Voltage Sag Source Location Method Based on the Fusion of Simulation and Measured Data

Abstract

The present invention relates to a method for locating voltage sag sources based on the fusion of simulation and measured data, comprising the following steps: Step S1: Acquiring simulated and measured data of voltage sags; Step S2: Calculating information on the sag's impact, grid structure, and fault type; Step S3: Pre-training a multi-classification model based on the simulation data; Step S4: Training the multi-classification model using transfer learning and measured data; and Step S5: Localizing the voltage sag source based on the trained multi-classification model. This method overcomes the poor applicability of existing methods and can effectively improve the accuracy of locating voltage sag sources.

Description

Technical Field

[0001] The present invention relates to the field of voltage sag positioning, and in particular to a voltage sag source positioning method based on the fusion of simulation and measured data. Background Art

[0002] In today's power systems, voltage sags are becoming an increasingly prominent issue in power quality. The Institute of Electrical and Electronics Engineers (IEEE) defines a voltage sag as a momentary decrease in the effective value of the system supply voltage to 10% to 90% of the rated value, lasting from 10ms to 1 minute. Modern loads are even more sensitive to voltage sags, which can cause significant economic losses to high-tech companies and many industrial users. In this context, research on locating the source of voltage sags is necessary to understand grid fault conditions, identify potential line hazards, and help users develop temporary power supply plans to mitigate the impact of voltage sags.

[0003] Existing research on voltage sag source location methods is all based on simulation analysis methods or measured data-driven methods. Although they can solve the problem of positioning accuracy, they still have certain limitations: the simulation analysis method ignores the influence of certain factors in actual scenarios (such as weather) and cannot be directly applied to actual situations; the measured data-driven method is restricted by the number of monitoring points, resulting in a small number of measured samples and uneven data distribution, which makes the model training insufficient and affects the accuracy of voltage sag source location. Summary of the Invention

[0004] In view of this, the object of the present invention is to provide a voltage sag source locating method based on the fusion of simulation and measured data, aiming to solve the above problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for locating a voltage sag source based on the fusion of simulation and measured data includes the following steps:

[0007] Step S1: Acquire simulation data and measured data of voltage sag;

[0008] Step S2: Calculate the sag impact information, grid structure information, and fault type information;

[0009] Step S3: pre-training a multi-classification model based on simulation data;

[0010] Step S4: Combine transfer learning and measured data to train a multi-classification model;

[0011] Step S5: Locate the voltage sag source based on the trained multi-classification model.

[0012] Furthermore, the calculation of the information on the degree of impact of the temporary sag is specifically as follows:

[0013] Assume that the calculation formula for the amplitude of the disturbed voltage sag at the monitoring point is as follows:

[0014]

[0015] In the above formula: o represents the matrix U i With D i Multiply the corresponding elements; U i represents the voltage sag amplitude of each monitoring node in the ith sag event; D i Indicates the disturbance condition of each monitoring node, where disturbed is 1 and undisturbed is 0;

[0016] Since the sag event list only contains the information of the disturbed monitoring nodes, and the simulation results contain the information of all nodes in the entire network, that is, using the number of disturbed monitoring nodes Ni of the ith sag event, the measured system is set to N iM ; The simulation system is N iS ; The simulation calculation results describe the advantage of including complete node information with the average voltage sag amplitude E of the disturbed node. The calculation formula is shown in formulas (2)-(4):

[0017]

[0018]

[0019] In formula (2) and formula (3): M represents the set of monitoring points; B represents the set of all nodes in the simulation system; b represents the disturbed node, num() is the number of calculations;

[0020] Under the i-th sag event, the average voltage sag amplitude E i The calculation formula is as follows:

[0021]

[0022] In the above formula: sum() represents the sum of matrix elements.

