Fault diagnosis method for DC distribution network based on Bayesian network information fusion

Abstract

A DC distribution network fault diagnosis method based on Bayesian network information fusion belongs to the technical field of electric DC distribution network. The purpose of the present invention is to improve the traditional Bayesian network-based power grid fault diagnosis model based on Bayesian network information fusion to address the commonalities and differences between DC distribution network and AC distribution network faults. The steps of the present invention are: Bayesian network information fusion, Bayesian network node conditional probability assignment, and DC distribution network fault diagnosis. The present invention improves the traditional Bayesian network-based power grid fault diagnosis model to realize DC distribution network fault diagnosis, and verifies the reliability of the method through actual examples.

Description

Technical Field

[0001] The invention belongs to the technical field of electric power direct current distribution network. Background Art

[0002] With the advancement of power semiconductor and power electronics technologies, and the widespread adoption of DC loads, DC distribution networks, with their strong transmission capacity, low cost, and efficient and reliable access to distributed energy generation and storage units, have become a research hotspot both domestically and internationally. When a DC distribution network fault occurs, the fault current rises rapidly and has a high amplitude, seriously threatening the safe and reliable operation of converters and other power electronic devices. Researching fast and effective DC distribution network fault diagnosis methods is crucial for rapidly restoring power and ensuring safe and stable grid operation.

[0003] Power grid fault diagnosis involves diagnosing suspected components using specific protection and circuit breaker action data, taking into account other useful information, while also diagnosing information about circuit breakers and protection devices. Accurately identifying faulty components is extremely valuable for restoring power and maintaining operation. Currently, typical methods for power grid fault diagnosis, both domestically and internationally, include Petri nets, artificial neural networks, expert systems, rough set theory, analytical models, multi-source information fusion, and Bayesian networks. Due to the "refusal to operate" and "maloperation" of protection and circuit breakers, the relay protection information received by the dispatch center is not entirely accurate, significantly complicating fault diagnosis.

[0004] This approach assesses the reliability of the protection and circuit breaker status and timing of each action in the power grid, calculates the reliability of each action, and incorporates it into the Bayesian inference formula. This reduces the impact of false alarms, missed alarms, malfunctions, or refusal to operate, as well as timing distortions, on the diagnostic results, thus enabling grid fault diagnosis. However, this reliability assessment method does not meet the requirements for relay protection speed in DC distribution networks and is therefore unusable in DC distribution networks.

[0005] A Bayesian network was constructed using information from the actual power grid topology and relay protection device operation data. Expert system rules were developed to identify protection failures and malfunctions, and the sequence of action of each layer of protection devices and circuit breakers was deduced when a fault occurred, enabling the simulation of complex power grid faults. The Bayesian network structure of this Bayesian network inference method results in a minimal impact of circuit breaker information on component fault diagnosis, making it prone to misjudgment when a circuit breaker fails to operate for some reason.

[0006] Currently, most power grid fault diagnosis methods are designed for AC distribution networks. DC and AC distribution networks share both similarities and differences when it comes to faults. The similarity lies in the same logic for relay protection operation when a DC or AC fault occurs. However, the differences lie in the fact that the fault characteristics and mechanisms of DC distribution networks differ from those of AC networks. Summary of the Invention

[0007] The purpose of the present invention is to improve the traditional Bayesian network-based power grid fault diagnosis model based on Bayesian network information fusion to address the similarities and differences between DC distribution network and AC distribution network faults.

[0008] The steps of the present invention are:

[0009] S1. Information fusion of Bayesian networks

[0010] Assume that there are two pieces of evidence E1 and E2 under the identification framework Θ, the basic probability assignment functions are M1 and M2, and the states corresponding to the evidence are represented by A i and B j To express; Dempster's synthesis rule is defined as:

[0011] is called the regularization parameter; if m is used to represent the probability assignment function, then the orthogonal sum of the two assignment functions m1 and m2 is recorded as When k=1, m1 and m2 are considered contradictory, and the combination rule cannot be used to synthesize the basic probability assignment. When k≠1, the synthesis formula is valid, and m determines a basic probability assignment. When there are multiple pieces of evidence to be integrated, the above combination formula can be expanded to obtain the result.

