Transform and particle swarm hybrid-based power distribution network fault section positioning detection method and related device

Abstract

The invention discloses a power distribution network fault section positioning detection method based on mixture of Transform and a particle swarm and a related device. The method comprises the following steps: constructing and initializing an upstream section matrix, a downstream section matrix, an upstream power supply matrix and a downstream power supply matrix at the current moment; reading operation parameters of the regional power distribution network and configuration data of the measurement equipment; carrying out Transform model training by using the operation parameters and the configuration data of the measurement equipment; current fault information is extracted from the feeder terminal unit, and a Transform model is called to complement missing data; and constructing a target function based on the constructed matrix and data obtained after completion of the missing data, and solving the target function to locate the position of the fault section. According to the method, the intelligent algorithm is introduced on the basis of original FTU configuration of the power distribution network, missing data is supplemented, the particle swarm optimization is used for solving, and the calculation precision and robustness of power distribution network fault positioning are improved.

Description

Technical Field

[0001] The present invention relates to the field of power system monitoring and data analysis, and in particular to a distribution network fault section location detection method and related devices based on a hybrid of Transformer and particle swarm. Background Art

[0002] Distribution network state fault detection and island state detection are key technologies for accurate perception of the power system state. They provide basic data for advanced applications such as relay protection systems, fault location, island detection, voltage control, and economic dispatch. With the large-scale incorporation of distributed power sources such as photovoltaics, wind turbines, and various energy storage devices, as well as flexible loads such as electric vehicles on the load side into the distribution network, the randomness and volatility of the system have increased significantly. The prospect of electric vehicles as energy storage devices for the power grid in the future may become a reality as verification progresses, making the current distribution network operation state under high-ratio inverter conditions more complex and uncertain than traditional distribution networks. The direction of the power flow in the distribution network under high-ratio inverters has changed significantly from the unidirectional power flow of traditional distribution networks. Therefore, it is particularly important to study the fast and accurate fault location and island detection algorithms for distributed distribution networks under high-ratio inverters.

[0003] Grid fault location is one of the long-standing research topics in power systems. Traditional distribution network fault location methods mainly include traveling wave method, impedance method, matrix algorithm, etc. However, with the current connection of distributed power sources, the phase of voltage and current in the distribution network changes, and impedance calculation becomes difficult due to power flow changes.

[0004] After a fault occurs in the distribution network, the relevant FTUs and RTUs transmit the real-time information of the corresponding section switches and contact switches to the control center through data communication, and the fault location software is used to locate the fault. However, in reality, due to some reasons, information distortion or packet loss may occur in the real-time information, causing misjudgment of ordinary algorithms. The traditional matrix fault detection method is to establish a mathematical model suitable for the distribution network by constructing an evaluation function. Different evaluation functions have different evaluation effects and redundancy. The characteristics of this type of method are that it is difficult to construct a mathematical model, the fault tolerance of redundant judgment has a certain degree of randomness, and the robustness is poor. It is not suitable for judgment after a high proportion of inverters form an island.

[0005] Therefore, a model with higher interpretability and robustness is needed, which can correctly reflect the fault status of the distribution network when the signal is true or distorted, to meet the current distribution network scenario with a high proportion of inverters and can quickly and accurately calculate the actual fault range. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a distribution network fault section location detection method and related devices based on a hybrid of Transformer and particle swarm. The present invention introduces an intelligent algorithm on the basis of the original FTU configuration of the distribution network, supplements the missing data, and uses the particle swarm algorithm for solving, thereby improving the calculation accuracy and robustness of distribution network fault location.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A distribution network fault section location detection method based on a hybrid of Transformer and particle swarm includes the following steps:

[0009] Construct and initialize the upstream section matrix, downstream section matrix, upstream power matrix and downstream power matrix at the current moment;

[0010] Read the operating parameters of the regional distribution network and the configuration data of the measuring equipment;

[0011] The operating parameters of the distribution network and the configuration data of the measuring equipment are divided into training set, validation set and test set as the data set to train the Transformer model;

[0012] Extract the current fault information from the feeder terminal unit FTU and evaluate the integrity of the data in the fault information. If the data is missing, call the trained Transformer model to make a single-step prediction of the missing data based on historical data and gradually complete the missing data.

