Power distribution network fault accurate positioning method based on graph neural network double-layer model under novel power system

Abstract

The invention discloses a novel power distribution network fault accurate positioning method based on a graph neural network double-layer model under a power system. The method comprises the following steps: 1) obtaining power distribution network graph data, FTU alarm information and distribution transformer voltage loss alarm information; 2) constructing a double-layer power distribution network fault positioning model which comprises a first-layer power distribution network fault positioning model and a second-layer power distribution network fault positioning model; 3) obtaining fault bus position information A and a fault section positioning result B; and 4) fusing the fault section positioning result A and the fault section positioning result B, and outputting a final power distribution network fault positioning result by using user complaint information as an auxiliary criterion. According to the invention, accurate positioning of the power distribution network fault can be realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent distribution network and fault diagnosis, and specifically to a method for accurately locating distribution network faults based on a graph neural network double-layer model in a new type of power system. Background Art

[0002] As an important part of modern industrial and commercial production and people's daily life, the power system plays a vital role in ensuring the stable and rapid development of my country. With the rapid expansion of urban scale and the continuous advancement of electric energy substitution, the demand for electricity in all walks of life has grown rapidly, and the dependence on electricity supply has become increasingly higher. In the process of power generation, transmission and distribution of the power system, the distribution network, as the "last mile" in the power supply link, realizes the connection between the large power grid and users. The reliability of the distribution network has the most direct impact on user satisfaction. According to statistics on the failure of power grid equipment in recent years, distribution network failures account for 80% to 90% of the total failures of the power system. Accurate and rapid positioning of distribution network faults is the prerequisite for subsequent timely isolation of faults and restoration of power supply. It can reduce power outage time and user losses, improve the reliability of power supply of distribution networks, and play a vital role in improving user satisfaction and the benefits of power supply companies.

[0003] As a key system for local distribution of electric energy, the distribution network usually adopts a closed-loop design and open-loop operation mode, and is composed of a large number of switches. There are many types of faults in the distribution network, and its wiring branches and equipment are very complex. In addition, the transition resistance is greatly affected by the environment during a fault. Therefore, it is very difficult to locate the fault, and it takes a long time to determine the location of the fault. Due to the rise of new energy, many distributed power sources have been added to the distribution network. After the distribution network fails, the fault location cannot be simply determined based on the traditional fault current direction. The actual fault conditions of the new power system are more complex and changeable, which increases the difficulty of fault location.

[0004] Traditional ranging positioning methods have slow positioning speed, low real-time performance, and high manpower and material costs. In contrast, artificial intelligence-based methods have great advantages in fault location accuracy and efficiency. They can use a large amount of historical data and real-time monitoring data to conduct autonomous learning and optimization, thereby obtaining more accurate and real-time fault location information. Graph neural networks have the advantages of comprehensive information utilization, spatial correlation modeling, nonlinear feature extraction, and strong adaptability in distribution network fault location. They can effectively process complex distribution network data, improve the accuracy and efficiency of fault location, and provide important support for the stable operation and reliability of distribution networks. At present, most existing studies use graph neural networks alone for bus-level fault location, and their accuracy still needs to be improved. Summary of the invention

[0005] The purpose of the present invention is to provide a method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, comprising the following steps:

[0006] 1) Obtain distribution network diagram data, FTU alarm information, and distribution transformer undervoltage alarm information;

[0007] 2) Constructing a two-layer distribution network fault location model, including a first-layer distribution network fault location model and a second-layer distribution network fault location model;

[0008] 3) Execute step 3-1) and step 3-2) in no particular order;

[0009] 3-1) Input the distribution network diagram data and FTU alarm information into the first-layer distribution network fault location model to obtain the fault bus location information A;

[0010] Then, the fault bus position information A and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result A;

[0011] 3-2) Input the distribution network diagram data and the distribution transformer undervoltage alarm information into the first-layer distribution network fault location model to obtain the fault bus location information B;

[0012] Then, the fault bus position information B and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result B;

[0013] 4) Fuse the fault section location results A and B, and use user complaint information as an auxiliary criterion to output the final distribution network fault location results.

