Secondary system fault location method based on network flow algorithm and deep neural network

Abstract

The present invention provides a secondary system fault location method based on a network flow algorithm and a deep neural network, proposes a new fault classification method, and performs binary coding on common faults; after constructing a deep neural network, it is trained, and when a fault occurs in the secondary system, fault feature information is obtained, and a fault feature set is constructed based on the fault feature information; the fault feature set is input into the trained deep neural network, and the deep neural network is used to process the fault to obtain a fault binary code; the fault position is preliminarily located through the fault binary code, and an improved network flow algorithm is introduced to solve the problem of ambiguity in the location of link faults and node faults in complex fault location, and node link information at the time of the fault is collected and compared with the normal operating conditions for further fault location. By deeply integrating the network flow algorithm with the deep neural network model, accurate positioning of the secondary system fault of the smart substation is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a secondary system fault location method based on a network flow algorithm and a deep neural network. Background Art

[0002] In recent years, with the promulgation of the IEC61850 communication protocol and the development of intelligent electronic devices and Ethernet technology, smart substation technology has gradually matured. Although smart substations provide great convenience for digital signal acquisition, the simultaneous influx of large amounts of data has caused the information to lose selectivity, which in turn increases the difficulty for operation and maintenance personnel to quickly locate faults after relay protection equipment failures. Therefore, research on fault location methods for the secondary system of smart substations is particularly important.

[0003] In recent years, experts and scholars have proposed some methods for the problem of fault location in smart substation protection systems. For example, some methods use the smart substation configuration file to extract the communication status of the virtual circuit as a criterion to determine the scope of the secondary circuit fault. However, due to the different SCD file specifications of different substations, this method lacks versatility. Some methods locate faults based on abnormal information in the message information flow and provide detailed criteria. However, this method can only determine device failures and it is difficult to locate communication link failures. In addition, when the power network changes dynamically, it is impossible to quickly correct the solution. In addition, some methods perform status assessment by analyzing the real-time information of the secondary equipment during operation. However, this method only uses the operating information of a single device and fails to integrate the operating information of multiple devices, making it difficult to accurately locate the secondary circuit of the smart substation. In summary, the current fault location methods of smart substation protection systems mainly have the following problems:

[0004] 1) Existing methods tend to process specific fault characteristics and lack the ability to comprehensively handle multiple fault types. However, the secondary system of a smart substation contains multiple secondary devices, and it is necessary to comprehensively locate the faults of multiple devices.

[0005] 2) Existing methods cannot quickly revise solutions when faced with dynamic changes in the power network. Summary of the Invention

[0006] In view of this, the present invention proposes a secondary system fault location method based on network flow algorithm and deep neural network. Combining network flow algorithm and deep neural network can achieve accurate location of complex faults and improve the accuracy of fault location.

[0007] The technical solution of the present invention is achieved as follows:

[0008] The secondary system fault location method based on network flow algorithm and deep neural network includes the following steps:

[0009] Step S1: Acquire common faults of the substation secondary system, classify and integrate the common faults, extract fault feature information, and perform binary coding based on the fault feature information of the common faults;

[0010] Step S2: Divide the fault feature information and binary codes of common faults into a training set and a test set, construct a deep neural network, train the deep neural network with the training set, and test it with the test set;

[0011] Step S3: Acquire fault feature information when a fault occurs in the secondary system, and construct a fault feature set based on the fault feature information;

[0012] Step S4: input the fault feature set into the trained deep neural network, and obtain the fault binary code through deep neural network processing;

[0013] Step S5: Preliminarily locate the fault location through fault binary coding, introduce an improved network flow algorithm, collect node link information at the time of the fault, and compare it with the normal operation to further locate the fault.

[0014] Preferably, the specific steps of step S1 are:

[0015] Step S11: obtaining a three-layer two-network structure of the smart substation, and extracting common devices from the three-layer two-network structure;

[0016] Step S12: Use a directed graph to perform simplified topology analysis on the three-layer two-network structure to obtain the connection status of each node in the smart substation;

[0017] Step S13: classify the fault type and fault complexity based on the connection status of each node in the smart substation and common equipment, and perform binary coding according to the fault characteristic information and fault complexity of the fault type.