[0013] Based on the constructed matrix and the data obtained after the Tranformer model completes the missing data, the overall objective function and the island detection-specific objective function are constructed, and the binary particle swarm algorithm is used to solve the objective function and locate the fault section.

[0014] According to the located fault section position, the fault section is analyzed and it is determined whether the island state in the distribution network has changed. If an island is detected, the generated section is recorded and the island position support is provided for the island-specific objective function. If the fault has been restored, the island-specific objective function is deleted in the next round of calculation.

[0015] 3. Furthermore, the constructed upstream section matrix, downstream section matrix, upstream power matrix and downstream power matrix are as follows:

[0016]

[0017] in is the upstream segment matrix, is the downstream segment matrix, ∪ is the logical OR operation

[0018] Calculate; M and N are the total number of upstream and downstream power supplies respectively; The upstream power supply switching status, is the switching status of the downstream power supply; M1 and N1 are The total number of sections between upstream and downstream power sources; They are The state quantity of the section between the upstream and downstream power sources; M2 and N2 are the total number of all upstream and downstream feeder sections respectively; They are the state quantities of the upstream and downstream sections respectively.

[0019] Furthermore, the operating parameters of the regional distribution network include the loads, equipment parameters and set values ​​of the feeder terminal units FTU in the distribution network, and the configuration data of the measuring equipment include the installation location of the measuring equipment, the measurement type and the measurement data of the measuring equipment.

[0020] Furthermore, the training steps of the Transformer model include: performing random masking on the input node feature data to simulate various node data missing situations that may occur in actual applications, so that the model faces various missing patterns during the training process. The Transformer model uses a self-attention mechanism to capture the association between known nodes and missing nodes on a global scale. When the model receives masked input data, the self-attention mechanism aggregates the information of the known nodes and calculates the most likely feature estimate of the missing nodes. The input of the trained Transformer model is missing data with a random mask, and the output is complete original data.

[0021] Furthermore, the overall objective function is constructed for the entire power grid system, and the island detection dedicated objective function is to determine whether there is an island after the first round of operation of the power grid system ends. If so, the objective function is performed separately for each island.

[0022] Furthermore, the overall objective function and the island detection-specific objective function are constructed in the same way, as shown in the following formula:

[0023]

[0024] in

[0025]

[0026] Where f(L) is the deviation between the actual value of the fault location in the distribution network and the predicted value determined by the switch state function, is the predicted value; n1 represents the total number of distribution network feeder switches; n2 is the total number of distribution network feeder sections; L mis the feeder section, and the state quantity is {0, 1}; is the total number of faulty sections, λ is the anti-misjudgment coefficient, and in the minimum set concept, λ takes values ​​in [0,1]; It is the state quantity of the feeder switch. When it is 1, it means that the short-circuit current flowing through the feeder switch flows from the system power supply to the end of the line. When it is -1, it means that the short-circuit overcurrent flowing through the feeder switch flows from the end of the line to the system power supply side.

[0027] 11. Further, the determining whether an isolated island exists specifically includes:

[0028] 1. Determine the area affected by the fault

[0029] According to the fault location L f , determine the upstream section A associated with the fault location up and downstream segment A down ;

[0030] Find the segment index corresponding to the fault location from the upstream segment matrix and the downstream segment matrix;

[0031] According to the segment index, set the upstream segment set affected by the failure to A. up , the downstream segment set is A down ;

[0032] (II) Power connectivity analysis

[0033] For upstream segment set A up :

[0034] Check upstream power matrix related elements Determine the switching status of the upstream power supply connected to each upstream section;

[0035] According to the logical operation in the matrix formula, it is calculated whether each upstream section can be connected through the power supply;

[0036] For upstream sections that cannot be connected by power supply, mark them as part of the potential island area;

[0037] For the downstream segment set A down :

[0038] Check downstream power matrix related elements Determine the switching status of the downstream power supply connected to each downstream section;

[0039] According to the logical operation in the matrix formula, it is calculated whether each downstream section can be connected by power supply;

[0040] For downstream sections that cannot be connected by power supply, mark them as part of the potential island area;

[0041] (III) Determination of isolated islands

[0042] Consider the power connections of the upstream and downstream sections:

[0043] If there is a section that belongs to both the upstream potential island area and the downstream potential island area, or there is an isolated section that cannot be connected to the power supply upstream or downstream, it is determined that an island is formed;

[0044] If all affected zones can be connected to power in some way, then no islanding has occurred.