[0012] The three component failure probability information is combined with the DS evidence theory synthesis rule, and the fusion formula is as follows:

[0013]

[0014] Where P P 、P CB and P L represent the probability values ​​obtained under the Bayesian network model for protection information, circuit breaker information, and emergency fault current limiting strategy information, respectively; X represents the parent node corresponding to each Bayesian network, that is, the component; the numbers "1" and "0" represent the component status, that is, the number "1" indicates that the component is faulty, and the number "0" indicates that the component is not faulty;

[0015] S2. Conditional probability assignment of Bayesian network nodes

[0016] Assume that the probability that the protection and circuit breakers in the DC distribution network do not refuse to operate or malfunction is 90% of the probability that the protection and circuit breakers in the AC distribution network do not refuse to operate or malfunction.

[0017] S3. DC distribution network fault diagnosis

[0018] a. Determine the faulty components in the DC distribution network

[0019] Let the failure probability of each component after reverse reasoning be a n , assuming the line fault threshold is 0.8, the bus fault threshold is 0.6, that is, if a n ≥ the fault threshold of the corresponding component, the component is determined to be a faulty component;

[0020] b. Relay protection device and circuit breaker operation

[0021] Calculate the expected operation probability of each protection device node and circuit breaker node; the specific method is as follows:

[0022] (1) Find the protection information Bayesian network model and the circuit breaker information Bayesian network model corresponding to the faulty component, and set the parent node in these two Bayesian network models to the state "1", that is, the component has failed;

[0023] (2) Perform forward inference of the Bayesian algorithm on the protection information Bayesian network model and the circuit breaker information Bayesian network model to calculate the expected action of the corresponding protection and circuit breaker in the event of a fault;

[0024] (3) Subtract the expected action probability of the protection device and circuit breaker obtained by forward reasoning of the Bayesian algorithm from the actual action situation E of the protection device and circuit breaker. The actual situation is obtained from the SCADA system, that is, the information of the protection device and circuit breaker is "1" if it is in action and "0" if it is not in action. This is used as the action judgment of the protection device and circuit breaker; that is, the following formula is used:

[0025]

[0026] Where a∈(0,1), b∈(0,1) and a+b=1. Taking into account the calculation results of other relevant literature and this paper, a and b are both reasonably set to 0.5.

[0027] The present invention obtains the difference between the far backup protection and near backup protection action behaviors and the actual action behaviors based on the Bayesian forward reasoning results:

[0028] (1) For misjudgment of remote backup protection action behavior, by finding Bayesian forward reasoning, we can preliminarily obtain the refusal circuit breaker in the action behavior analysis of the protection action device, find the remote backup protection and circuit breaker associated with the refusal circuit breaker, and modify the expected probability of their nodes;

[0029] (2) When the near backup protection action is misjudged, the Bayesian network is used to find the main protection element on the same side of the component. Under the premise of the main protection action, the expected action probability of the near backup protection action on the same side is modified. The expected action probability of the above two cases is added by 1.

[0030] The present invention improves the traditional Bayesian network-based power grid fault diagnosis model, divides relay protection information into two categories: protection action information and circuit breaker action information, constructs Bayesian network models according to their respective action logics, introduces a DC distribution network fault emergency control strategy, integrates the above information using DS evidence fusion theory, combines Bayesian forward and backward reasoning to obtain fault diagnosis and action behavior analysis of protection devices and circuit breakers, realizes DC distribution network fault diagnosis, and verifies the reliability of the method through actual examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a DC distribution network with IEEE 33-bus bus;

[0032] Figure 2 Bayesian network model for relay protection of DC distribution network;

[0033] Figure 3 To protect information corresponding to the Bayesian network model;

[0034] Figure 4 The Bayesian network model corresponding to the circuit breaker information;

[0035] Figure 5 The Bayesian network model corresponding to the current limiting strategy information;

[0036] Figure 6 It is the relay protection action process under the current limiting strategy;

[0037] Figure 7 For line L 6-7 Bayesian network model of protection information;

[0038] Figure 8 For line L 6-7 Bayesian network model of circuit breaker information;

[0039] Figure 9 For line L 6-7 The Bayesian network model of current limiting strategy information;

[0040] Figure 10 This is a comparison chart of the calculation results of the traditional method and the method of the present invention under scenario 2;

[0041] Figure 11 This is the behavior criterion diagram of relay protection and circuit breaker in scenario 2;

[0042] Figure 12 This is a comparison chart of the calculation results of the traditional method and the method of the present invention under scenario 3;

[0043] Figure 13 This is the behavior criterion diagram of relay protection and circuit breaker under scenario 3;

[0044] Figure 14 Comparison of calculation results between the traditional method and the method of the present invention under scenario 5

[0045] Figure 15 This is the behavior criterion diagram of relay protection and circuit breaker under scenario 5;

[0046] Figure 16 A comparison chart of the calculation results of the traditional method and the method of the present invention under scenario 6;

[0047] Figure 17 This is the behavior criterion diagram of relay protection and circuit breaker under scenario 6. DETAILED DESCRIPTION

[0048] The present invention is described in detail below with reference to the accompanying drawings:

[0049] 1 Bayesian network model under DC distribution network

[0050] 1.1 DC distribution network protection configuration and its operating principle

[0051] 1.1.1 DC Distribution Network Protection Configuration Based on the topology of the DC distribution network and the location of the converter, the protection of the DC distribution network can be divided into four areas: AC side protection, converter protection, load area protection, and DC side protection. Corresponding protection strategies can be configured for each area.

