threshold parameters of a power generation module diagnosis model, so that the generalization ability and the average accuracy of fault GM diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of renewable energy power generation, and in particular to a fault diagnosis technology of a modular multilevel converter half-bridge series microgrid (MMC-MG) system. Background Art

[0002] In the context of distributed energy with multi-energy complementarity, in order to improve the utilization rate of renewable energy, some scholars have proposed a modular multilevel converter half-bridge series microgrid (MMC-MG) system with an MMC half-bridge series structure. Each phase of the system is composed of two inductors L and 2K generation modules (GM) in cascade, where GM is composed of micro-sources such as photovoltaic or wind power, converters, energy storage devices (ESD), and half-bridge converters (HC). The system has the advantages of flexible distributed energy control, low output voltage harmonic content, and less grid-connected pollution. At the same time, the MMC-MG system plays a key role in modern power systems, and can effectively connect distributed energy (such as solar energy, wind energy, etc.) to the power grid to achieve efficient energy conversion and stable transmission.

[0003] Under the grid-connected double closed-loop control, V1 and V2 in the half-bridge converter of the power generation module GM are always in complementary conduction mode, that is, the input and removal of GM are achieved by controlling the conduction and shutdown of V1 and V2 in HC. When a fault occurs in the power generation module GM, if it cannot be diagnosed in a timely and accurate manner, it may cause serious consequences. If an IGBT in a submodule has an open circuit fault, it will cause abnormal output voltage of the submodule, thereby affecting the output performance of the entire converter. This may cause problems such as output current distortion and power fluctuation, reduce power quality, and may even damage other electrical equipment.

[0004] In the actual microgrid operating environment, fault diagnosis is crucial to ensure the reliability and stability of the system. Especially in situations where high power quality requirements are required, timely detection and handling of MMC-MG power generation module faults can avoid power outages and reduce economic losses. At present, in traditional fault diagnosis methods based on analytical models, it is very difficult to establish an accurate analytical model due to the complex internal structure of the MMC-MG power generation module, the presence of various nonlinear components and complex control strategies. Moreover, model parameters may change with factors such as ambient temperature and component aging, resulting in inaccurate diagnostic results. Signal processing-based methods mainly analyze and process the collected electrical signals (such as voltage and current signals). However, when faced with complex fault types and variable operating conditions, these methods may not be able to effectively extract fault features.

[0005] Therefore, the present invention provides a method for diagnosing power generation module faults based on voltage deviation signals and improved deep extreme learning machines (DELM), and selects the deviation signal of GM output voltage under each fault type as the characteristic attribute of fault diagnosis through analysis. At the same time, by learning a large amount of fault sample data, an effective and optimized fault diagnosis model is established to improve the accuracy and reliability of power generation module fault diagnosis. Summary of the invention

[0006] The purpose of the present invention is to improve the accuracy and reliability of diagnosis of a microgrid fault power generation module with an MMC half-bridge series structure, and to solve the problem of effectively extracting fault characteristics.

[0007] The present invention is a method for diagnosing a faulty MMC-MG power generation module based on a voltage deviation signal, and the steps are as follows:

[0008] Step 1: Collect the output voltage deviation value Δu of the power generation module when it is operating normally and when it fails yxi , deviation value Δu yxi It can be expressed as follows:

[0009] Δu yxi =u yxi -u yxi_c (Formula 1)

[0010] Where: u yxi (y=p,n;x=a,b,c;i=1,2,...,K) represents the actual output voltage of the i-th power generation module GM in the x-phase y-bridge arm; u yxi_c (y=p,n;x=a,b,c;i=1,2,...,K) represents the actual output voltage and ideal output voltage of the i-th power generation module GM in the x-phase y-bridge arm, u yxi_c is a fixed value, taking u yxi_c=160V; Δu yxi represents the output voltage deviation value of the i-th power generation module GM in the x-phase y-bridge arm; K represents the total number of power generation modules GM in the bridge arm, 2≤K≤50;

[0011] Step 2: Use the output voltage deviation value Δu of the power generation module yxi As its fault characteristic value;

[0012] After the i-th power generation module GM in the bridge arm fails, the actual output voltage u of the power generation module GM within a power frequency cycle is yxi With the ideal output voltage u yxi_c If they are not equal, there will be a certain deviation, and the deviation value of the output voltage will also be different when different faults occur.