[0045] Furthermore, the method also includes the steps of: returning to the step of reading the operating parameters of the regional distribution network and the configuration data of the measuring equipment at a set time interval, and continuing the next round of fault location and detection to ensure real-time perception and fault handling of the distribution network status.

[0046] A distribution network fault section location detection system based on a mixture of Transformer and particle swarm comprises a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the distribution network fault section location detection method based on a mixture of Transformer and particle swarm.

[0047] A computer-readable storage medium stores a computer program, wherein the computer program is used to be programmed or configured by a microprocessor to execute the distribution network fault section location detection method based on a hybrid of Transformer and particle swarm.

[0048] Compared with the prior art, the present invention mainly has the following advantages:

[0049] (1) Traditional fault location methods usually require manual construction of objective functions, which is time-consuming and complicated. The present invention is based on a hybrid algorithm of Transformer and particle swarm, which can automatically construct an objective function suitable for the distribution network, greatly reducing the preliminary preparation work and improving the applicability and flexibility of the algorithm.

[0050] (2) By using the Transformer model to process distribution network data, the present invention can effectively restore missing power flow parameter data, enhance the robustness and integrity of the data, and significantly improve the accuracy of fault section location. The Transformer model analyzes historical data based on the self-attention mechanism and is retrained every 72 hours, ensuring the model's high tracking of system status.

[0051] (3) The present invention can not only quickly locate the fault section, but also reduce the number of objective function variables by constructing a dedicated objective function for the island, thereby improving the detection accuracy of the island state and avoiding the algorithm from falling into local optimality or calculation errors.

[0052] (4) Traditional fault location algorithms mostly rely on simple objective functions, while the present invention significantly improves the algorithm's ability to interpret complex power grid scenarios by combining the prediction capability of the Transformer model with the optimization capability of the particle swarm algorithm. At the same time, the present invention can still maintain a high level of robustness in the case of signal loss or distortion, and adapt to the complex distribution network environment with a high proportion of inverters.

[0053] (5) The present invention combines neural networks and intelligent optimization algorithms to perceive the operating status of the distribution network in real time and quickly solve the fault section through the particle swarm algorithm, which greatly improves the calculation efficiency and is suitable for online monitoring and fault location of actual distribution network systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the fault section location detection method of the distribution network based on the hybrid of Transformer and particle swarm of the present invention.

[0055] Figure 2 It is a schematic diagram of the improved IEEE33 node model. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] See also Figure 1 The embodiment of the present invention provides a distribution network fault section location detection method based on a hybrid of Transformer and particle swarm, comprising the following steps:

[0058] Step (1), construct and initialize the upstream segment matrix at the current moment Downstream Segment Matrix The upstream power matrix and the downstream power matrix are as follows:

[0059]

[0060] Where ∪ is a logical OR operation; M and N are the total number of upstream and downstream power supplies respectively; The upstream power supply switching status, is the switching status of the downstream power supply; M1 and N1 are The total number of sections between upstream and downstream power sources; They are The state quantity of the section between the upstream and downstream power sources; M2 and N2 are the total number of all upstream and downstream feeder sections respectively; They are the state quantities of the upstream and downstream sections respectively.

[0061] These matrices provide necessary preparation for the construction of subsequent objective functions.

[0062] Step (2), reading the operating parameters of the regional distribution network and the configuration data of the measuring equipment: the operating parameters of the regional distribution network include the load, equipment parameters and the setting value of the feeder terminal unit (FTU) in the distribution network, and the configuration data of the measuring equipment includes the installation location of the measuring equipment, the measurement type and the measurement data of the measuring equipment (such as power flow detection data).

[0063] Step (3), data preparation and model training: The operating parameters of the distribution network and the configuration data of the measuring equipment are read as data sets and divided into a training set, a validation set and a test set to train the Transformer model.

