Inverter multi-tube fault diagnosis method of simplest feature set neural network
Through state feature extraction and rough set theory screening feature sets, combined with neural network for fault diagnosis, the problem of complex operation and slow speed in power electronic converter fault diagnosis is solved, and the three-phase T-type grid-connected inverter switch tube is realized Accurate location and quick identification of faults.
Patent Information
- Application Number
- CN202411198878.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-08-23
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-03
AI Technical Summary
In the fault diagnosis of power electronic converters, the model driving method has high accuracy but complex operation, and the data driving method has a slow diagnosis speed, making it difficult to achieve simple, easy to operate, fast and accurate fault diagnosis.
The minimum feature set is screened through state feature extraction and rough set theory, combined with neural networks for fault identification, and circuit topology analysis and BP neural network for training to identify switch tube faults.
It realizes accurate positioning and rapid identification of the switching tube faults of three-phase T-type grid-connected inverter switch tubes, simplifies the operation process, and improves the diagnosis speed and accuracy.
Smart Images

Figure CN119066470A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of power electronic power conversion, and in particular relates to a method for diagnosing open-circuit faults of switch tubes of a three-phase T-type grid-connected inverter by combining state feature extraction with a neural network. Background Art
[0002] At present, three-phase pulse width modulation (PWM) converters are widely used in motor drives, electric or hybrid electric vehicles, renewable energy conversion systems, etc. T-type three-level converters do not have the problem of shutdown and reordering, and are very convenient to control. In addition, compared with the neutral point clamped (NPC) three-level converter, the T-type three-level converter uses fewer components, which means lower cost, lower conduction loss power, and lower isolation drive power. Based on the above advantages, the T-type three-level converter has received widespread attention and has been used in various important occasions.
[0003] Among the converter failures, the most serious one is the switch failure. According to statistics, more than 30% of converter failures are caused by semiconductor power devices and their corresponding gate driver electronic circuits. Figure 1 As shown in Figure 2, among the converter failures, failures caused by semiconductors and solder joints account for 34% of the total converter failures. Therefore, it is very important to diagnose the converter failure in a timely manner.
[0004] Current research on fault diagnosis of power electronic converters mainly focuses on model-driven methods and data-driven methods.
[0005] In terms of model-driven fault diagnosis, fault diagnosis is mainly based on the changes of state characteristic signals such as current and voltage under fault conditions. There are diagnostic methods such as using instantaneous conversion power to detect switch faults in three-level active neutral point clamped (ANPC) converters, using electrical signals and their harmonic components to detect, classify and locate converter faults, and using input PWM and output current and output voltage signals to detect open-circuit faults in multi-level inverters that fly-through capacitors. The accuracy of the model-driven method is relatively high, and the diagnostic speed is also very fast, but the actual operation requirements are strict and high, and the formulation of diagnostic rules is relatively cumbersome.
[0006] In terms of data-driven fault diagnosis, there are methods such as probabilistic neural networks, deep neural networks, the combination of wavelet transform and neural networks, particle swarm optimization algorithms (PSO-BP neural networks) and the combination of neural networks and expert systems, or improved convolutional neural networks with global average pooling layer models. It would be very simple and easy to operate to diagnose faults only by relying on data. However, a large amount of data will consume a lot of time, which will slow down the diagnosis speed. Therefore, it is considered to extract effective state feature variables according to the state changes under fault conditions, and then combine neural networks for fault location and identification, so as to make the diagnosis method simple and easy to operate while having high diagnostic accuracy and fast diagnosis speed. Summary of the invention
[0007] The purpose of the present invention is to provide a method for diagnosing the open circuit fault of the switch tube of a three-phase T-type grid-connected inverter based on the combination of state feature extraction and neural network: first, the abnormal transition of the operating state of the T-type inverter is analyzed according to the circuit topology, and then, the state variables related to the state transition are optimized and screened by rough set theory (RST), so as to obtain the minimum feature set required to reflect various fault modes, and finally, the obtained minimum feature set is used as the input and output of the neural network for training and identification, so as to realize the accurate positioning of the fault point.
[0008] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme.
[0009] The three-phase T-type three-level inverter topology is as follows Figure 2 As shown in the figure, it can be roughly considered to consist of three parts: DC bus capacitor, T-type three-level converter and LCL filter.
[0010] T-type three-level converter. It consists of 12 power devices, T xn (x=a,b,c;n=1,2,3) is the power switch tube. Among them, T x1 and T x4 The vertical bridge arm is connected to the positive and negative busbars, T x2 and T x3 The horizontal bridge arm and the DC side capacitor midpoint O and the bridge arm output end point v x Connected. xn Represents the power device and its conduction and switching signals, where x represents the phase (a, b or c), n represents the power device number (1, 2, 3 or 4), S xn =1 means the power device is conducting, and S xn =0 means the power device is turned off.