[0014] Furthermore, the distribution network map data includes topological structure information and node attribute information in the distribution network topological map.

[0015] Furthermore, in the distribution network topology diagram, nodes and edges represent busbars and feeder branches, respectively.

[0016] Furthermore, the topological structure information is represented by the adjacency matrix A of the nodes V and the edges E, namely:

[0017]

[0018] In the formula, v i 、v j Represents a node. ij are the elements of the adjacency matrix A.

[0019] Furthermore, the node attribute information includes the amplitude and phase angle of the three-phase voltage of the line, the amplitude and phase angle of the three-phase current, and the node center characteristic information, which is recorded as The node center characteristic information includes degree centrality DC, eigenvector centrality EC, betweenness centrality BC and closeness centrality CC. V1, V2, and V3 are the amplitudes of the three-phase voltages; is the phase angle of the three-phase voltage; I1, I2, I3 are the amplitudes of the three-phase current; is the phase angle of the three-phase current.

[0020] Further, the step of obtaining the fault bus position information A includes:

[0021] a1) According to the fault section location result A, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0022] a2) Construct the real overcurrent signal expression I j ,Right now:

[0023]

[0024] a3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain:

[0025]

[0026] In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, respectively; P and Q are the numbers of feeder sections in the upstream and downstream of the switch, respectively; i1 and i2 are the current values;

[0027] a4) Selecting the fitness function, transforming the fault location problem into a problem of solving the minimum value of the objective function (4), and using the particle swarm optimization algorithm to solve it, obtaining the fault bus location information A;

[0028] The fitness function is as follows:

[0029]

[0030] In the formula, x i represents the state of faulty segment i, 0 represents non-faulty, 1 represents faulty; ω is the weight of the faulty segment summary. F(x) represents the fitness function; M represents the number of faulty segments.

[0031] Further, the step of obtaining the fault bus position information B includes:

[0032] b1) According to the fault section location result B, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0033] b2) Construct the real overcurrent signal expression I j ,Right now:

[0034]

[0035] b3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain:

[0036]

[0037] In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, P and Q are the feeder section numbers in the upstream and downstream of the switch, i1 and i2 are the current values; I * j (x) is a logic function representing the on / off status of the distribution network equipment; They are the status of the upstream and downstream distributed power sources in section i, where connected is 1 and not connected is 0;

[0038] b4) Selecting the fitness function, transforming the fault location problem into a problem of solving the minimum value of the objective function (4), and solving it using the particle swarm optimization algorithm to obtain the fault bus location information B;

[0039] The fitness function is as follows:

[0040]

[0041] In the formula, x i represents the state of faulty segment i, 0 represents non-faulty, 1 represents faulty; ω is the weight of the faulty segment summary. F(x) represents the fitness function; M represents the number of faulty segments.

[0042] Furthermore, the fault section location result A and the fault section location result B are fused through the DS evidence theory.

[0043] Furthermore, the user complaint information includes complaint information of downstream users in the fault section, complaint information of downstream users in the non-fault section, and fault information of the transformer to which the user belongs.

[0044] Furthermore, using user complaint information as an auxiliary criterion, the step of outputting the final distribution network fault location result includes: parsing the user complaint information, determining whether the fault location of the user complaint is within the interval corresponding to the location result, and if so, using the current location result as the final distribution network fault location result; if not, returning to step 3) and recalculating the fault section location result A and the fault section location result B.

[0045] The technical effect of the present invention is unquestionable. The present invention provides a method for accurately locating distribution network faults based on a two-layer model of a graph neural network under a new type of power system. The method constructs a distribution network fault location model based on a physics-preserving graph neural network (PPGN), introduces physical equations, reduces the model's dependence on samples, realizes layer-by-layer transmission of features, completes graph feature extraction from local to overall, and converts the distribution network fault location problem into a node classification problem, locating the fault to the bus. Then, based on the idea of ​​hierarchical partitioning, the fault section of the distribution network is further refined to consider the situation where the flow is reversed due to the access of distributed power sources, and more accurate section positioning is performed in combination with the line location results of the PPGN model. Finally, consider using user complaint information as an auxiliary criterion, integrating the analysis and judgment results, and realizing accurate location of distribution network faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a two-layer model of graph neural network;