[0064] The training of the Transformer model in the present invention is as follows: In order to effectively complete the missing node data in the power system, the present invention introduces a random masking mechanism based on the original Transformer architecture. In the training stage, random masking is performed on the input node feature data, that is, some nodes are randomly selected according to a preset probability, and the feature vectors of these nodes are replaced with special mask marks (such as zero vectors or specific identifiers). This operation simulates various node data missing situations that may occur in practical applications, so that the model faces a variety of missing modes during the training process. The Transformer model uses its self-attention mechanism to capture the association between known nodes and missing nodes in a global scope. When the model receives input data containing masks, the self-attention mechanism aggregates the information of known nodes and calculates the most likely feature estimate of the missing nodes. By continuously training the model to predict the masked node features under different masking conditions, the model gradually learns and masters the ability to infer global data based on partial visible data, and finally realizes the accurate completion of node data missing in the power system due to sensor failure, communication interruption and other reasons, and improves the integrity and reliability of system data. The input of the model is missing data with random masks, and the output is complete original data.

[0065] Step (4), fault information extraction and missing data prediction: extract the current fault information from the feeder terminal unit FTU and evaluate the integrity of the data in the fault information. If the data is missing, call the trained Transformer model to perform single-step prediction on the missing data based on historical data and gradually complete the data.

[0066] When completing missing data, traditional prediction methods often have limitations when processing data with complex time series relationships. The Transformer architecture of the present invention has achieved remarkable results in fields such as natural language processing by virtue of its self-attention mechanism that can effectively capture the dependencies between elements when processing long sequence data. Applying it to single-step prediction tasks can achieve more accurate data prediction.

[0067] Step (5), objective function construction and fault location: Based on the matrix constructed in step (1) and the data obtained after the Tranformer model in step (4) completes the missing data, the overall objective function and the island detection-specific objective function are constructed. The overall and special objective function construction methods are exactly the same. The overall objective function is constructed for the entire power grid, and the island detection-specific objective function is constructed for the nodes in the identified island using the same method.

[0068] The objective function is used to express the error between the actual switch status value uploaded by the FTU and the expected value, as shown in the following formula:

[0069]

[0070] In the formula, f(L) is the deviation between the actual value of the fault location of the distribution network and the predicted value determined by the switch state function. is the predicted value, which is a multi-valued function; n1 represents the total number of feeder switches in the distribution network.

[0071] In order to prevent misjudgment or omission of the fault location in the distribution network, the concept of minimum set can be used to improve the objective function. The improved objective function is shown in the following formula.

[0072]

[0073] Where n2 is the total number of feeder sections in the distribution network; is the total number of faulty sections, λ is the anti-false positive coefficient, and in the minimum set concept, λ is in the range of [0,1] to prevent missed or false positives, and in this embodiment, it is set to 0.5.

[0074] The overall objective function and the island detection-specific objective function are used to locate the fault section and determine whether there is a new island or island recovery. The overall objective function and the island detection-specific objective function are constructed in the same way, where the overall objective function is constructed for the entire power grid system, while the island detection-specific objective function is used to determine whether there is an island after the first round of power grid system operation. If there is, the objective function is constructed for each island separately to play a redundant role. The binary particle swarm algorithm is used to solve the objective function, which can accurately locate the fault section and further determine whether there is a new island or island recovery.

[0075] The binary particle swarm algorithm (BPSO) is an optimization algorithm developed based on the standard particle swarm algorithm (PSO), which is mainly used to solve discrete optimization problems. Its core idea is to find the optimal solution by simulating the foraging behavior of bird flocks. In each iteration, particles update their positions and speeds based on their own experience and group experience.

[0076] In terms of hyperparameter selection, this embodiment uses: T max =50, N=40, ω=0.7, c1=c2=1.49618

[0077] Through the above binary particle swarm algorithm, the objective function in the graph can be effectively solved and the accurate estimation of the fault location of the distribution network can be achieved.

[0078] Step (6) Fault analysis and island status detection: According to the calculation results of the objective function, the fault section is analyzed and it is determined whether the island status in the distribution network has changed. If the formation or recovery of an island is detected, the relevant data is recorded. If an island is detected, the generated section is recorded and the island position support is provided for the island-specific objective function in step (6). If the fault has been recovered, the island-specific objective function is deleted in the next round of calculation, so as to facilitate the establishment of a dedicated objective function for each island in step (5).