[0023] Furthermore, the grid structure information is specifically:

[0024] Introducing the scale of monitoring points F i , tightness C i And the distance L i Describe the grid structure information and define the monitoring point scale F by comprehensively considering the connection between the node and the adjacent nodes. i :

[0025]

[0026] In the above formula: f irepresents the number of adjacent nodes of monitoring point i, and G represents the set of adjacent nodes of monitoring point i;

[0027] Define the monitoring node density C i There are several characteristics that describe the transmission of voltage sags:

[0028]

[0029] In the above formula: c i represents the number of interconnected lines between monitoring point i and its adjacent nodes; n i represents the number of adjacent nodes of monitoring point i;

[0030] Assume that the endpoints of the changing branch are j and k, and the calculation formula is as follows:

[0031] L i =min{l ij , l ik} (7)

[0032] In the above formula: l ij Indicates the shortest electrical distance between monitoring point i and the endpoint j of the change branch; l ik It represents the shortest electrical distance between monitoring point i and endpoint k of the changing branch.

[0033] Furthermore, the fault cause information is characterized by a fault type T, and different fault types are represented by numerical values: T=1, 2, 3, 4 represent single-phase grounding, inter-phase short circuit, two-phase grounding, and three-phase short circuit, respectively.

[0034] Furthermore, the multi-classification model adopts MLP, and the network structure of MLP includes an input layer, a hidden layer and an output layer, and a full connection method is used between layers.

[0035] Furthermore, the MLP is specifically as follows:

[0036] The number of neurons m in the hidden layer is determined by formula (8):

[0037]

[0038] Where: Indicates rounding up; p indicates the dimension of the input layer parameter matrix; q indicates the number of multi-classifications; a indicates an adjustable constant of [1,10] to ensure the adaptability of the model; m indicates the number of neurons in the hidden layer;

[0039] According to the information of the sag impact degree, grid structure and fault type of the calculated simulation data, the features are constructed: [X i ,N i ,E i ,F i ,Ci ,L i ,T i ], then the formula from the input layer to the hidden layer is as follows:

[0040] H i =W1*[X i ,N i ,E i ,F i ,C i ,L i ,T i ] (9)

[0041] In the above formula: W1 represents a (4*A+3)*m-dimensional matrix;

[0042] The process from the input layer to the hidden layer is as follows: the (4*A+3)-dimensional input is compressed to m dimensions through full connection, and each dimension reflects the information of the degree of sag impact, grid structure information and fault type information;

[0043] The formula from hidden layer to output layer is as follows:

[0044] Y i =W2*H i (10)

[0045] In the above formula: W2 represents an m*M dimensional matrix, where element w yz (y∈[1,m],z∈[1,M]) represents the weight coefficient of the hidden layer neuron y being divided into the zth class under the i-th fault;

[0046] The process from the hidden layer to the output layer is as follows: the m-dimensional input is fully connected to obtain the membership results of each category of the m-dimensional input, and the category number with the largest membership is selected as the temporary sag source line number.

[0047] Furthermore, the step S4 is specifically as follows:

[0048] By calculating the impact degree of the measured data, the grid structure information and the fault type information, the feature is formed: [X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ], that is, the target domain dataset

[0049] The calculation formula for transfer learning is as follows:

[0050] P S =f S ([X i ,Ni ,E i ,F i ,C i ,L i ,T i ],W1) (11)

[0051] P M =f M ([X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ],W'1) (12)

[0052] In the above formula: W1, W'1 represent the parameters in the source domain and target domain respectively; [X i ,N i ,E i ,F i ,C i ,L i ,T i ]、[X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ] represent the datasets in the source domain and target domain respectively; f S 、f M Represent the mapping relationship between the source domain and the target domain respectively; P S 、P M Represent the predicted values of the source domain and the target domain respectively;

[0053] The objective function is set as follows:

[0054]

[0055]

[0056] In the above formula: y s 、y M Represent the true labels of the source domain and the target domain respectively; N S 、N M Represents the number of training samples in the source domain and the target domain respectively; L represents the loss function;

[0057] Using the multi-classification model pre-trained in step 2, the input layer and hidden layer parameters W1 are fixed and the parameters W'1 are adjusted;

[0058] The formula from the input layer to the hidden layer is as follows:

[0059] H' i =W'1*[X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ] (15)

[0060] Retrain the output layer parameters, similar to formula (10), and the formula from the hidden layer to the output layer is as follows:

[0061] Y' i =W'2*H' i (16)

[0062] In the above formula: W'2 represents an m*N dimensional matrix, where element w' yz (y∈[1,m],z∈[1,N]) represents the weight coefficient for classifying hidden layer neuron y into the zth class under the i-th fault. The process from the hidden layer to the output layer can be represented as follows: Taking an m-dimensional input and using a fully connected approach, we can obtain the membership results for each class of the m-dimensional input. The class number with the highest membership is selected as the line number of the sag source.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. The present invention combines the advantages of a large simulation sample size and comprehensive measured data information, taking into account the characteristics of both types of data and using them as data support for locating the source of voltage sags. This not only makes up for the deficiency of small measured data samples, but also considers the actual influencing factors at the simulation level.

[0065] 2. The present invention overcomes the problem of poor applicability of existing methods and can effectively improve the accuracy of locating the source of voltage sag. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a flow chart of the method of the present invention;

[0067] Figure 2 is the network structure of MLP in one embodiment of the present invention;

[0068] Figure 3 2 is a diagram of a transfer learning structure in one embodiment of the present invention. DETAILED DESCRIPTION

[0069] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0070] Please refer to Figure 1The present invention provides a method for locating a voltage sag source based on the fusion of simulation and measured data, comprising the following steps:

[0071] Step S1: Acquire simulation data and measured data of voltage sag;

[0072] Step S2: Calculate the sag impact information, grid structure information, and fault type information;

[0073] Step S3: pre-training a multi-classification model based on simulation data;

[0074] Step S4: Combine transfer learning and measured data to train a multi-classification model;

[0075] Step S5: Locate the voltage sag source based on the trained multi-classification model.

[0076] In this embodiment, the simulation data for voltage sag is based on the Engineering Production Management System (PMS2.0). The system obtains parameters of power grid equipment components (including line parameters, transformer parameters, etc.), and performs massive simulation calculations on different lines and different fault scenarios across the entire network based on the Monte Carlo method. The voltage amplitude data of each node is obtained, and the node voltage amplitude, fault type parameters, and line operation mode are used as the source of simulation data.

[0077] Measured data is based on a sag event list. This list contains information such as the sag magnitude, type, occurrence time, and duration at the monitoring point. It offers advantages such as ease of access and high real-time availability. However, the triggering time and duration of a single voltage sag event are random, and their magnitude is dependent on external environmental factors such as protection action time, making them unsuitable for use as features. Therefore, the sag magnitude, fault type, and grid structure information contained in the sag event list are used as the source of measured data.

[0078] In this embodiment, the sag impact information, grid structure information, and fault type information are calculated as follows:

[0079] Voltage amplitude, fault type, and grid structure are common data contained in both simulation and measured data. Therefore, to locate the source of voltage sags, this method combines the advantages of large amounts of simulation data with the comprehensiveness of measured data. Based on this common data, this method constructs three-dimensional attributes: information on the sag's impact, grid structure, and fault cause. This describes the specific information of a single voltage sag event. This is shown in Table 1.

[0080] Table 1 Location characteristics of voltage sag sources

[0081]

[0082] (1) Information on the impact of temporary sag

[0083] The impact of voltage sags can, to a certain extent, characterize the physical properties of the power grid. Specifically, voltage sag transmission is affected by distance, exhibiting a decaying characteristic. This information can be used to characterize the impact range and severity of a single voltage sag event. Definition: In a real system, a node where a monitoring point records a sag magnitude is considered an affected node. In a simulation model, a node where the sag magnitude is less than 0.9 is considered an affected node.

[0084] The calculation formula for the amplitude of the disturbed voltage sag at the monitoring point is as follows:

[0085]

[0086] In the above formula: Represents the matrix U i With D i Multiply the corresponding elements; U i represents the voltage sag amplitude of each monitoring node in the ith sag event; D i Indicates the disturbance status of each monitoring node, where disturbed is 1 and undisturbed is 0. The voltage amplitude values in the simulation and measured systems are as follows: In the actual system, if a monitoring point records a voltage sag event, the monitoring point is affected by the voltage sag event, and the sag amplitude is the recorded value; conversely, if the monitoring point is unaffected, the sag amplitude is 1; in the simulation system, the voltage amplitude is the simulation calculation result.