[0018] Preferably, the commonly used equipment includes protection devices, measurement and control devices, merging units and intelligent terminals, and the fault types include merging unit faults, intelligent terminal faults, protection device faults and link faults, and each of the fault types can be subdivided into several sub-fault types.

[0019] Preferably, the fault complexity classification in step S13 is performed based on the fault characteristic information of each fault, and the specific rules are:

[0020] Faults with only one device fault and that can be solved directly through fault characteristic information are classified as simple faults;

[0021] Faults in which multiple devices fail simultaneously and can be resolved using fault signature information are classified as pseudo-complex faults;

[0022] Faults that cannot be directly resolved through fault feature information are called complex faults.

[0023] Preferably, the binary code comprises 17 bits, wherein:

[0024] D16~17 represent the complexity of the fault, 00 represents no fault, 01 represents a simple fault, 10 represents a pseudo-complex fault, and 11 represents a complex fault;

[0025] D12 to D15 indicate the fault type, which corresponds to the link fault status, protection device fault status, intelligent terminal fault status, and merging unit fault status in order. 1 indicates a fault and 0 indicates no fault.

[0026] D5 to D11 indicate the fault location, where D10 and D11 indicate whether the fault occurs in the device or link. 00 indicates no fault, 01 indicates a device fault, and 10 indicates a link fault. D5 to D9 indicate the order of the devices or links.

[0027] D1~D4 represent the number of each fault.

[0028] Preferably, the deep neural network in step S2 includes two hidden layers, and the hidden layers use ReLU activation function, wherein the number of neurons in the first hidden layer is 64, and the number of neurons in the second hidden layer is 32.

[0029] Preferably, the expression of the fault feature set in step S3 is:

[0030] ;

[0031] Where X is the fault feature set, is the fault feature vector set of the nth fault event, The specific matrix behavior is:

[0032] ;

[0033] in They are matrices composed of merging unit, intelligent terminal, protection device and link fault feature information.

[0034] Preferably, the specific steps in step S5 are:

[0035] Step S51: Set the maximum flow capacity of all edges of each node of the smart substation to 1;

[0036] Step S52: Preliminarily analyze the fault signal through a deep neural network, locate the approximate fault area, and select the source node in the approximate fault area;

[0037] Step S53: Further fault location is performed by monitoring and comparing the changes in the total output flow of the source node before and after the fault.

[0038] Preferably, the method for selecting the source node in step S52 is: selecting the lowest layer device in the approximate fault area as the source node.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention discloses a secondary system fault location method based on a network flow algorithm and a deep neural network. Common faults in a substation secondary system are classified and integrated, and then binary-coded to obtain a unique binary code corresponding to each common fault. The common faults and binary codes are then used as a training set and a test set. The training set is used to train the deep neural network, and the accuracy is tested by the test set after training to a certain stage. When the test accuracy reaches a preset threshold, the training can be stopped. When a fault occurs in the secondary system, the fault feature information at the time of the fault can be transmitted to the deep neural network, and the deep neural network is used to process the fault to obtain a specified fault binary code. The fault binary code can be used to preliminarily locate the fault location. Then, a network flow algorithm is introduced to compare the flow information of the source node before and after the fault, so that further fault location can be performed, thereby accurately obtaining the fault location and improving the accuracy of fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 A fault location flow chart of the secondary system fault location method based on network flow algorithm and deep neural network of the present invention;

[0043] Figure 2 A three-layer, two-network structure diagram of a smart substation using a secondary system fault location method based on a network flow algorithm and a deep neural network according to the present invention;

[0044] Figure 3 A diagram showing the connection status of each node in the smart substation according to the secondary system fault location method based on the network flow algorithm and deep neural network of the present invention;

[0045] Figure 4 A structural diagram of a deep neural network of a secondary system fault location method based on a network flow algorithm and a deep neural network according to the present invention;

[0046] Figure 5 The smart substation bay diagram of embodiment 1 of the secondary system fault location method based on network flow algorithm and deep neural network of the present invention;

[0047] Figure 6 A deep neural network multiple optimization parameter diagram of embodiment 1 of the secondary system fault location method based on a network flow algorithm and a deep neural network of the present invention; DETAILED DESCRIPTION

[0048] In order to better understand the technical content of the present invention, a specific embodiment is provided below, and the present invention is further described in conjunction with the accompanying drawings.