[0079] The objective function is solved by the particle swarm algorithm to obtain the fault status of each branch. The faulty branch can be located by the fault status, and the following algorithm can be used to determine whether an island is formed.

[0080] 1. Determine the area affected by the fault

[0081] According to the fault location L f , determine the upstream section A associated with the fault location up and downstream segment A down ;

[0082] Find the segment index corresponding to the fault location from the upstream segment matrix and the downstream segment matrix;

[0083] According to the segment index, set the upstream segment set affected by the failure to A. up , the downstream segment set is A down .

[0084] (II) Power connectivity analysis

[0085] For upstream segment set A up :

[0086] Check upstream power matrix related elements Determine the switching status of the upstream power supply connected to each upstream section;

[0087] According to the logical operation in the matrix formula, it is calculated whether each upstream section can be connected through the power supply;

[0088] For upstream sections that cannot be connected by power supply, mark them as part of the potential island area;

[0089] For the downstream segment set A down :

[0090] Check downstream power matrix related elements Determine the switching status of the downstream power supply connected to each downstream section;

[0091] According to the logical operation in the matrix formula, it is calculated whether each downstream section can be connected by power supply;

[0092] Downstream sections that cannot be connected via power are marked as part of the potential island area.

[0093] (III) Determination of isolated islands

[0094] Consider the power connections of the upstream and downstream sections:

[0095] If there is a section that belongs to both the upstream potential island area and the downstream potential island area, or there is an isolated section that cannot be connected to the power supply upstream or downstream, it is determined that an island is formed.

[0096] If all affected zones can be connected to power in some way, then no islanding has occurred.

[0097] Step (7), polling and repeated detection: The system returns to step (2) at the set time interval and continues the next round of fault location and detection to ensure real-time perception and fault handling of the distribution network status.

[0098] The embodiment of the present invention is introduced by taking the improved IEEE33 node model as an example. Four distributed power sources are added to the IEEE33 node model, namely, 23 nodes, 24 nodes, 18 nodes, and 33 nodes. Figure 2It is represented as DG in the figure, where G is the infinite power source. The nodes are numbered from 1 to 37, and the segment number to the left of the node is the node number minus 1.

[0099] In step (1), taking the improved IEEE33 node system as an example, the system matrix is ​​read as follows:

[0100] Branch matrix:

[0101]

[0102] (1-1) The row vectors in the branch matrix B represent a set of connected nodes.

[0103] Power Matrix:

[0104] S = [1 23 24 18 33]

[0105] Each column in the power matrix S represents a power source.

[0106] The method of constructing an upstream section matrix, a downstream section matrix, an upstream power matrix, and a downstream power matrix specifically includes:

[0107] (1-2) Determine the number of branches and power sources, and initialize the upstream section matrix, downstream section matrix, upstream power matrix, and downstream power matrix.

[0108] (1-3) Traverse each branch, temporarily remove the current road for each branch to analyze connectivity. Find each connection point, determine the upstream and downstream nodes, and construct the upstream section matrix, downstream section matrix, upstream power matrix, and downstream power matrix;

[0109] (1-4) The generation principle of upstream nodes and downstream nodes is the same. The generation method is introduced by taking the upstream branch matrix as an example. The upstream matrix generation process is as follows:

[0110] 1. Create a zero matrix Au with dimension n1×n1, where n1 is the number of branches.

[0111] 2. Iterate each branch and temporarily remove the current iteration branch from the network to analyze the connectivity of the remaining network.

[0112] 3. Construct the construction diagram of the remaining branches.

[0113] 4. Find the connected components in the graph.

[0114] 5. Determine the upstream connected component and find all the nodes belonging to this component.

[0115] 6. Mark the positions of these nodes in the Au matrix. These nodes are the upstream nodes.

[0116] Because the matrix is ​​too large, only the contents of the upper left submatrix 10×10 are shown, as shown in the matrix below. Taking node 4 as an example, if the 1st, 2nd, and 3rd columns of the 4th row of the matrix are 1, then these three branches are the upstream branches of node 4.