[0047] Figure 2 Partitioning for the maximum voltage support range of distributed power sources;

[0048] Figure 3 The fault section location process for active distribution network;

[0049] Figure 4 Provide a flow chart for accurately locating distribution network faults under the new power system;

[0050] Figure 5 It is the topology diagram of IEEE123 node distribution network;

[0051] Figure 6 This is the effect diagram of fault section location based on particle swarm optimization algorithm. DETAILED DESCRIPTION

[0052] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0053] Embodiment 1:

[0054] See also Figures 1 to 6 , a new distribution network fault accurate location method based on a graph neural network double-layer model under a new power system, including the following steps:

[0055] 1) Obtain distribution network diagram data, FTU alarm information, and distribution transformer undervoltage alarm information;

[0056] 2) Constructing a two-layer distribution network fault location model, including a first-layer distribution network fault location model and a second-layer distribution network fault location model;

[0057] 3) Execute step 3-1) and step 3-2) in no particular order;

[0058] 3-1) Input the distribution network diagram data and FTU alarm information into the first-layer distribution network fault location model to obtain the fault bus location information A;

[0059] Then, the fault bus position information A and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result A;

[0060] 3-2) Input the distribution network diagram data and the distribution transformer undervoltage alarm information into the first-layer distribution network fault location model to obtain the fault bus location information B;

[0061] Then, the fault bus position information B and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result B;

[0062] 4) Fuse the fault section location results A and B, and use user complaint information as an auxiliary criterion to output the final distribution network fault location results.

[0063] The distribution network map data includes topological structure information and node attribute information in the distribution network topological map.

[0064] In the distribution network topology diagram, nodes and edges represent busbars and feeder branches respectively.

[0065] The topological structure information is represented by the adjacency matrix A of the nodes V and the edges E, namely:

[0066]

[0067] In the formula, v i 、v j Represents a node. ij are the elements of the adjacency matrix A.

[0068] The node attribute information includes the amplitude and phase angle of the three-phase voltage of the line, the amplitude and phase angle of the three-phase current, and the node center characteristic information, which is recorded as The node center characteristic information includes degree centrality DC, eigenvector centrality EC, betweenness centrality BC and closeness centrality CC. V1, V2, and V3 are the amplitudes of the three-phase voltages; is the phase angle of the three-phase voltage; I1, I2, I3 are the amplitudes of the three-phase current; is the phase angle of the three-phase current.

[0069] The steps of obtaining the fault bus position information A include:

[0070] a1) According to the fault section location result A, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0071] a2) Construct the real overcurrent signal expression I j ,Right now:

[0072]

[0073] a3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain:

[0074]

[0075] In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, P and Q are the feeder section numbers in the upstream and downstream of the switch, i1 and i2 are the current values; I * j (x) is a logic function representing the on / off status of the distribution network equipment; They are the status of the upstream and downstream distributed power sources in section i, where connected is 1 and not connected is 0;

[0076] a4) Selecting the fitness function, transforming the fault location problem into a problem of solving the minimum value of the objective function (4), and using the particle swarm optimization algorithm to solve it, obtaining the fault bus location information A;

[0077] The fitness function is as follows:

[0078]

[0079] In the formula, x i represents the state of faulty segment i, 0 represents non-faulty, 1 represents faulty; ω is the weight of the faulty segment summary. F(x) represents the fitness function; M represents the number of faulty segments.

[0080] The steps of obtaining the fault bus position information B include:

[0081] b1) According to the fault section location result B, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0082] b2) Construct the real overcurrent signal expression I j ,Right now:

[0083]

[0084] b3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain:

[0085]

[0086] In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, P and Q are the feeder section numbers in the upstream and downstream of the switch, i1 and i2 are the current values; I * j (x) is a logic function representing the on / off status of the distribution network equipment; They are the status of the upstream and downstream distributed power sources in section i, where connected is 1 and not connected is 0;

[0087] b4) Selecting the fitness function, transforming the fault location problem into a problem of solving the minimum value of the objective function (4), and solving it using the particle swarm optimization algorithm to obtain the fault bus location information B;

[0088] The fitness function is as follows:

[0089]

[0090] In the formula, x i represents the state of faulty segment i, 0 represents non-faulty, 1 represents faulty; ω is the weight of the faulty segment summary. F(x) represents the fitness function; M represents the number of faulty segments.