[0052] The goal of DC distribution network protection configuration is to prevent damage to equipment within the DC system and faults that could endanger the operation of the entire DC distribution network. Factors such as converter operation and control methods, and the direction of power transmission must be considered during configuration. For the AC system area, transformer overcurrent protection and busbar overvoltage / undervoltage protection are connected. For the AC / DC converter area, arm overcurrent protection, arm differential protection, and valve overcurrent protection are configured. For the load area, overcurrent protection, overvoltage protection, and converter interface protection are configured. For the DC line bus area on the DC side, voltage imbalance protection, low-voltage overcurrent protection, combined low-current and voltage protection, and differential protection are configured.

[0053] 1.1.2 Principle of DC Distribution Network Protection Operation This invention mainly analyzes the lines and buses in the DC distribution network. The following describes the operating principles of line and bus protection.

[0054] Figure 1This is a DC distribution network scenario with IEEE 33-bus busbars, where the block diagram is a local system diagram and configuration diagram of busbar B6 and its associated relay protection devices in the DC distribution network. The meanings of the letters are shown in Table 1. Figure 1 Middle Line L 5-6 Left end protection L 5-6 Lm, L 5-6 Lp and L 5-6 Ls is used to analyze the principle of line protection operation, where L 5-6 Lm is the main protection, which only protects the line itself; L 5-6 Lp is the near backup protection, which also protects the entire length of the line and is the main protection L 5-6 Lm backup protection; L 5-6 Ls is the remote backup protection, which is usually activated when the adjacent component fails. When the protection does not operate when the adjacent component (busbar B6) fails, L 5-6 Ls is the backup protection action to clear the fault. After the three protection actions, the circuit breaker CB is triggered. 5-6 Tripping.

[0055] For busbar protection, the main protection only protects the busbar itself. The line backup protection is its backup protection. Figure 1 B6m is the busbar protection. When busbar B6 fails, busbar protection B6m will trigger the circuit breaker CB. 6-5 , CB 6-7 and CB 6-25 Tripping.

[0056] Table 1 Protection configuration description

[0057]

[0058] 1.2 Bayesian Network Principle Bayesian network is a directed acyclic graph based on a network structure. Each node in the network represents a variable, the directed arcs represent the relationship between the variables, and the conditional probability between the node and its child nodes represents the dependency relationship between the variables. The mathematical description is: if the domain X = {x1, x2, ..., x n}, where x1, x2, …, x n For each node in the network, the joint probability of multiple nodes occurring is p(x1,x2,…,x n )for:

[0059]

[0060] Where π(x i ) represents x i The parent node collection.

[0061] 1.3 Bayesian Network Model for DC Distribution Network Relay Protection In DC distribution network relay protection, there is not only a temporal relationship between component (busbar or busbar) faults, protection activation, and circuit breaker operation, but also a certain logical relationship. Typically, DC distribution network protection activates after a component fault occurs, which in turn drives the circuit breaker to isolate the fault.

[0062] Considering that protection and circuit breaker failures and malfunctions often occur in DC distribution networks, fault isolation through relay protection can be divided into the following three situations:

[0063] Case 1: A DC distribution network component fails, and the primary protection (nearby backup protection) trips the near-end circuit breaker.

[0064] The second scenario: A DC distribution network component fails, the primary protection (local backup protection) operates, the local circuit breaker refuses to operate, and the remote backup protection operates to trip the remote circuit breaker.

[0065] The third situation: A DC distribution network component fails, the main protection (nearby backup protection) refuses to operate, and the far backup protection operates to trip the remote circuit breaker.