[0087] Since the sag event list only contains the information of the disturbed monitoring nodes, and the simulation results contain the information of all nodes in the entire network, that is, the number of disturbed monitoring nodes N in the ith sag event is used. i (Measured system: N iM ;Simulation system: N iS ) and the average voltage sag amplitude E of the disturbed node describe the advantages of the simulation calculation results containing full node information. The calculation formulas are shown in Equations (2)-(4):

[0088]

[0089]

[0090] In formula (2) and formula (3), M represents the set of monitoring points; B represents the set of all nodes in the simulation system; b represents the disturbed node, and num() is the number of calculations.

[0091] Under the i-th sag event, the average voltage sag amplitude E i The calculation formula is as follows:

[0092]

[0093] In the above formula: sum() represents the sum of matrix elements.

[0094] (2) Grid structure information

[0095] Changes in the grid structure will cause changes in the connection relationship between nodes, resulting in changes in the propagation path of voltage sags. However, the sag impact information only unilaterally describes the impact of voltage sags from the physical properties of the grid, and cannot characterize the impact level of voltage sags from the perspective of topological properties, that is, the grid structure. Therefore, this method introduces the monitoring point scale F i , tightness C i And the distance L i Describe the grid structure information.

[0096] If a node is connected to more lines, it is more susceptible to the impact of voltage sags at other nodes, reflecting the degree to which the node is directly affected by other nodes; if a node is connected to fewer lines, but its adjacent nodes are connected to more lines, the node is still susceptible to the impact of voltage sags at other nodes, reflecting the degree to which the node is indirectly affected by other nodes. Therefore, the monitoring point scale F is defined by comprehensively considering the connection between the node and its adjacent nodes. i :

[0097]

[0098] In the above formula: f i represents the number of adjacent nodes of monitoring point i, and G represents the set of adjacent nodes of monitoring point i.

[0099] Since the scale of monitoring points F i It can only describe the degree to which a node is affected by other nodes, without considering the voltage sag transmission characteristics. Therefore, the monitoring node tightness C is defined as i There are several characteristics that describe the transmission of voltage sags:

[0100]

[0101] In the above formula: c i represents the number of interconnected lines between monitoring point i and its adjacent nodes; n i Indicates the number of adjacent nodes of monitoring point i. Adjacent nodes and monitoring points can theoretically generate (n i +1)n i / 2 lines, so C i Indicates the closeness of monitoring point i.

[0102] Due to the changes in working conditions and line maintenance in the actual power grid, the connection mode between nodes changes, which further affects the severity of voltage sag at the monitoring point. Therefore, this patent defines the distance level L i Describes the degree of influence of the changing branch on the monitoring point. Assuming that the endpoints of the changing branch are j and k, the calculation formula is as follows:

[0103] L i =min{l ij , l ik} (7)

[0104] In the above formula: l ij Indicates the shortest electrical distance between monitoring point i and the endpoint j of the change branch; l ik It represents the shortest electrical distance between monitoring point i and endpoint k of the changing branch.

[0105] (3) Fault cause information

[0106] Because sag impact information and grid structure information only characterize the external characteristics of voltage sags, ignoring the impact of internal factors, and considering that voltage sags caused by different fault types have different impact ranges, voltage sags caused by three-phase faults are the most severe, this patent selects fault type T to represent fault cause information.

[0107] Preferably, different fault types are represented by numerical values: T=1, 2, 3, 4 represent single-phase grounding, inter-phase short circuit, two-phase grounding, and three-phase short circuit, respectively.

[0108] In this embodiment, for the result of locating the source of the temporary drop, on the one hand, upstream and downstream positioning can only divide the responsibility sharing relationship between the user and the power grid company, and locating the specific location puts higher requirements on the storage space and transmission channel of the data, and the amount of calculation is large; on the other hand, the power grid company aims to associate the fault on the grid side through the fault recording system, check the hidden dangers of the line, and improve the efficiency of targeted operation and maintenance. Users pay attention to the relative position of the temporary drop source in order to provide a basis for subsequent economic disputes, and also pay attention to the scope of the line fault in order to formulate a plan for resuming work and production, rather than focusing on the specific location of the temporary drop source. Therefore, this method takes into account the computational cost, takes into account the temporary drop source positioning needs of the grid layer and the user layer, and determines the temporary drop positioning result as the line where the temporary drop source is located.