[0091] The fault section location result A and the fault section location result B are fused through the DS evidence theory.

[0092] The user complaint information includes complaint information of downstream users in the fault section, complaint information of downstream users in the non-fault section, and fault information of the transformer to which the user belongs.

[0093] The steps of outputting the final distribution network fault location result using the user complaint information as an auxiliary criterion include: parsing the user complaint information to determine whether the fault location of the user complaint is within the interval corresponding to the location result; if so, using the current location result as the final distribution network fault location result; if not, returning to step 3) to recalculate the fault section location result A and the fault section location result B.

[0094] Embodiment 2:

[0095] A new distribution network fault accurate location method based on a graph neural network double-layer model in a new power system includes the following steps:

[0096] 1) Obtain distribution network diagram data, FTU alarm information, and distribution transformer undervoltage alarm information;

[0097] 2) Constructing a two-layer distribution network fault location model, including a first-layer distribution network fault location model and a second-layer distribution network fault location model;

[0098] 3) Execute step 3-1) and step 3-2) in no particular order;

[0099] 3-1) Input the distribution network diagram data and FTU alarm information into the first-layer distribution network fault location model to obtain the fault bus location information A;

[0100] Then, the fault bus position information A and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result A;

[0101] 3-2) Input the distribution network diagram data and the distribution transformer undervoltage alarm information into the first-layer distribution network fault location model to obtain the fault bus location information B;

[0102] Then, the fault bus position information B and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result B;

[0103] 4) Fuse the fault section location results A and B, and use user complaint information as an auxiliary criterion to output the final distribution network fault location results.

[0104] Embodiment 3:

[0105] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content of which is the same as that of Example 2. Furthermore, the distribution network graph data includes topological structure information and node attribute information in the distribution network topology graph.

[0106] Embodiment 4:

[0107] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content of which is the same as any one of Examples 2-3. Furthermore, in the distribution network topology diagram, nodes and edges represent busbars and feeder branches, respectively.

[0108] Embodiment 5:

[0109] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content of which is the same as any one of Embodiments 2-4, and further, the topological structure information is represented by an adjacency matrix A of nodes V and edges E, that is:

[0110]

[0111] In the formula, v i 、v j Represents a node.

[0112] Embodiment 6:

[0113] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content is the same as any one of embodiments 2-5, further, the node attribute information includes the amplitude and phase angle of the three-phase voltage of the line, the amplitude and phase angle of the three-phase current, and the node center characteristic information, recorded as The node centrality characteristic information includes degree centrality DC, eigenvector centrality EC, betweenness centrality BC and closeness centrality CC.

[0114] Embodiment 7:

[0115] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content of which is the same as any one of Embodiments 2-6, and further, the step of obtaining fault bus location information A includes:

[0116] 1) According to the fault section location result A, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0117] 2) Construct the real overcurrent signal expression, that is:

[0118]

[0119] 3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain:

[0120]

[0121] In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), xj (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, P and Q are the feeder section numbers in the upstream and downstream of the switch, i1 and i2 are the current values; I * j (x) is a logic function representing the on / off status of the distribution network equipment; are the status of the distributed power sources upstream and downstream of section i, respectively, where connected is 1 and not connected is 0;

[0122] 4) Select the fitness function, transform the fault location problem into a problem of solving the minimum value of the objective function (4), and use the particle swarm optimization algorithm to solve it, and obtain the fault bus location information A;

[0123] The fitness function is as follows:

[0124]

[0125] In the formula, x i represents the state of fault segment i, 0 represents non-fault and 1 represents fault; ω is the weight of the fault segment summary.