[0066] Based on the above three relay protection action logics, the established relay protection Bayesian network model is as follows: Figure 2 As shown in the figure, each node is connected to each other according to the relay protection action logic, with the connection order being: "element - primary protection - local backup protection - local circuit breaker - remote backup protection - remote circuit breaker." In the Bayesian network model, the numbers "0" and "1" represent the status of each node: for element nodes, the number "0" indicates that the element is in the "normal" working state, and the number "1" indicates that the element is in the "fault" state. For protection nodes and circuit breaker nodes, the number "0" indicates that the protection (circuit breaker) is in the "inoperative" state, and the number "1" indicates that the protection (circuit breaker) is in the "operative" state.

[0067] This traditional network model structure relies heavily on protection nodes. Since only protection nodes are connected to component nodes, and circuit breaker nodes serve only as connections, circuit breaker information has little influence on component fault diagnosis. This model structure maintains a high probability of component failure even when protection nodes activate but the circuit breaker does not, making misdiagnosis a possibility. To enhance the role of circuit breaker information in component fault diagnosis, the traditional Bayesian network model structure has been improved.

[0068] 2 DC distribution network fault diagnosis based on Bayesian network information fusion

[0069] 2.1 Improved Bayesian Network Model

[0070] The information obtained in traditional SCADA systems can be divided into two categories: one is protection action information, and the other is circuit breaker action information. Therefore, Bayesian networks can be established for these two types of information respectively.

[0071] 2.2.1 Protection Information Bayesian Network Model

[0072] The Bayesian network structure corresponding to the protection action information is "element-main protection-near backup protection-remote backup protection". The Bayesian network model structure is as follows: Figure 3 shown.

[0073] 2.2.2 Circuit Breaker Information Bayesian Network Model

[0074] The Bayesian network structure corresponding to the circuit breaker action information is "element-near-end circuit breaker (acted by the main protection and the near backup protection)-remote circuit breaker (acted by the remote backup protection)", and its Bayesian network model structure is as follows: Figure 4 shown.

[0075] 2.2 Introduction of the Bayesian Network Model for Current Limiting Strategies Currently, there are two main strategies for limiting current in specific DC distribution networks. One is to implement an emergency current limiting control strategy for the converter. The basic idea is: after a DC fault occurs, the number of submodules in the converter's MMC bridge arm is adaptively changed according to the DC voltage at the DC fault point to limit the growth of the DC fault current. The other is to add a fault current limiter to suppress the fault current. When the DC distribution network is operating normally, the fault current limiter is not put into use due to the mechanical bypass switch. When a fault occurs in the DC distribution network, the fault current limiter is put into use by controlling the coordination of the semiconductor switches (IGBTs) in the fault current limiter, thereby suppressing the fault current.

[0076] Based on this, a Bayesian network model for emergency fault current limiting strategy information is established. Figure 5 As shown in the figure, the parent node is the corresponding component in the DC distribution network, and its child nodes correspond to two types of current limiting strategies. For the fault current limiter, the number "0" indicates that it is not in use, and the number "1" indicates that it is in use. For the converter current limiting control, the number "0" indicates that the converter current limiting control mode is not activated, and the number "1" indicates that the converter current limiting control mode is activated.

[0077] The overall logic of the Bayesian network model for current limiting strategy information is as follows: when a component in the DC distribution network fails, the network implements a current limiting strategy to suppress the fault current. This fault current suppression serves two main purposes: first, it significantly reduces the interrupting current of the DC circuit breaker, lowering the cost of the DC circuit breaker; second, it prevents converter lockup, which could cause a temporary outage in the DC distribution network.

[0078] Figure 6The DC fault clearing process, achieved by the coordinated use of a current limiting strategy and traditional relay protection, begins after a certain delay after the fault occurs. The converter pole controller or fault current limiter detects the fault and switches the corresponding converter to emergency current limiting control mode, adjusting the number of submodules in each bridge arm. The current-limiting reactors are then connected to the grid through a control strategy. Once the converter or fault current limiter detects that the DC fault has been cleared by the DCCB, the corresponding converter immediately switches from emergency current limiting control to normal control mode. The current-limiting reactors are then disconnected from the grid via a mechanical bypass switch, restoring normal DC grid operation. The flow chart shows that the current limiting control strategy operates in parallel with traditional relay protection, meaning that the fault emergency current limiting control strategy does not affect normal relay protection operation.

[0079] 2.3 Information Fusion in Bayesian Networks

[0080] The comprehensive analysis of the component failure probabilities obtained under the above three types of Bayesian network models is converted into a probability, and the DS evidence theory is needed at this time.

[0081] The Dempster-Shafer theory of evidence, also known as the DS theory of evidence or evidence theory, is centered around the Dempster combination rule, which allows for the fusion of independent bodies of evidence. This rule defines a basic probability assignment function, which can then be used to fuse different bodies of evidence using the combination rule.