[0049] See also Figure 1 The secondary system fault location method based on network flow algorithm and deep neural network provided by the present invention includes the following steps:

[0050] Step S1: Acquire common faults of the substation secondary system, classify and integrate the common faults, extract fault feature information, and perform binary coding based on the fault feature information of the common faults;

[0051] Step S2: Divide the fault feature information and binary codes of common faults into a training set and a test set, construct a deep neural network, train the deep neural network with the training set, and test it with the test set;

[0052] Step S3: Acquire fault feature information when a fault occurs in the secondary system, and construct a fault feature set based on the fault feature information;

[0053] Step S4: input the fault feature set into the trained deep neural network, and obtain the fault binary code through deep neural network processing;

[0054] Step S5: Preliminarily locate the fault location through fault binary coding, introduce an improved network flow algorithm, collect node link information at the time of the fault, and compare it with the normal operation to further locate the fault.

[0055] The secondary system fault location method based on network flow algorithm and deep neural network of the present invention classifies and integrates common faults of substation secondary system, including classification of type and complexity. Common faults include multiple types, which can be divided into multiple complexities based on whether the fault type can be quickly resolved. Different fault types and complexities can form different combinations, and each combination is binary coded. The identity identifiers of common faults are obtained, and then the fault feature information and binary codes of common faults are used as a set of data. All the data are divided into a training set and a test set, where the training set is used to train the deep neural network. After the training is completed, the accuracy can be tested through the test set. When the accuracy reaches the preset threshold, the training can be stopped. At this time, the deep neural network can be put into use. When a fault occurs in the secondary system of the substation, the fault feature information at the time of the fault can be collected, and the fault feature information can be constructed into a fault feature set. The fault feature set is used as the input of the deep neural network. After the deep neural network obtains the fault feature set, it can automatically process it to obtain the corresponding fault binary code. The fault location can be preliminarily located through the fault binary code. At this time, the fault location needs to be further determined. Therefore, the present invention introduces an improved network flow algorithm. By collecting the node link information in the preliminarily located fault location and comparing it with the normal operating conditions, the fault can be further located. By adopting the network flow algorithm to fuse the deep neural network, complex faults can be accurately located and the fault location accuracy can be improved.

[0056] Preferably, the specific steps of step S1 are:

[0057] Step S11: obtaining a three-layer two-network structure of the smart substation, and extracting common devices from the three-layer two-network structure, where the common devices include protection devices, measurement and control devices, merging units, and smart terminals;

[0058] Step S12: Use a directed graph to perform simplified topology analysis on the three-layer two-network structure to obtain the connection status of each node in the smart substation;

[0059] Step S13: Based on the connection status of each node in the smart substation and common equipment, the fault type and fault complexity are divided. Binary coding is performed according to the fault characteristic information and fault complexity of the fault type. The fault types include merging unit fault, smart terminal fault, protection device fault and link fault. Each fault type can be subdivided into several sub-fault types. The fault complexity is divided according to the fault characteristic information of each fault. The specific rules are as follows:

[0060] Faults with only one device fault and that can be solved directly through fault characteristic information are classified as simple faults;

[0061] Faults in which multiple devices fail simultaneously and can be resolved using fault signature information are classified as pseudo-complex faults;

[0062] Faults that cannot be directly resolved through fault feature information are called complex faults.