[0013] Δu yxi =u yxi -u yxi_c ≠0 (Formula 2)

[0014] When the i+1th power generation module GM in the bridge arm is not faulty, the actual output voltage u of the power generation module GM is yx(i+1) With the ideal output voltage u yx(i+1)_c equal, the deviation is 0, that is

[0015] Δu yx(i+1) =u yx(i+1) -u yx(i+1)_c =0 (Formula 3)

[0016] The output voltage deviation Δu of each power generation module in the bridge arm when it is operating normally and when a fault occurs is used. yx1 , Δu yx2 , Δu yx3 , ..., Δu yx(i+1) , Δu yxi As a characteristic signal for fault diagnosis of power generation module;

[0017] Step 3: Combine the K fault feature deviation values ​​Δu obtained in step 2 yx1 , Δu yx2 , Δu yx3 , ..., Δu yx(i+1) , Δu yxi , and arbitrarily select 75% of the K fault feature deviation values ​​as the training set, and the remaining 25% as the test set, to establish a GM diagnosis model for faulty power generation modules based on voltage deviation signals and Deep Extreme Learning Machine (DELM);

[0018] The input voltage fault characteristic deviation value sample is (r yxi ,t yxi), r yxi =[Δu yxi1 ,Δu yxi2 ,...,Δu yxK ] T , t yxi =[t yx1 ,t yx2 ,...,t yxm ] T r yxi ∈R n , t yxi ∈R m ; Among them, r yxi is the sample data with dimension K, t yxi is the expected output of dimension m; in the deep extreme learning machine, the weight w i and threshold b i Randomly generated, the output of the network structure is expressed as follows

[0019]

[0020] Where: L is the number of hidden layer nodes; w j =[w j1 ,w j2 ,...,w jK ] T is the weight vector connecting the input node and the jth hidden layer node; g is the activation function; β j =[β j1 ,β j2 ,...,β jm ] T is the weight vector between the output node and the jth hidden layer node; b j is the bias of the jth hidden layer node; then Get the unique least squares solution of β As shown below:

[0021]

[0022] In the formula, G represents the k×l-dimensional hidden layer output matrix, G is a singular matrix; G + is the generalized inverse matrix of G; G T is a singular matrix; is the least squares solution of β;

[0023] The input voltage fault feature data sample r is used as the target output r1 of the first extreme learning autoencoder, that is, r = r1, and the output weight β1 is obtained; then the output matrix G1 of the extreme learning machine is used as the input and target output of the next extreme learning autoencoder, that is, r = r2, and so on, to obtain the output weight β of the last hidden layer. i+1 .

[0024] Step 4: Use the Seagull algorithm to calculate the weight w of the power generation module fault diagnosis model based on the voltage deviation signal and deep extreme learning machine built in steps 2 and 3. i and threshold b i The parameters are optimized to obtain an improved power generation module fault diagnosis optimization model with higher accuracy;

[0025] Step 5: Use the test set data to classify the faults of the power generation module;

[0026] The invention is beneficial in that: the power generation module fault diagnosis method based on voltage deviation signal and improved deep extreme learning machine first selects the deviation signal of GM output voltage under each fault type as the characteristic attribute of fault diagnosis through theoretical analysis method, and at the same time, establishes an effective and optimized fault diagnosis model by learning a large amount of voltage deviation fault sample data, thereby improving the accuracy and reliability of power generation module fault diagnosis. In addition, the seagull algorithm is used to optimize the weight and threshold parameters of the power generation module fault diagnosis model based on voltage deviation signal, thereby improving the accuracy of fault classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is the topological structure diagram of the MMC half-bridge series structure microgrid system; Figure 2 It is the bridge arm topology of the system; Figure 3 (a) is a schematic diagram of the actual output voltage and its deviation of the power generation module when an open circuit fault occurs in V1 of the half-bridge converter; Figure 3 (b) is a schematic diagram of the actual output voltage and its deviation of the power generation module when V1 in the half-bridge converter has a short circuit fault; Figure 4 It is the fault diagnosis process of the power generation module in the bridge arm; Figure 5 This is a schematic diagram of the fault output result of the first submodule of the lower bridge arm of phase A; Figure 6 It is a schematic diagram of the results of multiple experiments on the faults of the power generation modules in the lower bridge arm of phase A. DETAILED DESCRIPTION