[0126] Embodiment 8:

[0127] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content of which is the same as any one of Embodiments 2-7, and further, the step of obtaining the fault bus location information B includes:

[0128] 1) According to the fault section location result B, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0129] 2) Construct the real overcurrent signal expression, that is:

[0130]

[0131] 3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain:

[0132]

[0133] In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, respectively; P and Q are the numbers of feeder sections in the upstream and downstream of the switch, respectively; i1 and i2 are the current values;

[0134] 4) Select the fitness function, transform the fault location problem into a problem of solving the minimum value of the objective function (4), and use the particle swarm optimization algorithm to solve it, and obtain the fault bus location information B;

[0135] The fitness function is as follows:

[0136]

[0137] In the formula, x i represents the state of fault segment i, 0 represents non-fault and 1 represents fault; ω is the weight of the fault segment summary.

[0138] Embodiment 9:

[0139] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new power system, the technical content of which is the same as any one of Examples 2-8. Furthermore, the fault section location result A and the fault section location result B are fused through the DS evidence theory.

[0140] Embodiment 10:

[0141] A method for accurately locating distribution network faults based on a two-layer graph neural network model in a new type of power system, the technical content of which is the same as any one of Examples 2-9. Furthermore, the user complaint information includes complaint information of downstream users in the faulty section, complaint information of downstream users in the non-faulty section, and fault information of the transformer to which the user belongs.

[0142] Embodiment 11:

[0143] A method for accurately locating distribution network faults based on a graph neural network double-layer model in a new type of power system, the technical content of which is the same as any one of Examples 2-10. Furthermore, the step of outputting the final distribution network fault location result using user complaint information as an auxiliary criterion includes: parsing the user complaint information to determine whether the fault location of the user complaint is within the interval corresponding to the location result, and if so, using the current location result as the final distribution network fault location result; if not, returning to step 3) to recalculate the fault section location result A and the fault section location result B.

[0144] Embodiment 12:

[0145] A new distribution network fault accurate location method based on a graph neural network double-layer model in a new power system includes the following steps:

[0146] 1) Abstract the distribution network into graph data;

[0147] 2) Extract the structural information of the graph and the attribute information of the nodes as the input of the model;

[0148] 3) The first layer of the two-layer model: construct a distribution network fault location model based on the physics-preserving graph neural network (PPGN), introduce physical equations, reduce the model's dependence on samples, realize the layer-by-layer transmission of features, complete the graph feature extraction from local to global, transform the distribution network fault location problem into a node classification problem, and locate the fault to the bus.

[0149] 4) The second layer of the two-layer model: locating the fault section of the distribution network based on the optimization algorithm. Based on the idea of ​​hierarchical partitioning, the fault section of the distribution network is further refined (such as Figure 2 As shown in the figure), considering the reverse flow caused by the access of distributed power sources, more accurate section positioning is performed in combination with the line positioning results of the PPGN model. (The second layer judgment process of the two-layer model is as follows Figure 3 (shown)

[0150] 5) Input the FTU alarm information and graph data into the two-layer model to obtain the fault location result A, input the distribution transformer undervoltage alarm information and graph data into the two-layer model to obtain the fault location result B, and fuse the location results A and B based on the DS evidence theory. Use user complaint information as an auxiliary criterion, fuse the judgment results, and output the fault location result. (A new distribution network fault accurate location method based on the graph neural network two-layer model under the power system The overall process is as follows Figure 4 (shown)

[0151] The distribution network is abstracted into topological graph data, where nodes and edges represent busbars and feeder branches respectively.