[0063] Common equipment in smart substations include switches, protection devices, measurement and control devices, merging units, and smart terminals. They can be divided into line interval and busbar detection according to their functions, such as Figure 2 As shown in the figure, different process layer devices and networks can be analyzed using the conventional "multi-node side" method. The research objects of this invention include merging units, intelligent terminals, protection devices and link faults. When the secondary system is operating normally, it is necessary to focus on the on-off of the link and the normal operation of the device. Therefore, a directed graph can be used to analyze the Figure 2 At the same time, since the fault of the communication link of the smart substation can directly transmit the fault information to the monitoring system, it can locate the fault more accurately and quickly than the link of the physical power grid, especially the type of fault. Therefore, the present invention uses the connection relationship between the communication link and the smart terminal, and draws on the positioning logic of the physical power grid to locate the communication link using the device-link-device method, so that the positioning accuracy of the communication link is higher and the positioning accuracy is improved. Figure 2 A detailed analysis of the line bays and bus bays in the corresponding devices and links is conducted, and the node connections are as follows: Figure 3 As shown, the circle represents the node, the number in the circle represents the node number, the curve represents the link, and the number in brackets in the curve represents the link number.

[0064] In the fault location of the secondary system of smart substation, the classification and refined processing of fault information are important ways to improve the accuracy of fault location. Figure 3 In the fault classification model shown, the present invention divides secondary system faults into four types: merging unit faults, intelligent terminal faults, protection device faults, and link faults. Merging unit faults are further subdivided into faults MU1 and MU2, intelligent terminal faults into faults IT1 and IT2, protection device faults into faults R1 and R2, and link faults into 17 line faults, GOOSE network faults, and SV network faults. This detailed classification allows for high accuracy in handling single faults, and the operation is relatively simple and effective.

[0065] However, this classification approach has limitations, particularly in diagnosing link faults. Specifically, when a link fault occurs between merging unit MU1 and protection device R1, relying solely on the fault information provided by the alarm message often fails to accurately determine the specific location of the fault. This can lead to the link fault being misidentified as a merging unit fault or a protection device fault.

[0066] Based on the above analysis, it is necessary not only to distinguish between the four types of device failures, but also to distinguish between simple failures and complex failures. The existing classification standard is based on the number of equipment failure types. The specific classification standards are as follows:

[0067] 1) When all devices are in good condition, there is no fault.

[0068] 2) When only a single device fails, it is a simple failure.

[0069] 3) When multiple devices fail at the same time, it is a complex failure.

[0070] This method can accurately locate simple faults, but complex faults require locating based on on-site conditions and experience. This classification standard is applicable to most problems, but for a few faults, better solutions may be available. For example, the network storm on the GOOSE_A network of the merging unit. Based on the above classification, this fault is both a merging unit failure and a link failure, making it a complex fault. However, this fault has a specific fault signature that can be used to directly locate the fault and diagnose its type.

[0071] Therefore, it is necessary to re-establish rules to better locate such faults. The specific rules are as follows:

[0072] 1) A simple fault is one that has only one device fault and can be solved directly using fault characteristic information.

[0073] 2) Faults in which multiple devices fail simultaneously and can be resolved through specific fault characteristic information are called pseudo-complex faults.

[0074] 3) Faults that cannot be directly resolved through fault feature information are called complex faults.

[0075] Although this method is slightly cumbersome, the accuracy of judgment will be significantly improved.

[0076] Preferably, the binary code comprises 17 bits, wherein:

[0077] D16~17 represent the complexity of the fault, 00 represents no fault, 01 represents a simple fault, 10 represents a pseudo-complex fault, and 11 represents a complex fault;

[0078] D12 to D15 indicate the fault type, which corresponds to the link fault status, protection device fault status, intelligent terminal fault status, and merging unit fault status in order. 1 indicates a fault and 0 indicates no fault.

[0079] D5-D11 indicate the fault location. D10 and D11 indicate whether the fault occurs in a device or link. 00 indicates no fault, 01 indicates a device fault, and 10 indicates a link fault. D5-D9 indicate the order of the devices or links. When D10-11 is 00, the last five digits can only represent 00000, indicating no fault. When D10-11 is 01, the last five digits represent a device fault. In this case, 00001 indicates a fault in device 1. When D10-11 is 10, the last five digits represent a link fault. In this case, 00001 indicates a fault in link 1. Because there are 12 devices and 20 links in total, and the maximum number is between 16 and 32, a 5-digit binary representation is used.

[0080] D1-D4 represent the fault number of each type. Each number represents a fault. Because the number of fault types ranges from 8 to 16, a 4-bit binary code is required to store the data. If the fault code has more than 16 bits, 5 or more bits can be used to represent it.