[0011] from Figure 2 It can be seen that each phase of the T-type three-level converter contains four power devices, namely the vertical power device (Sx1 and S x4 ) and horizontal power devices ((S x2 and S x3 ), which together form the horizontal and vertical bridge arms of the T-type three-level converter. The existence of the horizontal bridge arm allows the converter to output zero level. In order to facilitate the understanding of the working principle of the T-type three-level converter, the following definitions are made.
[0012] Define the state function S x As shown in formula (1):
[0013] The state function indicates that the operating state of each phase of the inverter corresponds to the switch state [P], [O], [N]. The switch state [P], [O], [N] corresponds to the output positive level (0.5V dc ), zero level (0) and negative level (-0.5V dc ) voltage. According to the state function, each bridge arm output terminal v x The output voltage V between the DC side capacitor midpoint O xo , as shown in formula (2):
[0014] Since the three phases of the inverter are symmetrical, the following analysis will take phase A as an example. Six operating states are defined according to the current flow path during the operation of the inverter as shown in Table 1. Table 1 Operation status of T-type three-level inverter
[0015] When the anti-parallel diode fails to open, it will cause the switch tube to overvoltage. At present, the driving circuit of the IGBT module often integrates short-circuit and overvoltage protection circuits. When the anti-parallel diode fails to open, the overvoltage protection circuit will be triggered. Therefore, this article will only consider the impact of the power switch tube failure, and will not analyze and discuss the states V and VI.
[0016] Next, we will analyze the operating state of the T-type three-level inverter when a power switch failure occurs, and summarize the changes in the operating state when different failures occur. Figure 3 As shown in Figure 2, the failure of the power switch tube will cause the current path to change, resulting in the bridge arm output voltage V xo Taking the operating state I as an example, the current path in normal operation is P—T a1 —v a , the switch state is [P], the output voltage is 0.5V dc .
[0017] When T a1 After an open circuit fault occurs, if Figure 3 As shown in (a), the current will flow through T a3 and diode D a2 Continue current, at this time the current path becomes O-D a2 —T a3 —v a , the switch state becomes [O], and the output voltage becomes 0. a1 and T a3 When a fault occurs at the same time, Figure 3 As shown in (b), the current can only flow through diode D a4 Continue current, the current path becomes N-D a4 —v a , the switch state changes to [N], and the output level changes to negative level -0.5V dc .
[0018] Through the above analysis, it can be seen that when a power switch fails, the current path and operating state will change. Different failures will cause different changes. Therefore, the switch failure can be used as a trigger condition for state change. Define the switch failure function F TAX (X=1,2,3,4), when F TAX =1 indicates that the switch tube has an open circuit fault. TAX = 0 indicates that the switch tube is normal. Similarly, the state changes of other working conditions when the corresponding switch tube fails are analyzed and listed in a table, as shown in Table 2. Table 2 Operation status changes of T-type three-level inverter under different trigger conditions
[0019] It can be seen from Table 2 that when a switch tube at different positions fails, it may cause different state changes and path changes. Therefore, it is possible to detect the location of the fault by reversely detecting the change of state and path.
[0020] In order to express the operating transfer conditions in Table 2 with more easily accessible electrical signals, the switch signal S xn and the bridge arm output terminal to midpoint voltage V xo They represent the operating status before and after the transfer respectively. The current path and current direction use the current value i of each branch. PA 、i OA and i NA Then the A phase switch failure can be expressed by formula (3): (F TA1 ,F TA2 ,FTA3 ,F TA4 )=f(V AO ,i PA ,i OA ,i NA ,S A1 ,S A2 ,S A3 ,S A4 ) (3) Wherein, f(·) is the fault indication function of each power switch of phase A of the inverter.
[0021] Then, the rough set theory is used to process and simplify the relevant information in the above formula to mine the simplest and effective information that can reveal the fault characteristics of the power switch, and then simplify the fault diagnosis strategy. In the rough set theory, the information contained in the research object is called a knowledge expression system, which is expressed by formula (4). H S =(U,At,VA,f(·)) (4) Among them, H s is the knowledge expression system under study; U is a non-empty finite set of research objects, also called the domain of the system; At represents the attribute set of the research objects in the domain, which can be divided into the conditional attribute set CA and the decision attribute set DA; VA represents the set of attribute value domains contained in At; f(·) is the information function, which is used to represent the mapping from U→VA.