[0028] The MMC-MG fault power generation module diagnosis method of the improved deep extreme learning machine involved in the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] The grid-connected topology structure of the MMC half-bridge series structure microgrid system of the present invention is as follows: Figure 1 As shown. The system has six bridge arms YX (Y = P, N; X = A, B, C); PA, PB, PC represent the three upper bridge arms of the system, and NA, NB, NC represent the three lower bridge arms of the system. Each bridge arm is composed of K power generation modules connected in series with an inductor L, as shown in Figure 2As shown in the figure, GM is composed of a photovoltaic or wind power micro source, an energy storage device ES and a half-bridge converter (HC). Among them, HC contains two IGBTs (V1 and V2) and two anti-parallel diodes VD1 and VD2.

[0030] The present invention improves the accuracy and reliability of diagnosis of a microgrid fault power generation module with an MMC half-bridge series structure and solves the problem of effective extraction of fault features. A method for diagnosing a MMC-MG fault power generation module based on a voltage deviation signal and an improved deep extreme learning machine is provided, and the steps are as follows:

[0031] Step 1: When the system is operating normally, the number of power generation modules GM actually put into use in the bridge arm is K on The number of GMs K required for system control on_c Equal, that is, the actual output voltage of the power generation module is equal to the ideal output voltage, that is:

[0032]

[0033] In the formula, K on , K on_c They represent the number of power generation modules actually put into use and the number of power generation modules required to be put into use in the bridge arm respectively; u yxi 、u yxi_c (y=p,n;x=a,b,c) are respectively the actual output voltage and ideal output voltage of the i-th GM in the x-phase y-bridge arm;

[0034] In addition, the actual total output voltage of the bridge arm u yx Should be equal to the actual output voltage u of each power generation module yxi The sum of the GM output voltage u yxi_c The sum of the ideal voltage value and the actual voltage value is 0:

[0035]

[0036] Among them, u yx_c is the ideal value of the total output voltage of the x-phase y-bridge arm;

[0037] Step 2: Collect and calculate the output voltage deviation value Δu of the power generation module GM when it fails yxi , deviation value Δu yxi It can be expressed as follows:

[0038] Δu yxi =u yxi -u yxi_c (Formula 3)

[0039] Where: u yxi(y=p,n;x=a,b,c;i=1,2,...,K) represents the actual output voltage of the i-th power generation module GM in the x-phase y-bridge arm; u yxi_c (y=p,n;x=a,b,c;i=1,2,...,K) represents the actual output voltage and ideal output voltage of the i-th power generation module GM in the x-phase y-bridge arm, u yxi_c is a fixed value, taking u yxi_c =160V; Δu yxi represents the output voltage deviation value of the i-th power generation module GM in the x-phase y-bridge arm; K represents the total number of power generation modules GM in the bridge arm, 2≤K≤50;

[0040] Step 3: Use the output voltage deviation value Δu of the power generation module yxi As its fault characteristic value;

[0041] When V1 in GM_an1 in the lower bridge arm of phase A is open or short-circuited, the actual output voltage, expected output voltage, and voltage deviation of the power generation module are as follows: Figure 3 (a) and 3(b). Figure 3 It can be seen that within a cycle, when V1 has an open circuit fault and is in the SPWM state, the deviation of its output voltage is not equal to 0 and the deviation is large; when GM needs to be cut off, the output voltage deviation is always 0. When V1 has a short circuit fault, the output voltage deviation is always not equal to 0, and the deviation is larger during the period when it is in the SPWM state.

[0042] After the i-th power generation module GM in the bridge arm fails, the actual output voltage u of the power generation module GM within a power frequency cycle is yxi With the ideal output voltage u yxi_c If they are not equal, there will be a certain deviation, and the deviation value of the output voltage will also be different when different faults occur.