[0082] Define the identification framework Θ as a series of complete and mutually exclusive discriminant hypotheses, representing the set of all possible propositions for a problem. At any time, the answer to the problem can only be an element of Θ. Assume that there are two pieces of evidence E1 and E2 under the identification framework Θ, and the basic probability assignment functions are M1 and M2. The states corresponding to the evidence are represented by A and M2 respectively. i and B j To express.

[0083] set up Dempster's synthesis rule is defined as:

[0084]

[0085] The above combination formula reflects the relationship between two pieces of evidence. It is called the regularization parameter. If m is used to represent the probability assignment function, then the orthogonal sum of the two assignment functions m1 and m2 is recorded as When k=1, m1 and m2 are considered contradictory, and the basic probability assignment cannot be synthesized using the combination rule. When k≠1, the synthesis formula is valid, and m determines a basic probability assignment. When there are multiple pieces of evidence (more than two) to be fused, the above combination formula can be expanded to obtain it.

[0086] The Bayesian network model for protection information, the Bayesian network model for circuit breaker information, and the Bayesian network model for emergency fault current limiting strategy information correspond to three independent bodies of evidence in DS evidence theory. Each body of evidence will generate corresponding fault probability information. The fusion formula of these three component fault probability information is combined with the DS evidence theory synthesis rules as shown below:

[0087]

[0088] Where P P 、P CB and P L They represent the probability values ​​obtained under the Bayesian network model for protection information, circuit breaker information, and emergency fault current limiting strategy information; X represents the parent node corresponding to each Bayesian network, that is, the component (bus or line); the numbers "1" and "0" represent the component status, that is, the number "1" indicates that the component is faulty, and the number "0" indicates that the component is faulty. For example, P CB (X=1) represents the probability of component failure under the Bayesian network corresponding to the circuit breaker information, P P P(X=0) represents the probability that the component has not failed under the Bayesian network corresponding to the protection information, and so on. P(X=1) represents the probability that the component has failed under the fusion of the three Bayesian network information.

[0089] 2.4 Bayesian network node conditional probability assignment Due to the lack of data on DC-related prior probabilities and conditional probability tables, DC circuit breakers are relatively new in technology and have been in engineering applications for a short time, and there is no actual engineering data for reference. At the same time, when a DC distribution network fails, the emergency fault current limiting strategy limits the fault current, which will have a certain impact on the DC distribution network relay protection action, increasing the probability of protection and circuit breaker refusal and malfunction. AC relay protection has rich historical operation data and expert knowledge. The present invention believes that DC prior probability data can refer to the relevant AC prior probability data. Therefore, it is reasonable to assume that the probability that the DC distribution network protection and circuit breaker do not refuse to operate or malfunction is 90% of the probability that the protection and circuit breaker do not refuse to operate or malfunction in the AC distribution network. The set component node failure prior probability and the relay protection device and circuit breaker operation failure probability are shown in Table 2 and Table 3 respectively.

[0090] Table 2 Prior probability table of component node failure

[0091]

[0092] Table 3 Failure probability of relay protection device and circuit breaker operation

[0093]

[0094] 2.5 DC distribution network fault diagnosis

[0095] DC distribution network fault diagnosis based on Bayesian networks mainly consists of two parts: the first is to identify the faulty components in the DC distribution network; the second is to analyze the operating behavior of relay protection devices and circuit breakers. These two parts are introduced in turn below.

[0096] 2.5.1 Determining Faulty Components in DC Distribution Networks Since each component in the power system network can be modeled as its own Bayesian network, each component can only appear in its own Bayesian network. The formula for determining faulty components is shown in the formula below. Let the fault probability of each component after reverse reasoning be a n , assuming the line fault threshold is 0.8, the bus fault threshold is 0.6, that is, if a n ≥ the fault threshold of the corresponding component, the component is determined to be a faulty component.

[0097] 2.5.2 Analysis of relay protection device and circuit breaker operation behavior

[0098] The forward reasoning of the Bayesian algorithm is used in combination with the Bayesian prior probability assignment to obtain the posterior probability of each node of the faulty element in the DC distribution network under its corresponding protection information Bayesian network model and circuit breaker information Bayesian network model. That is, in the case of element failure, the expected action probability of each protection device node and circuit breaker node is obtained. The specific method is as follows: (1) Find the protection information Bayesian network model and circuit breaker information Bayesian network model corresponding to the faulty element, and set the parent node (element node) in these two Bayesian network models to state "1", that is, this element has failed;

[0099] (2) Perform forward inference of the Bayesian algorithm on the protection information Bayesian network model and the circuit breaker information Bayesian network model to calculate the expected action of the corresponding protection and circuit breaker in the event of a fault;

[0100] (3) The expected action probability of the protection device and the circuit breaker obtained by forward reasoning of the Bayesian algorithm is subtracted from the actual action situation E of the protection device and the circuit breaker. The actual situation is obtained from the SCADA system, that is, the information of the protection device and the circuit breaker is "1" if it is in action and "0" if it is not in action. This is used as the action judgment of the protection device and the circuit breaker.