[0081] Based on the above complexity classification standard, the present invention classifies and integrates multiple fault types and performs binary encoding as the output of the deep neural network. The specific fault types and fault feature information are shown in Table 1:

[0082] Table 1 Typical fault types and their characteristic information

[0083]

[0084] Since the coding involves the location of the fault, the binary coding needs to be written for specific devices and links. The fault complexity and coding are shown in Table 2:

[0085] Table 2 Fault complexity and its coding

[0086]

[0087] Preferably, the deep neural network in step S2 includes two hidden layers, and the hidden layers use ReLU activation function, wherein the number of neurons in the first hidden layer is 64, and the number of neurons in the second hidden layer is 32. The specific structure is as follows: Figure 4 shown.

[0088] Preferably, the expression of the fault feature set in step S3 is:

[0089] ;

[0090] Where X is the fault feature set, is the fault feature vector set of the nth fault event, The specific matrix behavior is:

[0091] ;

[0092] in They are matrices composed of merging unit, intelligent terminal, protection device and link fault feature information.

[0093] Smart substation faults generate fault signature information and report it to the monitoring system, making it easier for staff to identify the fault type and location. In order to organize these fault signature information more systematically, the present invention designs a matrix structure to form a fault signature set.

[0094] Specifically, the compilation of the fault feature set follows the above four fault types, with the aim of more efficiently identifying and diagnosing various types of faults and making it easier for relevant staff to query and handle them in actual work.

[0095] In a smart substation, all fault alarm signals are represented by switching quantities. Therefore, when the system detects a fault alarm signal, the corresponding matrix element value is 1; if the fault alarm signal is not detected, the matrix element value is 0.

[0096] After screening and data analysis, 19 types of fault signature information were selected that effectively reflect faults. These include two types of merging unit fault signature information, two types of intelligent terminal fault signature information, one type of protection device fault signature information, and 14 types of link fault information. If the fault signature information is used as matrix column elements, the fault signature set matrix should have 19 column vectors, with a total of 19 columns. This number of columns should remain constant when the fault signature information is fixed. This method allows the constructed fault signature set to clearly reflect the characteristics of each fault type and provide strong support for subsequent fault diagnosis and resolution.

[0097] Preferably, the specific steps in step S5 are:

[0098] Step S51: Set the maximum flow capacity of all edges of each node of the smart substation to 1;

[0099] Step S52: Preliminarily analyze the fault signal through a deep neural network, locate the approximate fault area, and select the lowest-level device in the approximate fault area as the source node;

[0100] Step S53: Further fault location is performed by monitoring and comparing the changes in the total output flow of the source node before and after the fault.

[0101] For simple and pseudo-complex faults, deep neural networks can directly determine the fault type and location based on specific fault information. However, when faced with complex faults, particularly those involving merging unit I / O plug-in failures and link failures, deep neural networks can only determine the most likely fault point based on a large set of fault features. This approach has low accuracy and cannot meet fault location requirements unless human intervention is performed.

[0102] Network flow algorithms can efficiently process multi-source heterogeneous data in complex network topologies, possess global optimization capabilities, and accurately locate link and node faults, resolving the ambiguity inherent in traditional methods. Their dynamic adaptability supports real-time updates to traffic distribution, adapting to changes in power system topology, while maintaining low computational complexity, meeting the real-time requirements of fault location. Therefore, the present invention considers integrating network flow algorithms to improve the accuracy of neural networks in the face of such faults.

[0103] The basic flow of network flow is as follows:

[0104] 1) Construct a directed graph G = (V, E), where V is the node set and E is the edge set. Figure 2 Consider information transmission from bottom to top as a directed graph.

[0105] 2) Determine the source point, starting from a specific node, and all flows flow into this node.

[0106] 3) Determine the sink point where all flows terminate.

[0107] 4) Determine the capacity function c to give the maximum capacity of each edge.

[0108] 5) Determine the flow. This involves satisfying the constraint function f. This generally involves capacity constraints and flow conservation. The capacity constraint states that for an edge flow e∈, the corresponding edge can accommodate the flow E, 0 ≤ f(e) ≤ c(e). Flow conservation states that for all non-source and non-sink nodes v∈V, the total flow into node v equals the total flow out of node v.