[0152] The structural information of the graph and the attribute information of the nodes are extracted as the input of the first layer of the two-layer model, and the bus level positioning is realized by using PPGN. The main steps are as follows:

[0153] a1) Assume that the distribution network graph data is G = (V, E), where the nodes V and edges E in the graph represent the busbar and feeder branches respectively. Construct the adjacency matrix A, and the calculation formula is as follows:

[0154]

[0155] a2) Obtain the attribute information of the node. The main steps are as follows:

[0156] a2.1) The amplitude and phase angle of the three-phase voltage and the amplitude and phase angle of the three-phase current of the line are detected by FTU;

[0157] a2.2) Four central properties of nodes calculated based on graph theory, namely degree centrality, eigenvector centrality, betweenness centrality and closeness centrality;

[0158] a2.3) Combine steps 2.1 and 2.2 to express the node's feature vector X as

[0159]

[0160] a2.4) The adjacency matrix A of the distribution network graph G and the eigenvector matrix X of the nodes are combined as the input of the first layer PPGN of the two-layer model, and the distribution network fault location is converted into a classification problem to locate the faulty bus.

[0161] Combined with the positioning results of the first layer model and the on / off status of the distribution network equipment switches, the area to which the distribution network fault line belongs is divided into layers and partitions, and a more accurate fault section positioning is achieved through the particle swarm optimization algorithm. The main steps are as follows:

[0162] b1) Based on the output results of the first-layer model, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point;

[0163] b2) Use logic functions to represent the opening and closing states of distribution network equipment (circuit breakers, switches, etc.). The overcurrent signal is expressed as:

[0164]

[0165] b3) For a total of M feeder sections in the distribution network, a variable x ranging from 0 to 1 is introduced i (i=1,2,...,M) represents the state of fault segment i, 0 represents non-fault, 1 represents fault, are the status of the upstream and downstream distributed power sources in section i, respectively. Connected is 1, and not connected is 0. are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections upstream and downstream of switch j, respectively, and P and Q are the number of feeder sections upstream and downstream of the switch, respectively. The calculation formula of the switch function is:

[0166]

[0167] Divide the area according to the maximum voltage support range of the distributed power source to improve the efficiency of the subsequent fault section location algorithm;

[0168] The fitness function is selected to transform the fault location problem into a problem of solving the minimum value of the objective function (4), and the particle swarm optimization algorithm is used to solve it and obtain the final fault section. j is the real overcurrent signal of FTU, ω is the weight of the fault section summary, and the fitness function calculation formula is:

[0169]

[0170] Input the FTU alarm information and graph data into the double-layer model to obtain the fault location result A, input the distribution transformer undervoltage alarm information and graph data into the double-layer model to obtain the fault location result B, fuse the location results A and B, and use the user complaint information as an auxiliary criterion to output the fault location result. The main steps are as follows:

[0171] Input the FTU alarm information into the double-layer model to obtain the fault location result A, and input the FTU alarm information into the double-layer model to obtain the fault location result B;

[0172] Based on DS evidence theory, the positioning results A and B are integrated;

[0173] Use user complaint information as auxiliary criteria, integrate the analysis results, and output the final fault location results.

[0174] Embodiment 13:

[0175] The verification of a distribution network fault accurate location method based on a two-layer graph neural network model in a new power system is as follows:

[0176] The IEEE123 node distribution network is abstracted into graph data G = (V, E). The specific abstracted graph structure type is an undirected unweighted isomorphic graph. The nodes V and edges E in the graph represent the busbar and feeder branch, respectively. The topology diagram of the IEEE123 node distribution network is as follows: Figure 5 shown.

[0177] The structural information of the graph and the attribute information of the nodes are extracted as the input of the model.

[0178] Extract the structural information of the graph and the attribute information of the nodes. The main steps are as follows:

[0179] 2.1) For the distribution network graph data G = (V, E), construct the adjacency matrix A, and the calculation formula is as follows:

[0180]

[0181] 2.2) Get the attribute information of the node. The main steps are as follows:

[0182] 2.2.1) The amplitude and phase angle of the three-phase voltage of the line are detected by FTU Amplitude and phase angle of three-phase current

[0183] 2.2.2) The four central characteristics of nodes calculated based on graph theory are degree centrality DC, eigenvector centrality EC, betweenness centrality BC and closeness centrality CC.