[0022] Replace the information in Table 2 with electrical signals, and according to the rough set theory, the information table of the T-type inverter operating state transfer can be obtained as shown in Table 3: Table 3 Information table of T-type inverter operation status transfer
[0023] In order to analyze and extract the characteristics of the voltage signal in the table, the upper bridge arm switch tube voltage V Ta1 Replace the bridge arm output terminal to midpoint voltage V xo , to ensure that the voltage value is always greater than 0. Then, according to the rough set theory, the discernibility matrix of the knowledge expression system is obtained as shown in Table 4, and the attributes are simplified to extract the simplest attribute set. Table 4. Discrimination matrix of knowledge representation system
[0024] The knowledge matrix reflects the distinguishing and comparing relationship between any working conditions. For example, the U(1) state and the U(1) state are the same working condition and cannot be distinguished, so the distinguishing characteristic variable between them is an empty set φ. For example, the characteristic variable I between the U(1) state and the U(4) state is PA , I OA , V Ta1 All of them are different, so as long as one of these three exists in the final characteristic variable set, U(1) normal state and U(4)T A1 The fault status is distinguished. The distinction between other working conditions is similar.
[0025] According to the rough set theory, the simplest characteristic variables of the knowledge matrix are extracted and selected. In addition, since the current will be affected by load changes, etc., it is best to avoid selecting the current characteristic quantity as the final simplest characteristic variable set during the selection process. According to the above principles, the simplest characteristic variable set finally selected is S A1 , S A3 , V Ta1 After obtaining the simplest set of feature variables, the neural network can be used for training and recognition. Figure 4 Flowchart of the inverter multi-tube fault diagnosis method based on the simplest feature set neural network.
[0026] Based on the results of the above state feature extraction, it can be determined that S A1 , S A3 , and V Ta1 The fault can be located and identified by using it as the input feature of the neural network.
[0027] Due to the application of the above technical solution, the present invention has the following characteristics: 1. Based on circuit topology state transition and rough set theory, the present invention extracts the simplest feature variable set as the input of the neural network, which greatly reduces the amount of calculation driven by pure data. 2. In order to solve the classification difficulty problem of fault identification based on characteristic variables, the present invention uses a method combining state feature extraction with neural network, which greatly improves the feasibility of operation and the rapidity of identification. 3. The present invention realizes accurate and rapid identification of fault locations of single and multiple switch tubes of the inverter, and is simple and feasible to implement. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 : The proportion diagram of different fault types of the inverter in the present invention;
[0029] Figure 2 : Topological structure diagram of the three-phase T-type three-level inverter in the present invention;
[0030] Figure 3: A diagram showing the current path variation of a T-type three-level converter under different fault conditions in the present invention;
[0031] Figure 4 : Flow chart of the inverter multi-tube fault diagnosis method of the simplest feature set neural network in the present invention;
[0032] Figure 5 :A general back propagation (BP) neural network structure diagram in the present invention;
[0033] Figure 6 : The specific BP neural network structure diagram in the present invention;
[0034] Figure 7 : The bridge arm fault identification training result diagram in the present invention; DETAILED DESCRIPTION
[0035] The technical solution will be clearly and completely described below in conjunction with the preferred examples and drawings of the present invention. It should be understood that the preferred examples are only for illustrating the present invention, not for limiting the scope of protection of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.
[0036] The present invention provides a method for diagnosing open-circuit faults of switch tubes of three-phase T-type grid-connected inverters based on state feature extraction combined with neural networks. Firstly, the state changes are analyzed based on the circuit topology structure and the simplest feature variable set is extracted according to the rough set theory to determine the neural network input. Then, the BP neural network is used for network training, thereby realizing accurate and rapid identification of the fault point of the switch tube.
[0037] A general BP neural network structure is as follows Figure 5 As shown, the specific BP neural network structure constructed in this scheme is as follows Figure 6 As shown in the figure, the operation and implementation process of the fault identification scheme is as follows:
[0038] Input a shutdown signal to a certain switch tube, record the switch tube number at the same time, and collect the S A1 , S A3 , and V Ta1 data.
[0039] Replace the switch tube, input the shutdown signal and repeat the above operation until all the switch tubes and the multi-tube faults of the switch tube combination are traversed.
[0040] For the bridge arm fault, each phase contains four single-tube faults and two multi-tube faults, plus a set of data sets when all switch tubes are normal, a total of 7 sets of input data sets are collected.
[0041] Among them, the input quantity determined according to the simplest feature set obtained in the previous article can be obtained through the transportation of the collected data set, and the output feature value is the label of the faulty switch tube corresponding to the data set.
[0042] The determined BP neural network input and output are shown in Table 5: Table 5 Input and output of BP neural network In the table, F TAn (n=1, 2, 3, 4) are the output parameters set. TAn =1, it indicates that the corresponding inverter No. n is in a fault state at this time.