[0109] When the same type of fault occurs on the same line while the grid operation remains unchanged, and the features selected in step 1 have a certain degree of similarity, locating the source of the voltage sag can be considered a multi-classification problem, with the line as the classification result and the number of classifications determined based on the number of lines. MLP is selected as the classification algorithm for voltage sag location.

[0110] The network structure of MLP includes input layer, hidden layer and output layer, and the layers are fully connected. The structural diagram is as follows: Figure 2 shown.

[0111] The number of neurons in the hidden layer will affect the final classification results. If the number of neurons in the hidden layer is too small, it will be difficult to complete the learning requirements and cannot accurately complete the classification requirements; if the number of neurons in the hidden layer is too large, it is easy to cause overfitting, resulting in poor generalization ability of the network. The number of neurons in the hidden layer m is determined according to formula (8):

[0112]

[0113] Where: Indicates rounding up; p represents the dimension of the input layer parameter matrix; q represents the number of multi-classifications; a represents an adjustable constant [1,10] to ensure the adaptability of the model; m represents the number of neurons in the hidden layer.

[0114] Assuming that a network contains A monitoring terminals and the number of lines is M, a simulation model of the network is constructed using PMS2.0. The Monte Carlo method is used to obtain the voltage sag amplitude and fault type of each node. The voltage sag impact degree information, grid structure information and fault type information of the simulation data are calculated using equations (1) to (7), forming the feature: [X i ,N i ,E i ,F i ,C i ,L i ,T i ].

[0115] The formula from the input layer to the hidden layer is as follows:

[0116] H i =W1*[X i ,N i ,E i ,F i ,C i ,L i ,T i ] (9)

[0117] In the above formula: W1 represents a (4*A+3)*m-dimensional matrix, where element w dg (d∈[1,4*A+3],g∈[1,m]) represents the weight coefficient of node d's output at the gth fault. The process from the input layer to the hidden layer can be reflected as follows: the (4*A+3)-dimensional input is compressed to m dimensions through a fully connected approach, and each dimension reflects information about the impact of the temporary sag, the grid structure, and the fault type.

[0118] The formula from hidden layer to output layer is as follows:

[0119] Y i =W2*H i (10)

[0120] In the above formula: W2 represents an m*M dimensional matrix, where element w yz (y∈[1,m],z∈[1,M]) represents the weight coefficient for classifying hidden layer neuron y into the zth class under the i-th fault. The process from the hidden layer to the output layer can be represented as follows: Taking an m-dimensional input and using a fully connected approach, we can obtain the membership results for each class of the m-dimensional input. The class number with the highest membership is selected as the line number of the sag source.

[0121] In this embodiment, a transfer learning mechanism is introduced to train measured data. Transfer learning is a machine learning concept that fine-tunes existing models to adapt them to new tasks or new functions. In transfer learning, the learning domain containing a large amount of labeled data is called the source domain, and the learning domain with less labeled data is called the target domain. The model is pre-trained on the source domain data and fine-tuned using the target domain data to fully utilize the source domain data and improve the accuracy of the model in the target domain. This method uses the learning domain of simulation data as the source domain and the learning domain of measured data as the target domain, and utilizes transfer learning to construct a multi-classification model for locating voltage sag sources.

[0122] The influence degree of the temporary sag of the measured data, the grid structure information and the fault type information are calculated by formula (1) to formula (7), and the characteristics are formed: [X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ], which is the target domain dataset.