[0101] That is, through the following formula:

[0102]

[0103] Where a∈(0,1), b∈(0,1) and a+b=1. Taking into account the calculation results of other relevant literature and this paper, a and b are both reasonably set to 0.5.

[0104] Based on the results of Bayesian forward reasoning, the action behaviors of far backup protection and near backup protection are different from the actual action behaviors. There are certain misjudgments in reflecting the logical sequence of main protection, near backup protection and far backup protection actions.

[0105] And when the following two situations occur, the results will be misjudged:

[0106] (1) The local backup protection and circuit breaker refuse to operate, and the remote backup protection and circuit breaker operate correctly, which is judged to be a false operation;

[0107] (2) The main protection operates correctly, and the near backup protection operates incorrectly, which is judged as correct operation. The near backup protection does not operate and is judged as refusal to operate. For the above two situations, after analysis, it is found that: for the first situation, when the main protection and the near backup corresponding circuit breaker fail to operate, since the Bayesian forward reasoning cannot reflect the logic of the protection action, the near backup protection will refuse to operate and the far backup protection will be judged as malfunction, resulting in misjudgment in the analysis of the far backup protection action behavior. For the second situation: when the main protection operates correctly, the main protection will operate correctly, and the near backup protection action behavior will be misjudged.

[0108] Therefore, after the preliminary judgment of the protective device and circuit breaker action behavior through Bayesian forward reasoning, a feedback process is required to determine whether the above two special situations occur. If the above two action situations occur, it is necessary to formulate expert system rules to correct the results. The expert system rules are as follows:

[0109] (1) For misjudgment of remote backup protection action behavior, by finding Bayesian forward reasoning, we can preliminarily obtain the refusal circuit breaker in the action behavior analysis of the protection action device, find the remote backup protection and circuit breaker associated with the refusal circuit breaker, and correct the expected probability of their nodes.

[0110] (2) When the near backup protection action is misjudged, the Bayesian network is used to find the main protection element on the same side of the component. Under the premise of the main protection action, the expected action probability of the near backup protection action on the same side is modified. The expected action probability of the above two cases is added by 1.

[0111] 3 Simulation

[0112] Based on MATLAB R2018a simulation platform, according to Figure 1 The protection configuration diagram of the DC distribution network shown in the figure compares the fault diagnosis between the traditional Bayesian network algorithm and the Bayesian network-based information fusion method proposed in the present invention.

[0113] 3.1 Line Fault

[0114] Regarding line faults in a DC distribution network, the present invention designs three fault scenarios, which are as follows: Scenario 1: Line L 6-7 Fault, converter emergency current limiting control starts, circuit breaker CB 6-7 and CB 7-6 Tripping; the specific relay protection and circuit breaker action information is shown in Table 4.

[0115] Table 4 Relay protection and circuit breaker action information for scenario 1

[0116]

[0117] Taking scenario 1 as an example, the method of the present invention is briefly described. Referring to the protection and circuit breaker action information combined with the DC distribution network topology, it can be seen that there is only one element in the fault area, namely line L 6-7 ; For line L 6-7 Establish three Bayesian network models such as Figure 7 — Figure 9 As shown in the figure, by combining the established Bayesian network model with the protection and circuit breaker action information, the fault probabilities obtained by the three Bayesian networks are fused through the DS evidence fusion method. The final fault probability is 99.4597%, while the fault probability obtained by the traditional method is 91.252%. From the failure probability results, it can be seen that the calculation results of the method of the present invention are better than those of the traditional method when there is no refusal or malfunction of the protection and circuit breaker.

[0118] Scenario 2: Line L 6-7 Fault, fault current limiter input, circuit breaker CB 6-7 L refuses to operate, the related backup protection operates and triggers the circuit breaker to trip; the specific protection and circuit breaker operation information is shown in Table 2. Combining the relay protection operation information with the actual DC distribution network topology, it can be seen that there are four elements in the fault area, among which the line element is L 5-6 , L 6-7 and L 6-25 , the busbar element is B6, and the corresponding Bayesian network model is established for these elements and the corresponding probability is calculated. The traditional method is compared with the method adopted by the present invention. Figure 10 shown.