[0043] For the collected data, only 80% of them are used as training data for the BP neural network, and the remaining 20% of the data will be used to verify the effectiveness of the designed BP neural network. Simulation Results
[0044] In order to verify the correctness of the proposed method, the present invention has carried out simulation verification. The simulation parameters of the grid-connected inverter are shown in Table 6. Table 6 Simulation parameters
[0045] Build a simulation module in MATLAB, set parameters according to Table 6, run the simulation and collect data sets.
[0046] The collected data is used to input the neural network for training. The results of bridge arm fault identification training are shown in the figure below. Figure 7 As shown in the figure, it can be seen that the final recognition accuracy is 100%.
[0047] In summary, the proposed fault identification method based on state feature extraction combined with neural network can make a quick and accurate judgment on the fault location of the grid-connected inverter switch tube when a fault occurs. Compared with the traditional model-driven or data-driven diagnosis method based on feature variable identification, it not only improves the feasibility of operation, but also greatly reduces the amount of data calculation and improves the rapidity of diagnosis.
[0048] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. The general principles defined in this article can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in this article, but should conform to the widest scope consistent with the principles and novel features disclosed in this article. All equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for inverter multi-tube fault diagnosis based on a minimal feature set neural network. It is characterized in that The method comprises: In the simulation analysis of the normal and faulty operating conditions of the T-type three-level inverter, the characteristic variables are extracted according to the changes in the operating status, and then the rough set theory is used to screen and optimize these characteristic variables to obtain the simplest characteristic variable set. Finally, the back propagation (BP) neural network is used to train the acquired data to realize the identification of the fault location of the switch tube, provide information for subsequent repairs, and ensure the stable operation of the power grid; First, in order to obtain the characteristic variables related to the state change of the T-type three-level inverter, according to the normal and fault operating states, the changes in the current path under different operating conditions are analyzed. The state transition simulation is used to extract the relevant state characteristic variables, including the switch signal, current signal and bridge arm voltage signal. Therefore, the state can be expressed by the following formula: And establish the corresponding knowledge expression system matrix, the specific construction formula is as follows: (F TA1 ,F TA2 ,F TA3 ,F TA4 )=f(V AO ,i PA ,i OA ,i NA ,S A1 ,S A2 ,S A3 ,S A4 ) (1) Among them: F TAX (X=1,2,3,4) represents the switch failure function. When F TAX =1 indicates that the switch tube has an open circuit fault. TAX = 0 indicates that the switch is normal, V AO represents the bridge arm voltage, i XA (X=P,O,N,) represents the current on the three branches of phase A, S AX (X=1, 2, 3, 4) indicates that the switching signal f(·) of the four switching tubes of phase A is a fault indication function of each power switch of phase A of the inverter. And establish the corresponding knowledge expression system matrix, the specific construction formula is as follows: H S =(U,At,VA,f(·)) (2) Among them, H s is the knowledge expression system under study; U is a non-empty finite set of research objects, also called the domain of the system; At represents the attribute set of the research objects in the domain, which can be divided into the conditional attribute set CA and the decision attribute set DA; VA represents the set of attribute value domains contained in At; f(·) is the information function, which is used to represent the mapping from U→VA. The switch signal S of the discernible simplest feature set in the knowledge expression system matrix is selected by the rough set theory method. A1 , S A3 , and the upper bridge arm voltage signal V Ta1 As the input feature variable of BP neural network; The neural network is trained by extracting characteristic variable data, and then the trained neural network is used to monitor the operating status of the T-type three-level inverter in real time, and a fault signal is output in time when a fault occurs.
2. The method according to claim 1, characterized in that Methods to improve the speed and accuracy of inverter multi-transistor fault diagnosis include: The simplest feature variable set obtained by screening is used to obtain the corresponding variable data set from the simulation, input into the BP network and give the corresponding output. The network model is improved by adjusting the hidden layer weights through continuous training and learning. The relationship between the output and weight of each layer is as follows: in, represents the output of layer L, f represents the activation function, Represents the calculation weight coefficient from layer l-1 to layer l, Indicates the corresponding bias coefficient. Finally, a four-layer feedforward BP neural network with 3 inputs and 7 outputs is constructed. The data of the T-type three-level inverter operation process is input in real time into the trained network, and the real-time output can be quickly used for fault diagnosis.
3. The method according to claim 1, wherein the inverter multi-tube fault diagnosis method is characterized by: First, the operating conditions of the T-type three-level inverter are analyzed, and the simplest feature set is extracted using the rough set theory to reduce the amount of data required for diagnosis. Then, the BP neural network is used to classify and identify the obtained feature variables, thereby improving the accuracy and speed of classification and identification.
Citation Information
Patent Citations
Fault diagnosis method and device for cascaded three-level inverter
CN110208721A
Digital-analog hybrid driven T-type three-level inverter open-circuit fault diagnosis method
CN116774030A