[0123] The calculation formula for transfer learning is as follows:

[0124] P S =f S ([X i ,N i ,E i ,F i ,C i ,L i ,T i ],W1) (11)

[0125] P M =f M ([X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ],W'1) (12)

[0126] In the above formula: W1, W'1 represent the parameters in the source domain and target domain respectively; [X i ,N i ,E i ,F i ,C i ,L i ,T i ]、[X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ] represent the datasets in the source domain and target domain respectively; f S 、f M Represent the mapping relationship between the source domain and the target domain respectively; P S 、P M Represent the predicted values of the source domain and the target domain respectively. In order to minimize the error between the predicted value and the true value label, the objective function is set as follows:

[0127]

[0128]

[0129] In the above formula: y s 、y M Represent the true labels of the source domain and the target domain respectively; N S 、N M They represent the number of training samples in the source domain and the target domain respectively; L represents the loss function.

[0130] Since the features extracted by neural networks with higher layers are more distinct, i.e., the information in the first few layers is more general, this method uses the pre-trained multi-classification model in step 2, fixes the input layer and hidden layer parameters W1, and only uses the fine-tuning parameter W'1.

[0131] The formula from the input layer to the hidden layer is as follows:

[0132] H' i =W'1*[X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ] (15)

[0133] Retrain the output layer parameters, similar to formula (10), and the formula from the hidden layer to the output layer is as follows:

[0134] Y' i=W'2*H' i (16)

[0135] In the above formula: W'2 represents an m*N dimensional matrix, where element w' yz (y∈[1,m],z∈[1,N]) represents the weight coefficient for classifying hidden layer neuron y into the zth class under the i-th fault. The process from the hidden layer to the output layer can be represented as follows: Taking an m-dimensional input and using a fully connected approach, we can obtain the membership results for each class of the m-dimensional input. The class number with the highest membership is selected as the line number of the sag source.

[0136] This method can be used to obtain a multi-classification model that can accurately locate the measured small sample data. Figure 3 shown.

[0137] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A method for locating voltage sag sources based on the fusion of simulation and measured data, characterized in that: The following steps are involved: Step S1: Acquire simulation data and measured data of voltage sag; Step S2: Calculate the sag impact information, grid structure information, and fault type information; Step S3: pre-training a multi-classification model based on simulation data; Step S4: Combine transfer learning and measured data to train a multi-classification model; Step S5: locating the voltage sag source based on the trained multi-classification model; The step S4 is specifically as follows: By calculating the impact degree of the measured data, the grid structure information and the fault type information, the feature is formed: [X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ], that is, the target domain dataset The calculation formula for transfer learning is as follows: P S =f S ([X i ,N i ,E i ,F i ,C i ,L i ,T i ],W1) (11) P M =f M ([X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ],W'1) (12) In the above formula: W1, W'1 represent the parameters in the source domain and target domain respectively; [X i ,N i ,E i ,F i ,C i ,L i ,T i ]、[X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ] represent the datasets in the source domain and target domain respectively; f S 、f M Represent the mapping relationship between the source domain and the target domain respectively; P S 、P M Represent the predicted values of the source domain and the target domain respectively; The objective function is set as follows: In the above formula: y s 、y M Represent the true labels of the source domain and the target domain respectively; N S 、N M Represents the number of training samples in the source domain and the target domain respectively; L represents the loss function; Using the multi-classification model pre-trained in step 2, the input layer and hidden layer parameters W1 are fixed and the parameters W'1 are adjusted; The formula from the input layer to the hidden layer is as follows: H' i =W'1*[X' i ,N' i ,E' i ,F' i ,C' i ,L' i ,T' i ] (15) Retrain the output layer parameters, similar to formula (10), and the formula from the hidden layer to the output layer is as follows: Y' i =W'2*H' i (16) In the above formula: W'2 represents an m*N dimensional matrix, where element w' yz (y∈[1,m],z∈[1,N]) represents the weight coefficient of the hidden layer neuron y being divided into the zth class under the i-th fault. The process from the hidden layer to the output layer can be reflected as follows: taking the m-dimensional input and using the full connection method, the membership results of each class of the m-dimensional input can be obtained, and the class number with the largest membership is selected as the temporary sag source line number.