[0109] Based on the above basic theory, Figure 3 For the fault location scenario shown, the following optimization strategies are proposed:

[0110] 1) The maximum traffic capacity of all edges is uniformly set to a unit value of 1, ensuring that the traffic carried by each communication link is constant at 1. This measure ensures that the amount of traffic reduction caused by any single link failure remains consistent, thereby simplifying the fault detection process and improving the efficiency and accuracy of fault location.

[0111] 2) Generalize the concepts of source and target nodes. When a network encounters an anomaly, deep neural networks are used to initially analyze the fault signal and pinpoint the approximate fault area. The source and target nodes are then selected within this area, with the lowest-level device in the fault area designated as the source node and the highest-level device as the sink. By comparing the status of the source and target nodes, as well as the intermediate transmission paths, more accurate fault location is achieved.

[0112] 3) Use the total outbound traffic of a source node as a key evaluation metric. Given that the traffic limit for each link is set to 1, the total outbound traffic of a source node is effectively equivalent to the number of its external links. The number of external links can be easily determined using the node association matrix. This parameter is crucial for fault diagnosis.

[0113] By implementing the three improvement measures mentioned above, we can effectively address the problem of link failures and input / output port failures being difficult to accurately locate. Specifically, by monitoring and comparing the changes in the total output traffic of the source node before and after the failure, we can significantly enhance the effectiveness of the neural network in fault location.

[0114] Example 1

[0115] The following takes a 220kV smart substation bay as an example to verify the effectiveness of the method proposed in the present invention. Figure 5 As shown in the figure, the transformer bay structure consists of line bays, bus bays and transformer bays. The overall architecture includes 20 nodes and 31 connection links, and is equipped with 3 merging units, 3 intelligent terminals and 3 sets of protection devices. By setting a rich set of fault features obtained from a variety of fault scenarios, the fault feature information matrix is ​​a matrix of 51 rows and 19 columns. By collecting actual smart substation fault sample data and data obtained from specific experiments, they are classified into simple fault samples, pseudo-complex fault samples and complex fault samples after screening. Samples of various fault types are used to construct training sets and test sets at a ratio of 70% and 30% to evaluate the generalization ability of the model. After multiple optimization of parameters, the results are as follows: Figure 6 shown.

[0116] The accuracy of the deep neural network model of the mixed sample composed of 3 types of samples under different iteration times and different initial learning rates is as follows: Figure 6 As shown, the X coordinate represents the initial learning rate, the Y coordinate represents the number of iterations, the Z coordinate represents the accuracy, and the color bar represents the accuracy value.

[0117] from Figure 6As can be seen, when the initial learning rate is fixed, the model's accuracy peaks after approximately 2500 iterations and stabilizes, with only a slight increase. When the number of iterations is less than 2500, the model's accuracy continues to rise slightly as the number of iterations increases. However, when the number of iterations exceeds 2500, the accuracy does not increase significantly, while the training time increases significantly. Taking both training time and accuracy into consideration, 2500 iterations was ultimately selected as the optimal number of iterations for model training.

[0118] from Figure 6 It is not difficult to see that the model performs best when the learning rate is set between 0.1 and 0.15. When the learning rate is lower than 0.1, the training process converges slowly, making it difficult for the model to achieve high performance within a limited number of iterations. When the learning rate is higher than 0.15, although the initial convergence speed is faster, the accuracy does not improve significantly and the growth is relatively slow. This embodiment ultimately selects an initial learning rate of 0.1 to ensure the efficiency and stability of the model during training, thereby achieving optimal fault localization performance.

[0119] Example verification:

[0120] 1) Simple fault type (merging unit module power failure)

[0121] For simple faults such as merging unit module failure, simulate Figure 5 The middle unit MU1 (node ​​15) is faulty.

[0122] When a fault occurs, a specific fault feature information, merging unit module power failure alarm, is generated. Converting this information into a fault feature vector set yields the matrix element 1 at row 46 and column 1, X46 1. The element 1 at X46 1 indicates the merging unit module power failure.