[0184] 2.2.3) Combine steps 2.1 and 2.2 to express the node's feature vector X as

[0185] 3) The first layer of the two-layer model: construct a distribution network fault location model based on physical preserving graph neural network (PPGN) to locate the fault to the bus. The specific steps are as follows:

[0186] 3.1) Introduce physical equations:

[0187] Y ij ΔU=ΔC+Δ ij

[0188] Among them, Y ij ∈C 3×3 is the admittance matrix between node i and node j before the fault occurs, ΔU = U-U0, ΔC = C-C0 represent the voltage and current changes caused by the fault, respectively. The learning strategy benefits by extracting the relationship between the partial weighted voltage and the fault location, thereby reducing the model's dependence on samples.

[0189] 3.2) The adjacency matrix A of the distribution network graph G and the feature vector matrix X of the nodes are combined as the input of the first layer PPGN of the two-layer model. Through local aggregation and global transformation, graph embedding learning is completed, and the distribution network fault location is transformed into a classification problem to locate the faulty bus.

[0190] 4) The second layer of the two-layer model: locating the fault section of the distribution network based on the optimization algorithm. The specific steps are as follows:

[0191] 4.1) Based on the output results of the first-layer model, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas using the sectionalizer as the area demarcation point.

[0192] 4.2) Use logic functions to represent the opening and closing states of distribution network equipment (circuit breakers, switches, etc.), and the overcurrent signal is expressed as:

[0193]

[0194] 4.3) For a total of M feeder sections in the distribution network, a variable x ranging from 0 to 1 is introduced i (i=1,2,...,M) represents the state of fault segment i, 0 represents non-fault, 1 represents fault, are the status of the upstream and downstream distributed power sources in section i, respectively. Connected is 1, and not connected is 0. are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections upstream and downstream of switch j, respectively, and P and Q are the number of feeder sections upstream and downstream of the switch, respectively. The calculation formula of the switch function is:

[0195]

[0196] 4.4) Divide the regions according to the maximum voltage support range of the distributed power sources;

[0197] 4.5) Select the fitness function and transform the fault location problem into the problem of solving the minimum value of the objective function. j is the real overcurrent signal of FTU, ω is the weight of the fault section summary, and the fitness function calculation formula is:

[0198]

[0199] Taking the IEEE 33-node distribution network as an example, the fault sections are named L1, L2, ..., L 32 ,L 33 , the nodes are named S1,S2,...,S 32 ,S 33 , the distributed power access points are K1, K2, and K3, and the fault settings are shown in Table 1.

[0200] Table 1 Single point failure settings

[0201]

[0202] The particle swarm optimization algorithm is used to solve the final fault section. The maximum number of iterations is set to 200 and the number of particles is set to 2000. The results are as follows: Figure 6 shown.

[0203] 5) Input the FTU alarm information and graph data into the double-layer model to obtain the fault location result A, input the distribution transformer undervoltage alarm information and graph data into the double-layer model to obtain the fault location result B, fuse the location results A and B, and use the user complaint information as an auxiliary criterion to output the fault location result. The main steps are as follows:

[0204] 5.1) Input the FTU alarm information into the double-layer model to obtain the fault location result A, and input the FTU alarm information into the double-layer model to obtain the fault location result B;

[0205] 5.2) Based on DS evidence theory, the positioning results A and B are integrated;

[0206] 5.3) Use user complaint information as auxiliary criteria, integrate the analysis results, and output the final fault location results.

Claims

1. A new distribution network fault accurate location method based on a graph neural network double-layer model in a new power system, characterized by: The following steps are involved: 1) Obtain distribution network diagram data, FTU alarm information, and distribution transformer undervoltage alarm information. 2) Constructing a two-layer distribution network fault location model, including a first-layer distribution network fault location model and a second-layer distribution network fault location model; 3) Execute step 3-1) and step 3-2) in no particular order; 3-1) Input the distribution network diagram data and FTU alarm information into the first-layer distribution network fault location model to obtain the fault bus location information A; Then, the fault bus position information A and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result A; 3-2) Input the distribution network diagram data and the distribution transformer undervoltage alarm information into the first-layer distribution network fault location model to obtain the fault bus location information B; Then, the fault bus position information B and the opening and closing status of the distribution network equipment switch are input into the second-layer distribution network fault location model to obtain the fault section location result B; 4) The fault section location result A and the fault section location result B are integrated to obtain the location result C, and then the user complaint information is used as an auxiliary criterion to output the final distribution network fault location result.