[0117]

[0118] The generation principles of the upstream power matrix and the downstream power matrix are similar. The generation method is introduced by taking the upstream power matrix as an example. The specific process is as follows:

[0119] 1. Create a zero matrix Ku with dimensions m1×n2. n2 is the number of branches and m1 is the number of power supplies.

[0120] 2. For each power utilization matrix Au, detect the connected components in which it is located.

[0121] 3. If the power supply is connected to a heavy component upstream, set the corresponding position of Ku to 1.

[0122] Since the matrix is ​​too large, only the matrix of nodes 9-18 is shown.

[0123]

[0124] The embodiment of the present invention is introduced by taking the improved IEEE33 node model as an example. Four distributed power sources are added to the IEEE33 node model, namely, 23 nodes, 24 nodes, 18 nodes, and 33 nodes. Figure 2 It is represented as DG in the figure, where G is the infinite power source. The nodes are numbered from 1 to 37, and the segment number to the left of the node is the node number minus 1.

[0125] Step (2) Reading distribution network parameters and measurement configuration: First, by reading the regional parameters and topological structure of the distribution network, obtain the relevant information of the Supervisory Control And Data Acquisition (SCADA) configuration, including the load, equipment parameters and the setting value of the feeder terminal unit FTU in the distribution network. At the same time, read the installation location, measurement type and measurement data of the measurement equipment to provide data support for subsequent steps. At the same time, establish the association matrix between the distribution network nodes.

[0126] In step (3), it specifically includes:

[0127] Read the voltage and power data collected by each node at the current moment, and use the data of the previous 72 hours to form a single-step prediction data set. Perform classification data training and classify the n-1, n-2, n-3, and n-4 situations in the collected signals.

[0128] The model was retrained every 24 hours, and each model was retrained again after 72 hours of training.

[0129] The Transformer model processes the linear and nonlinear components of the measured values ​​through the self-attention mechanism, and optimizes the state estimation process by combining historical state information. The self-attention formula in the Transformer model is expressed as:

[0130]

[0131] In formula (1), Q is the query matrix; K is the key matrix; V is the value matrix; d k is the dimension of the key vector, which is used to scale the dot product to prevent the dot product from being too large, resulting in the gradient of the softmax function being too small; softmax is the activation function, which is used to calculate the weight.

[0132] Step (4): Determine whether the fault information matrix is ​​complete and select the corresponding trained model to perform single-step prediction on the missing parameters.

[0133] The data generated by FTU is judged. If there is data loss, the Transformer model trained in step (4) is called for single-step prediction. Assuming that the model is missing 4 sets of FTU data (setting judgment and power flow information are all missing), n-4 prediction is used, and in subsequent predictions it becomes n-3 model. After n-3 prediction, n-2 model is obtained. In this way, complete data information can be obtained recursively.

[0134] Then, according to the parameters selected during setting, the actual value is calculated and compared with the set value to obtain the simulated FTU fault signal and the original state quantity, which are then substituted into step (6) for calculation.

[0135] If the judgment matrix is ​​not missing, skip this step.

[0136] Step (5): Construct the overall objective function and the objective function of the island part ()

[0137] Using the island positions obtained in the previous round, a dedicated objective function for island detection is constructed to improve positioning accuracy.

[0138] Feeder switch status And when it is 1, it means that the current flows through the feeder switch The short-circuit current flows from the system power supply to the end of the line. When it is -1, it means that it flows through the feeder switch. The short-circuit overcurrent flows from the end of the line to the system power supply side, L m(m=1,2,3,4,…,36) is the feeder section, and its state quantity is {0,1}. The feeder switch state function is the distribution network defective line segment state L=[L1 L2…L 36 ], reflecting the distribution network section L m (m=1,2,3,4,…,36) operating status.