[0119] Table 5 Relay protection and circuit breaker action information for scenario 2

[0120]

[0121] After determining that the faulty component is line L 6-7 After that, the Bayesian forward reasoning calculation is performed on the Bayesian network corresponding to the fault component. According to the relay protection and circuit breaker behavior judgment rules, the calculation results are as follows: Figure 11 shown.

[0122] From the calculation results of relay protection and circuit breaker behavior, it can be seen that CB 6-7 L is in the malfunction area, so the overall fault diagnosis result is line L 6-7 Fault, circuit breaker CB 6-7 L refuses to move, consistent with the preset scenario.

[0123] Table 6 Analysis of diagnostic results

[0124]

[0125] Scenario 3: Line L 6-7 Fault, the converter emergency current limiting control starts, the main protection L 6-7 Lm refuses to operate, the related backup protection operates and triggers the circuit breaker to trip; the analysis results of component failure and relay protection operation behavior are shown in Table 8. Figure 12 and Figure 13 shown.

[0126] Table 7 Scenario 3 protection and circuit breaker action information

[0127]

[0128] Table 8 Analysis of diagnostic results

[0129]

[0130] When the line main protection operates but the proximal circuit breaker fails to operate (Scenario 2), the proposed method improves the accuracy of the fault line by 0.5% compared to the traditional method. This is because the line main protection has a significant impact on fault diagnosis, that is, when the main protection operates, it can be determined that the line corresponding to the main protection has a fault. To avoid the influence of the main protection on the identification of the fault component, when the line main protection fails to operate (Scenario 3), the proposed method improves the accuracy of the fault line by 5.33% compared to the traditional method, demonstrating the advantages of the proposed algorithm. For the busbar without the fault, the comparison results show that the proposed method has a lower accuracy than the traditional method, which is more conducive to the identification of the fault component.

[0131] 3.2 Busbar Fault

[0132] Similarly, for busbar faults in DC distribution networks, the present invention designs three fault scenarios, which are as follows: Scenario 4: Busbar B6 fails, the converter emergency current limiting control is activated, and the circuit breaker CB 5-6 R, CB 6-7 L and CB 6-25 L trips;

[0133] Table 9 Relay protection and circuit breaker operation information for scenario 4

[0134]

[0135] Scenario 5: Busbar B6 fault, fault current limiter activated, circuit breaker CB 6-25 L refuses to move, protect L 6-7 Rs malfunctions, the related backup protection operates and triggers the circuit breaker to trip.

[0136] Table 10: Relay protection and circuit breaker operation information for scenario 5

[0137]

[0138] Scenario 6: Busbar B6 fails, the converter emergency current limiting control is activated, the busbar main protection B6m refuses to operate, the relevant backup protection operates and triggers the circuit breaker to trip.

[0139] Table 11 Relay protection and circuit breaker operation information for scenario 6

[0140]

[0141] The fault components are diagnosed and the relay protection behavior is analyzed for each of the three fault scenarios. The diagnostic results for scenario 5 are as follows: Figure 14 and Figure 15 As shown, the diagnosis results of scenario 6 are as follows Figure 16 and Figure 17 The overall diagnosis results in the three scenarios are shown in Table 12.

[0142] Table 12 Analysis of diagnostic results in three scenarios

[0143]

[0144] When a busbar fault occurs in a DC distribution network, the calculation results show that the probability of the busbar fault in the fault area calculated by the method of the present invention is greater than that of the traditional method; the probability of the busbar and line without fault in the fault area calculated by the method of the present invention is less than that of the traditional method, that is, the method of the present invention can increase the probability of the faulty busbar and reduce the probability of the non-faulty busbar and line, which is more conducive to fault judgment. Scenario 6 sets a serious fault situation in which the main busbar protection refuses to operate. Figure 16 The results show that the traditional method can no longer perform fault identification, but the method of the present invention can still accurately perform fault diagnosis.

[0145] This paper proposes a DC distribution network fault diagnosis method based on Bayesian network information fusion. This method enables diagnosis of components and relay protection behaviors after a DC distribution network fault occurs, improving the accuracy of fault diagnosis. Compared with existing fault diagnosis methods, the research method of this paper has the following characteristics:

[0146] (1) The traditional relay protection information is divided into two categories: protection action information and circuit breaker action information. Bayesian network models are established according to their respective action logics, which effectively avoids the problem that the circuit breaker information has little influence on the component fault judgment in the traditional Bayesian network model.