2. According to a new type of power system according to claim 1, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The distribution network map data includes topological structure information and node attribute information in the distribution network topological map.

3. According to the novel power system according to claim 2, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: In the distribution network topology diagram, nodes and edges represent busbars and feeder branches respectively.

4. According to a new type of power system according to claim 2, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The topological structure information is represented by the adjacency matrix A of the nodes V and the edges E, namely: In the formula, v i 、v j Represents a node; A ij are the elements of the adjacency matrix A.

5. According to a new type of power system according to claim 2, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The node attribute information includes the amplitude and phase angle of the three-phase voltage of the line, the amplitude and phase angle of the three-phase current, and the node center characteristic information, which is recorded as The node center characteristic information includes degree centrality DC, eigenvector centrality EC, betweenness centrality BC and closeness centrality CC; V1, V2, V3 are the amplitudes of the three-phase voltages; is the phase angle of the three-phase voltage; I1, I2, I3 are the amplitudes of the three-phase current; is the phase angle of the three-phase current.

6. According to a new type of power system according to claim 1, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The steps of obtaining the fault bus position information A include: a1) According to the fault section location result A, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point; a2) Construct the real overcurrent signal expression I j ,Right now: a3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain: In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, P and Q are the feeder section numbers in the upstream and downstream of the switch, i1 and i2 are the current values; I * j (x) is a logic function representing the on / off status of the distribution network equipment; are the status of the distributed power sources upstream and downstream of section i, respectively, where connected is 1 and not connected is 0; a4) Selecting the fitness function, transforming the fault location problem into a problem of solving the minimum value of the objective function (4), and using the particle swarm optimization algorithm to solve it, obtaining the fault bus location information A; The fitness function is as follows: In the formula, x i represents the state of fault segment i, 0 represents non-fault and 1 represents fault; ω is the weight of the fault segment summary; F(x) represents the fitness function; M represents the number of fault segments.

7. According to a new type of power system according to claim 1, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The steps of obtaining the fault bus position information B include: b1) According to the fault section location result B, the area to which the fault line of the distribution network belongs is divided, and the feeder is divided into several areas with the sectionalizer as the area demarcation point; b2) Construct the real overcurrent signal expression I j ,Right now: b3) Use the logic function to characterize the opening and closing state of the distribution network equipment, and obtain: In the formula, are the fault state values ​​from switch j to the upstream power source S1 and the downstream power source S2, respectively, j (p), x j (q) are the fault status values ​​of the sections in the upstream and downstream of switch j, P and Q are the feeder section numbers in the upstream and downstream of the switch, i1 and i2 are the current values; I * j (x) is a logic function representing the on / off status of the distribution network equipment; They are the status of the upstream and downstream distributed power sources in section i, where connected is 1 and not connected is 0; b4) Selecting the fitness function, transforming the fault location problem into a problem of solving the minimum value of the objective function (4), and using the particle swarm optimization algorithm to solve it, obtaining the fault bus location information B; The fitness function is as follows: In the formula, x i represents the state of faulty segment i, 0 represents non-faulty, 1 represents faulty; ω is the weight of the faulty segment summary. F(x) represents the fitness function; M represents the number of faulty segments.

8. According to a new type of power system according to claim 1, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The fault section location result A and the fault section location result B are fused through the DS evidence theory.

9. According to a new type of power system according to claim 1, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The user complaint information includes complaint information of downstream users in the fault section, complaint information of downstream users in the non-fault section, and fault information of the transformer to which the user belongs.

10. According to a new type of power system according to claim 1, the method for accurately locating distribution network faults based on a graph neural network double-layer model is characterized in that: The steps of outputting the final distribution network fault location result using the user complaint information as an auxiliary criterion include: parsing the user complaint information to determine whether the fault location of the user complaint is within the interval corresponding to the location result; if so, using the current location result as the final distribution network fault location result; if not, returning to step 3) to recalculate the fault section location result A and the fault section location result B.

Images