[0139] Using the matrix constructed from upstream and downstream, the switch state function of the upstream area containing the system power supply is The switch state function of the downstream section is: Construct the following objective function:

[0140]

[0141] in

[0142]

[0143] Where f(L) is the deviation between the actual value of the fault location in the distribution network and the predicted value determined by the switch state function, is the predicted value; n1 represents the total number of distribution network feeder switches; n2 is the total number of distribution network feeder sections; L m is the feeder section, and the state quantity is {0, 1}; is the total number of faulty sections, λ is the anti-misjudgment coefficient, and in the minimum set concept, λ takes values ​​in [0,1]; It is the state quantity of the feeder switch. When it is 1, it means that the short-circuit current flowing through the feeder switch flows from the system power supply to the end of the line. When it is -1, it means that the short-circuit overcurrent flowing through the feeder switch flows from the end of the line to the system power supply side.

[0144] Finally, the binary particle swarm method is used to solve the objective function The fault location can be obtained by calculation. The example matrix below is the simulated FTU signal and particle swarm calculation results.

[0145] Simulate FTU signal matrix (1-36 node one-dimensional vector):

[0146] [1 -1-1 -1-1 -1-1 -1-1 -1 1 1 1 1 -1-1 -1-1 -1-1 -1-1 -1-1 -1-1 -1-1 -1-1 -1-1 -1 1]' The output is:

[0147] [10000100000001000000000000000000] T This indicates that faults 1, 6, and 14 are as expected.

[0148] If there are new or fewer fault points, determine whether the node has formed an island or recovered from an island. Save data to

[0149] This is convenient for use in step (5) in the next round of calculations.

[0150] Step (6) Fault analysis and island status detection: According to the calculation results of the objective function, analyze the fault section and determine whether the island status in the distribution network has changed. If the formation or recovery of an island is detected, record the relevant data to provide a basis for subsequent calculations. If an island is detected, record the generated section and provide island location support for the island-specific objective function in step 5, so that step 5 can detect each island.

[0151] Island establishes a dedicated objective function.

[0152] Step (7) Polling and repeated detection: The system returns to step (2) at the set time interval and continues the next round of fault location and detection to ensure real-time perception and fault handling of the distribution network status.

[0153] In addition, an embodiment of the present invention also provides a distribution network fault section location detection system based on a hybrid of Transformer and particle swarm, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to perform the distribution network fault section location detection based on a hybrid of Transformer and particle swarm.

[0154] method.

[0155] In addition, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is used to be programmed or configured by a microprocessor to execute the

[0156] This paper describes a distribution network fault section location detection method based on a hybrid of Transformer and particle swarm.

[0157] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A distribution network fault section location detection method based on a hybrid of Transformer and particle swarm, characterized in that: The steps include: Construct and initialize the upstream section matrix, downstream section matrix, upstream power matrix and downstream power matrix at the current moment; Read the operating parameters of the regional distribution network and the configuration data of the measuring equipment; The operating parameters of the distribution network and the configuration data of the measuring equipment are divided into training set, validation set and test set as the data set to train the Transformer model; Extract the current fault information from the feeder terminal unit FTU and evaluate the integrity of the data in the fault information. If the data is missing, call the trained Transformer model to make a single-step prediction of the missing data based on historical data and gradually complete the missing data. Based on the constructed matrix and the data obtained after the Tranformer model completes the missing data, the overall objective function and the island detection-specific objective function are constructed, and the binary particle swarm algorithm is used to solve the objective function and locate the fault section. According to the located fault section position, the fault section is analyzed and it is determined whether the island state in the distribution network has changed. If an island is detected, the generated section is recorded and the island position support is provided for the island-specific objective function. If the fault has been restored, the island-specific objective function is deleted in the next round of calculation.

2. The distribution network fault section location detection method based on the hybrid of Transformer and particle swarm as claimed in claim 1 is characterized by: The constructed upstream section matrix, downstream section matrix, upstream power matrix and downstream power matrix are as follows: in is the upstream segment matrix, is the downstream section matrix; ∪ is the logical OR operation; M and N are the total number of upstream and downstream power supplies respectively; The upstream power supply switching status, is the switching status of the downstream power supply; M1 and N1 are The total number of sections between upstream and downstream power sources; They are The state quantity of the section between the upstream and downstream power sources; M2 and N2 are the total number of all upstream and downstream feeder sections respectively; They are the state quantities of the upstream and downstream sections respectively.