2. The voltage sag source location method based on the fusion of simulation and measured data according to claim 1 is characterized in that: The calculation of the information on the degree of impact of the temporary sag is specifically as follows: Assume that the calculation formula for the amplitude of the disturbed voltage sag at the monitoring point is as follows: In the above formula: Represents the matrix U i With D i Multiply the corresponding elements; U i represents the voltage sag amplitude of each monitoring node in the ith sag event; D i Indicates the disturbance condition of each monitoring node, where disturbed is 1 and undisturbed is 0; Since the sag event list only contains the information of the disturbed monitoring nodes, and the simulation results contain the information of all nodes in the entire network, that is, the number of disturbed monitoring nodes N in the ith sag event is used. i , let the measured system be N iM ; The simulation system is N iS ; The simulation calculation results describe the advantage of including complete node information with the average voltage sag amplitude E of the disturbed node. The calculation formula is shown in formulas (2)-(4): In formula (2) and formula (3): M represents the set of monitoring points; B represents the set of all nodes in the simulation system; b represents the disturbed node, num() is the number of calculations; Under the i-th sag event, the average voltage sag amplitude E i The calculation formula is as follows: In the above formula: sum() represents the sum of matrix elements.

3. The voltage sag source location method based on the fusion of simulation and measured data according to claim 1 is characterized in that: The grid structure information is specifically: Introducing the scale of monitoring points F i , tightness C i And the distance L i Describe the grid structure information and define the monitoring point scale F by comprehensively considering the connection between the node and the adjacent nodes. i : In the above formula: f i represents the number of adjacent nodes of monitoring point i, and G represents the set of adjacent nodes of monitoring point i; Define the monitoring node density C i There are several characteristics that describe the transmission of voltage sags: In the above formula: c i represents the number of interconnected lines between monitoring point i and its adjacent nodes; n i represents the number of adjacent nodes of monitoring point i; Assume that the endpoints of the changing branch are j and k, and the calculation formula is as follows: L i =min{l ij ,L ik } (7) In the above formula: l ij Indicates the shortest electrical distance between monitoring point i and the endpoint j of the change branch; l ik It represents the shortest electrical distance between monitoring point i and endpoint k of the changing branch.

4. The voltage sag source location method based on the fusion of simulation and measured data according to claim 1 is characterized in that: The fault cause information is characterized by the fault type T, and different fault types are represented by numerical values: T=1, 2, 3, and 4 represent single-phase grounding, interphase short circuit, two-phase grounding, and three-phase short circuit, respectively.

5. The voltage sag source location method based on the fusion of simulation and measured data according to claim 1 is characterized in that: The multi-classification model adopts MLP, and the network structure of MLP includes an input layer, a hidden layer and an output layer, and a full connection method is used between layers.

6. The voltage sag source location method based on the fusion of simulation and measured data according to claim 5 is characterized in that: The MLP is specifically as follows: The number of neurons m in the hidden layer is determined by formula (8): Where: Indicates rounding up; p indicates the dimension of the input layer parameter matrix; q indicates the number of multi-classifications; a indicates an adjustable constant of [1,10] to ensure the adaptability of the model; m indicates the number of neurons in the hidden layer; According to the information of the sag impact degree, grid structure and fault type of the calculated simulation data, the features are constructed: [X i ,N i ,E i ,F i ,C i ,L i ,T i ], then the formula from the input layer to the hidden layer is as follows: H i =W1*[X i ,N i ,E i ,F i ,C i ,L i ,T i ] (9) In the above formula: W1 represents a (4*A+3)*m-dimensional matrix; The process from the input layer to the hidden layer is as follows: the (4*A+3)-dimensional input is compressed to m dimensions through full connection, and each dimension reflects the information of the degree of sag impact, grid structure information and fault type information; The formula from hidden layer to output layer is as follows: AND i =W2*H i (10) In the above formula: W2 represents an m*M dimensional matrix, where element w yz (y∈[1,m],z∈[1,M]) represents the weight coefficient of the hidden layer neuron y being divided into the zth class under the i-th fault; The process from the hidden layer to the output layer is as follows: the m-dimensional input is fully connected to obtain the membership results of each category of the m-dimensional input, and the category number with the largest membership is selected as the temporary sag source line number.

Images