[0043] Δu yxi =u yxi -u yxi_c ≠0 (Formula 4)

[0044] When the i+1th power generation module GM in the bridge arm is not faulty, the actual output voltage u of the power generation module GM is yx(i+1) With the ideal output voltage u yx(i+1)_c equal, the deviation is 0, that is

[0045] Δu yx(i+1) =u yx(i+1) -u yx(i+1)_c =0 (Formula 5)

[0046] The output voltage deviation value Δu of each power generation module GM of the bridge arm is adopted yx1 , Δuyx2 , Δu yx3 , ..., Δu yx(i+1) , Δu yxi As a characteristic value for fault diagnosis of power generation module;

[0047] Step 4: Combine the K fault feature deviation values ​​Δu obtained in step 3 yx1 , Δu yx2 , Δu yx3 , ..., Δu yx(i+1) , Δu yxi , and arbitrarily select 75% of the K fault feature deviation values ​​as the training set, and the remaining 25% as the test set, to establish a GM diagnosis model for faulty power generation modules based on the Deep Extreme Learning Machine (DELM);

[0048] The input voltage fault characteristic deviation value sample is (r yxi ,t yxi ), r yxi =[Δu yxi1 ,Δu yxi2 ,...,Δu yxK ] T , t yxi =[t yx1 ,t yx2 ,...,t yxm ] T r yxi ∈R n , t yxi ∈R m ; Among them, r yxi is the sample data with dimension K, t yxi is the expected output of dimension m; in the deep extreme learning machine, the weight w i and threshold b i Randomly generated, the output of the learning machine network structure is expressed as follows

[0049]

[0050] Where: L is the number of hidden layer nodes; w j =[w j1 ,w j2 ,...,w jK ] T is the weight vector connecting the input node and the jth hidden layer node; g is the activation function; β j =[β j1 ,β j2 ,...,β jm ] T is the weight vector between the output node and the jth hidden layer node; bj is the bias of the jth hidden layer node; then Get the unique least squares solution of β As shown below:

[0051]

[0052] In the formula, G represents the k×l-dimensional hidden layer output matrix, G is a singular matrix; G + is the generalized inverse matrix of G; G T is a singular matrix; is the least squares solution of β;

[0053] Step 5: Take the input voltage fault feature data sample r as the target output r1 of the first extreme learning autoencoder, that is, r = r1, and obtain the output weight β1; then take the output matrix G1 of the extreme learning machine as the input and target output of the next extreme learning autoencoder, that is, r = r2, and so on, to obtain the output weight β of the last hidden layer i+1 .

[0054] Step 6: Use the Seagull algorithm to optimize the weights and threshold parameters of the power generation module fault diagnosis model based on the voltage deviation signal built in steps 3, 4, and 5 to obtain a power generation module fault diagnosis optimization model with high accuracy; and use the test set data to classify the faults of the power generation module;

[0055] Set the population size to 20, the number of iterations to 50, the weight boundary to [-1,1], and the number of hidden layers of ELM-AE to 20. Figure 5 The schematic diagram of the fault result of the first submodule of the lower bridge arm of phase A is given. As can be seen from the figure, the accuracy is 91.70% when the deep extreme learning machine DELM is used, and 100% when the improved deep extreme learning machine DELM is used. In order to further verify the effectiveness of the power generation module fault diagnosis model based on the voltage deviation signal and the improved deep extreme learning machine, multiple diagnosis experiments are carried out for each submodule fault, and the change curve of its accuracy is shown in the figure. Figure 6 As shown. Figure 6 It can be seen that in the diagnosis of each fault generating sub-module in the bridge arm of this system, when DELM is used, the fluctuation of its accuracy is large, between 35% and 100%; when the improved deep extreme learning machine is used, the output results of the fault diagnosis of each sub-module are relatively stable and its accuracy is close to 100%, which verifies the effectiveness of the fault diagnosis model based on voltage deviation signal.

[0056] The invention is beneficial in that: the power generation module fault diagnosis method based on voltage deviation signal and improved deep extreme learning machine first selects the deviation signal of GM output voltage under each fault type as the characteristic attribute of fault diagnosis through theoretical analysis method, and at the same time, establishes an effective and optimized fault diagnosis model by learning a large amount of voltage deviation fault sample data, thereby improving the accuracy and reliability of power generation module fault diagnosis. In addition, the seagull algorithm is used to optimize the weight and threshold parameters of the power generation module fault diagnosis model based on voltage deviation signal, thereby improving the accuracy of fault classification.