[0123] At the same time, according to the network flow algorithm, the total outgoing network flow under normal operation of node 15 is 3. The monitoring system uses monitoring information to perform network flow verification on all branches of node 15. The monitoring system cannot obtain information about the branches connected to node 15 and determines that the total outgoing network flow of node 15 is 0. By comparing, we get formula (1), which further verifies that the merging unit is faulty.

[0124] (1)

[0125] Where: is the total outgoing traffic of node 15 during normal operation; is the total outgoing traffic of node 15 when the failure occurs.

[0126] Combining the above judgment rules, the fault binary code is output as formula (2). 01 1000 0101111 0001(2)

[0128] The fault location is 0101111, which is the merging unit module and is merging unit module 15. The fault number is 0001, which means the merging unit has lost power. This matches the simulated fault described above and the judgment is correct.

[0129] 2) Complex fault type (merging unit I\O plug-in fault - merging unit to protection device)

[0130] For complex faults such as merging unit I\O plug-in failure - merging unit to protection device, simulation Figure 5 The middle unit MU1 (node ​​19) fails.

[0131] When a fault occurs, multiple fault feature information will appear, including merging unit GOOSE\SV interruption\alarm and protection device GOOSE\SV interruption\alarm. When converted into the fault feature information vector set, it can be obtained that the 26th, 27th, 28th, 38th and 50th rows, that is, X26j, X27j, X28j, X38j and X50j, will have element 1. The matrix can only locate the merging unit MU3 and protection device R3 and the link 27 between them, and no further positioning can be done.

[0132] According to the network flow algorithm, the total network flow sent by node 19 under normal operation is 3. The monitoring system uses monitoring information to perform network flow verification on the branches of node 19. It is found that the monitoring system cannot collect information on the links connected to node 19, and the total network flow sent by node 19 is determined to be 0. By comparing Equation (3), it can be determined that the fault occurred in the merging unit.

[0133] (3)

[0134] Where: is the total outgoing traffic of node 19 during normal operation; is the total outgoing traffic of node 19 when the failure occurs.

[0135] Combining the above judgment rules, the fault binary code is output as formula (4). 11 1011 0110011 0011(4)

[0137] Therefore, the fault location is 0110011, which is the merging unit module and merging unit module No. 19. The fault number is 0011, which means the merging unit I / O plug-in is faulty (merging unit to protection device). This is consistent with the simulated fault and the judgment is correct.

[0138] A large number of different types of fault examples are simulated for verification, and the discrimination results of some examples are listed in Table 3. It can be seen that the method proposed in the present invention can make correct discriminations.

[0139] Table 3. Some simulated fault identification results

[0140]

[0141] Method comparison:

[0142] The four types of samples mentioned in this invention were used for deep neural network training: samples containing only simple faults T1, samples containing only pseudo-complex faults T2, fault samples containing only complex faults T3, and mixed samples of the three types T4. The effects of the network flow algorithm and the particle swarm optimization algorithm proposed in this invention were compared under the above four sample types. The results are listed in Table 4:

[0143] Table 4 Fault location accuracy and time of different models

[0144]

[0145] As can be seen from Table 4, the network flow algorithm has a higher accuracy rate than the particle swarm optimization algorithm when processing fault samples. At the same time, compared with the particle swarm optimization algorithm, the network flow algorithm takes less time to locate faults and can obtain results more quickly.

[0146] This example addresses the challenges of locating faults in the secondary system of smart substations, which face difficulties in integrating multiple fault types and adjusting models to dynamic changes in the power system. A new fault classification method is proposed. A fault feature set is constructed using information on multiple fault types and their characteristics. A model integrating a network flow algorithm and a deep neural network is constructed. This method is compared with a particle swarm optimization algorithm, and a numerical example verifies the effectiveness of the proposed method. The conclusions are as follows:

[0147] 1) The advantage of this method is that it integrates information on various fault types from multiple devices and can quickly adjust to dynamic changes in the power system.

[0148] 2) This method can effectively solve the problem of locating complex faults, especially merging unit I / O plug-in faults and link faults, and improve the accuracy of fault location.