[0147] (2) The fault emergency current limiting control strategy of DC distribution network is considered, quantified into a Bayesian network model and introduced into the fault component diagnosis of DC distribution network. The relationship between the fault emergency current limiting control strategy and relay protection is analyzed, which shows that its control strategy does not conflict with traditional relay protection.

[0148] (3) The DS evidence fusion theory is used to integrate the fault emergency current limiting control strategy with the component fault diagnosis results obtained by traditional relay protection, thereby improving the accuracy of DC distribution network fault diagnosis results.

[0149] DC distribution networks contain numerous controlled power electronic devices. When a fault occurs, fault diagnosis requires a comprehensive integration of control methods, in addition to relay protection. In addition to emergency current limiting, the characteristics of the DC distribution network structure and other control systems should be fully considered. Exploring new DC distribution network fault diagnosis methods tailored to these characteristics is a topic for further research.

Claims

1. A DC distribution network fault diagnosis method based on Bayesian network information fusion, characterized by: The steps are: S1. Information fusion of Bayesian networks Assume that there are two pieces of evidence E1 and E2 under the identification framework Θ, the basic probability assignment functions are M1 and M2, and the states corresponding to the evidence are represented by A i and B j To express; Dempster's synthesis rule is defined as: is called the regularization parameter; if m is used to represent the probability assignment function, then the orthogonal sum of the two assignment functions m1 and m2 is recorded as When k = 1, m1 and m2 are considered contradictory, and the combination rule cannot be used to synthesize the basic probability assignment; when k ≠ 1, the synthesis rule formula is valid, and m determines a basic probability assignment. When there are multiple pieces of evidence to be fused, it is obtained by expanding the above formula (2); The three component failure probability information is combined with the DS evidence theory synthesis rule, and the fusion formula is as follows: Where P P 、P CB and P L represents the probability values ​​obtained under the Bayesian network model for protection information, circuit breaker information, and emergency fault current limiting strategy information; X represents the parent node corresponding to each Bayesian network, that is, the component; the numbers "1" and "0" represent the component status, that is, the number "1" indicates that the component has failed, and the number "0" indicates that the component has not failed; S2. Conditional probability assignment of Bayesian network nodes Assume that the probability of the protection and circuit breaker in the DC distribution network not to refuse to operate or to malfunction is 90% of that in the AC distribution network; S3. DC distribution network fault diagnosis a. Determine the faulty components in the DC distribution network Let the failure probability of each component after reverse reasoning be a n , assuming the line fault threshold is 0.8, the bus fault threshold is 0.6, that is, if a n ≥ the fault threshold of the corresponding component, the component is determined to be a faulty component; b. Relay protection device and circuit breaker operation Calculate the expected operation probability of each protection device node and circuit breaker node; the specific method is as follows: (1) Find the protection information Bayesian network model and the circuit breaker information Bayesian network model corresponding to the faulty component, and set the parent node in these two Bayesian network models to the state "1", that is, the component has failed; (2) Perform forward inference of the Bayesian algorithm on the protection information Bayesian network model and the circuit breaker information Bayesian network model to calculate the expected action of the corresponding protection and circuit breaker in the event of a fault; (3) Subtract the expected action probability of the protection device and circuit breaker obtained by forward reasoning of the Bayesian algorithm from the actual action situation E of the protection device and circuit breaker. The actual situation is obtained from the SCADA system, that is, the information of the protection device and circuit breaker is "1" if it is in action and "0" if it is not in action. This is used as the action judgment of the protection device and circuit breaker; that is, the following formula is used: Among them, a∈(0,1), b∈(0,1) and a+b=1, and the reasonable values ​​of a and b are both 0.

5.

2. The DC distribution network fault diagnosis method based on Bayesian network information fusion according to claim 1 is characterized by: Based on the Bayesian forward reasoning results, the differences between the far backup protection and near backup protection action behaviors and the actual action behaviors are obtained: (1) For the misjudgment of remote backup protection action behavior, by finding Bayesian forward reasoning, we can preliminarily obtain the refusal circuit breaker in the action behavior analysis of the protection action device, find the remote backup protection and circuit breaker associated with the refusal circuit breaker, and correct the expected action probability; (2) When the near backup protection action is misjudged, the main protection element on the same side of the component is found in the Bayesian network. Under the premise of the main protection action, the expected action probability of the near backup protection action on the same side is modified and the above expected action probability is added by 1.

Images