3. The fault section location detection method of distribution network based on hybrid Transformer and particle swarm as claimed in claim 1 is characterized by: The operating parameters of the regional distribution network include the loads and equipment parameters in the distribution network and the setting values ​​of the feeder terminal units FTU, and the configuration data of the measuring equipment include the installation location of the measuring equipment, the measurement type and the measurement data of the measuring equipment.

4. The fault section location detection method of distribution network based on hybrid Transformer and particle swarm as claimed in claim 1, characterized in that: The training steps of the Transformer model include: performing random masking on the input node feature data to simulate various node data missing situations that may occur in actual applications, so that the model faces various missing modes during the training process. The Transformer model uses a self-attention mechanism to capture the association between known nodes and missing nodes on a global scale. When the model receives masked input data, the self-attention mechanism aggregates the information of the known nodes and calculates the most likely feature estimation of the missing nodes. The input of the trained Transformer model is missing data with random masks, and the output is complete original data.

5. The distribution network fault section location detection method based on the hybrid of Transformer and particle swarm as claimed in claim 2 is characterized by: The overall objective function is to construct the entire power grid system, and the island detection dedicated objective function is to determine whether there is an island after the first round of power grid system operation. If so, the objective function is performed separately for each island.

6. The fault section location detection method of distribution network based on hybrid Transformer and particle swarm as claimed in claim 5, characterized in that: The overall objective function and the island detection dedicated objective function are constructed in the same way, as shown in the following formula: in Where f(L) is the deviation between the actual value of the fault location in the distribution network and the predicted value determined by the switch state function, is the predicted value; n1 represents the total number of distribution network feeder switches; n2 is the total number of distribution network feeder sections; L m is the feeder section, and the state quantity is {0, 1}; is the total number of faulty sections, λ is the coefficient for preventing misjudgment, and in the concept of the minimum set, λ takes values ​​in [0,1]; It is the state quantity of the feeder switch. When it is 1, it means that the short-circuit current flowing through the feeder switch flows from the system power supply to the end of the line. When it is -1, it means that the short-circuit overcurrent flowing through the feeder switch flows from the end of the line to the system power supply side.

7. The fault section location detection method of distribution network based on hybrid Transformer and particle swarm as claimed in claim 6, characterized in that: The determining whether there is an isolated island specifically includes:

1. Determine the area affected by the fault According to the fault location L f , determine the upstream section A associated with the fault location up and downstream segment A down ; Find the segment index corresponding to the fault location from the upstream segment matrix and the downstream segment matrix; According to the segment index, set the upstream segment set affected by the failure to A. up , the downstream segment set is A down ; (II) Power connectivity analysis For upstream segment set A up : Check upstream power matrix related elements Determine the switching status of the upstream power supply connected to each upstream section; According to the logical operation in the matrix formula, it is calculated whether each upstream section can be connected through the power supply; For upstream sections that cannot be connected by power supply, mark them as part of the potential island area; For the downstream segment set A down : Check downstream power matrix related elements Determine the switching status of the downstream power supply connected to each downstream section; According to the logical operation in the matrix formula, it is calculated whether each downstream section can be connected by power supply; For downstream sections that cannot be connected by power supply, mark them as part of the potential island area; (III) Determination of isolated islands Consider the power connections of the upstream and downstream sections: If there is a section that belongs to both the upstream potential island area and the downstream potential island area, or there is an isolated section that cannot be connected to the power supply upstream or downstream, it is determined that an island is formed; If all affected zones can be connected to power in some way, then no islanding has occurred.

8. The distribution network fault section location detection method based on the hybrid of Transformer and particle swarm as claimed in claim 1 is characterized by: The method also includes the steps of: returning to the step of reading the operating parameters of the regional distribution network and the configuration data of the measuring equipment at a set time interval, and continuing the next round of fault location and detection to ensure real-time perception and fault handling of the distribution network status.

9. A distribution network fault section location detection system based on a hybrid of Transformer and particle swarm, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the distribution network fault section location detection method based on a hybrid of Transformer and particle swarm according to any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, wherein the computer program is used to be programmed or configured by a microprocessor to execute the distribution network fault section location detection method based on a hybrid of Transformer and particle swarm according to any one of claims 1 to 8.

Images