[0149] 3) Compared with the particle swarm optimization algorithm, this method has higher accuracy and shorter time for fault location.

[0150] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A secondary system fault location method based on network flow algorithm and deep neural network, characterized by: The following steps are involved: Step S1: Acquire common faults of the substation secondary system, classify and integrate the common faults, extract fault feature information, and perform binary coding based on the fault feature information of the common faults; Step S2: Divide the fault feature information and binary codes of common faults into a training set and a test set, construct a deep neural network, train the deep neural network with the training set, and test it with the test set; Step S3: Acquire fault feature information when a fault occurs in the secondary system, and construct a fault feature set based on the fault feature information; Step S4: input the fault feature set into the trained deep neural network, and obtain the fault binary code through deep neural network processing; Step S5: Preliminarily locate the fault location through fault binary coding, introduce an improved network flow algorithm, collect node link information at the time of the fault, and compare it with the normal operation to further locate the fault; The specific steps of step S1 are: Step S11: obtaining a three-layer two-network structure of the smart substation, and extracting common devices from the three-layer two-network structure; Step S12: Use a directed graph to perform simplified topology analysis on the three-layer two-network structure to obtain the connection status of each node in the smart substation; Step S13: classify the fault type and fault complexity based on the connection status of each node in the smart substation and common equipment, and perform binary coding according to the fault characteristic information and fault complexity of the fault type; The specific steps in step S5 are: Step S51: Set the maximum flow capacity of all edges of each node of the smart substation to 1; Step S52: Preliminarily analyze the fault signal through a deep neural network, locate the approximate fault area, and select the source node in the approximate fault area; Step S53: Further fault location is performed by monitoring and comparing the changes in the total output flow of the source node before and after the fault.

2. The secondary system fault location method based on network flow algorithm and deep neural network according to claim 1 is characterized in that: The commonly used equipment includes protection devices, measurement and control devices, merging units and intelligent terminals. The fault types include merging unit faults, intelligent terminal faults, protection device faults and link faults. Each of the fault types can be subdivided into several sub-fault types.

3. The secondary system fault location method based on network flow algorithm and deep neural network according to claim 1 is characterized in that: In step S13, the fault complexity is divided according to the fault characteristic information of each fault, and the specific rules are as follows: Faults with only one device fault and that can be solved directly through fault characteristic information are classified as simple faults; Faults in which multiple devices fail simultaneously and can be resolved using fault signature information are classified as pseudo-complex faults; Faults that cannot be directly resolved through fault feature information are called complex faults.

4. The secondary system fault location method based on network flow algorithm and deep neural network according to claim 1 is characterized in that: The binary code comprises 17 bits, where: D16~17 represent the complexity of the fault, 00 represents no fault, 01 represents a simple fault, 10 represents a pseudo-complex fault, and 11 represents a complex fault; D12 to D15 indicate the fault type, which corresponds to the link fault status, protection device fault status, intelligent terminal fault status, and merging unit fault status in order. 1 indicates a fault and 0 indicates no fault. D5 to D11 indicate the fault location, where D10 and D11 indicate whether the fault occurs in the device or link. 00 indicates no fault, 01 indicates a device fault, and 10 indicates a link fault. D5 to D9 indicate the order of the devices or links. D1~D4 represent the number of each fault.

5. The secondary system fault location method based on network flow algorithm and deep neural network according to claim 1 is characterized in that: The deep neural network in step S2 includes two hidden layers, and the hidden layers use the ReLU activation function, wherein the number of neurons in the first hidden layer is 64, and the number of neurons in the second hidden layer is 32.

6. The secondary system fault location method based on network flow algorithm and deep neural network according to claim 1 is characterized in that: The expression of the fault feature set in step S3 is: ； Where X is the fault feature set, is the fault feature vector set of the nth fault event, The specific matrix behavior is: ； in They are matrices composed of merging unit, intelligent terminal, protection device and link fault feature information.

7. The secondary system fault location method based on network flow algorithm and deep neural network according to claim 1 is characterized in that: The method for selecting the source node in step S52 is: selecting the lowest layer device in the approximate fault area as the source node.

Images