[0057] The above is one of the implementation methods of the present invention. For ordinary technicians in this field, the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be clear that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A method for diagnosing a faulty MMC-MG power generation module based on a voltage deviation signal, characterized in that: The steps are: Step 1: Collect the output voltage deviation value Δu of the power generation module when it is operating normally and when it fails yxi , deviation value Δu yxi It can be expressed as follows: Δu yxi =u yxi -u yxi_c (Formula 1) Where: u yxi (y=p,n;x=a,b,c;i=1,2,...,K) represents the actual output voltage of the i-th power generation module GM in the x-phase y-bridge arm; u yxi_c (y=p,n;x=a,b,c;i=1,2,...,K) represents the actual output voltage and ideal output voltage of the i-th power generation module GM in the x-phase y-bridge arm, u yxi_c is a fixed value, taking u yxi_c =160V; Δu yxi represents the output voltage deviation value of the i-th power generation module GM in the x-phase y-bridge arm; K represents the total number of power generation modules GM in the bridge arm, 2≤K≤50; Step 2: Use the output voltage deviation value Δu of the power generation module yxi As its fault characteristic value; After the i-th power generation module GM in the bridge arm fails, the actual output voltage u of the power generation module GM within a power frequency cycle is yxi With the ideal output voltage u yxi_c If they are not equal, there will be a certain deviation, and the deviation value of the output voltage will also be different when different faults occur. Δu yxi =u yxi -u yxi_c ≠0 (Formula 2) When the i+1th power generation module GM in the bridge arm is not faulty, the actual output voltage u of the power generation module GM is yx(i+1) With the ideal output voltage u yx(i+1)_c equal, the deviation is 0, that is Δu yx(i+1) =u yx(i+1) -u yx(i+1)_c =0 (Formula 3) The output voltage deviation Δu of each power generation module in the bridge arm when it is operating normally and when a fault occurs is used. yx1 , Δu yx2 , Δu yx3 , ..., Δu yx(i+1) , Δu yxi As a characteristic signal for fault diagnosis of power generation module; Step 3: Combine the K fault feature deviation values ​​Δu obtained in step 2 yx1 , Δu yx2 , Δu yx3 , ..., Δu yx(i+1) , Δu yxi , and arbitrarily select 75% of the K fault feature deviation values ​​as the training set, and the remaining 25% as the test set, to establish a GM diagnosis model for the faulty power generation module based on voltage deviation signals and Deep Extreme Learning Machine (DELM); The input voltage fault characteristic deviation value sample is (r yxi ,t yxi ), r yxi =[Δu yxi1 ,Δu yxi2 ,...,Δu yxK ] T , t yxi =[t yx1 ,t yx2 ,...,t yxm ] T r yxi ∈R n , t yxi ∈R m ; Among them, r yxi is the sample data with dimension K, t yxi is the expected output of dimension m; in the deep extreme learning machine, the weight w i and threshold b i Randomly generated, the output of the network structure is expressed as follows Where: L is the number of hidden layer nodes; w j =[w j1 ,w j2 ,...,w jK ] T is the weight vector connecting the input node and the jth hidden layer node; g is the activation function; β j =[β j1 ,β j2 ,...,β jm ] T is the weight vector between the output node and the jth hidden layer node; b j is the bias of the jth hidden layer node; then Get the unique least squares solution of β As shown below: In the formula, G represents the k×l-dimensional hidden layer output matrix, G is a singular matrix; G + is the generalized inverse matrix of G; G T is a singular matrix; is the least squares solution of β; The input voltage fault feature data sample r is used as the target output r1 of the first extreme learning autoencoder, that is, r = r1, and the output weight β1 is obtained; then the output matrix G1 of the extreme learning machine is used as the input and target output of the next extreme learning autoencoder, that is, r = r2, and so on, to obtain the output weight β of the last hidden layer. i+1 ; Step 4: Use the Seagull algorithm to calculate the weight w of the power generation module fault diagnosis model based on the voltage deviation signal and deep extreme learning machine built in steps 2 and 3. i and threshold b i The parameters are optimized to obtain an improved power generation module fault diagnosis optimization model with higher accuracy; Step 5: Use the test set data to classify the faults of the power generation module.

Images