A multi-objective optimization method for distribution network relay protection setting

Abstract

The present invention discloses a multi-objective-based distribution network relay protection setting optimization method, which relates to the technical field of distribution network relay protection; by real-time monitoring of the operating state of the distribution network, the operating data and topology of the distribution network are obtained; the protection type and protection level of each protection switch are obtained through the distribution network topology, and the corresponding deep learning model is called according to the protection type and protection level, and the setting value of each protection switch is calculated according to the real-time operating data of the distribution network, and the setting value of each protection switch is updated synchronously. By calling the intelligent model to calculate the setting value, the intelligence and automation of the setting calculation are realized, the error caused by artificial factors is reduced, and the efficiency and accuracy of the setting calculation are improved; the setting value is calculated through the real-time distribution network topology, so that the calculation of the setting value can adapt to the changes in the distribution network structure, and can be performed according to the characteristics of different protection switches, which improves the adaptability and flexibility of the setting calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of power system distribution network relay protection, and in particular to a distribution network relay protection setting optimization method based on multiple objectives. Background Art

[0002] Distribution network relay protection is an important measure to detect faults or abnormal conditions in the power system, thereby issuing alarm signals, or directly isolating and removing the faulty part. Distribution network relay protection switches are one of the most important secondary equipment in the power system and play a vital role in the safe and stable operation of the power system. In order to meet the requirements of the distribution network for the speed, selectivity, sensitivity and reliability of distribution network relay protection, reasonable protection settings must be set to give full play to the effectiveness of distribution network relay protection switches.

[0003] In the existing technology, if the selectivity or sensitivity cannot meet the requirements, it is necessary to manually adjust the relevant protection settings, which often relies on the experience and judgment of engineers. There may be errors caused by human factors, resulting in low accuracy of the setting calculation. In addition, the traditional distribution network relay protection setting method is often based on fixed rules and formulas. These rules and formulas are designed based on past experience and fixed distribution network structure. It is difficult to adapt to changes in distribution network structure and operation mode, and lacks adaptability. When the distribution network structure changes, the traditional setting method cannot make timely adjustments, thus affecting the safe and stable operation of the distribution network. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of low setting calculation accuracy and insufficient adaptability mentioned in the above background technology, and to propose a distribution network relay protection setting optimization method based on multiple objectives.

[0005] The first aspect of the present invention provides a multi-objective distribution network relay protection setting optimization method, the method comprising:

[0006] Monitor the operating status of the target distribution network in real time, and obtain the operating data and topology of the target distribution network; the target distribution network consists of a power supply trunk and power supply branches, each of which is equipped with a corresponding protection switch; the function of the protection switch is to disconnect the faulty power supply line from the target distribution network when a fault occurs in the power supply trunk and power supply branch it protects;

[0007] According to the target distribution network topology diagram, the protection type and protection level of each protection switch are determined; the protection type of the protection switch determines the calculation method of the setting value of the protection switch, and the protection level of the protection switch determines the order of the setting calculation of the protection switch;

[0008] For each protection switch, according to the protection type and protection level of the protection switch, a corresponding feature matrix is ​​extracted from the operation data, and a corresponding deep learning model is called from the model library, and the feature matrix is ​​used as the input of the deep learning model to obtain the setting value of the protection switch;

[0009] At the same time, the setting value of each protection switch is updated.

[0010] Optionally, the protection level of the protection switch includes level one and level two; and according to the target distribution network topology, determining the protection type and protection level of each protection switch includes:

[0011] For each protection switch, determine the protection type of the protection switch according to the marking information of the protection switch in the target distribution network topology map; the marking information includes information of the power supply trunk and power supply branch protected by the protection switch;

[0012] Determine the current protection range of the protection switch according to the position of the protection switch in the target distribution network topology map; when the protection switch is actuated, the equipment in the current protection range of the protection switch is disconnected from the target distribution network;

[0013] If there is no other protection switch in the current protection range of the protection switch, the protection switch is determined to be a primary protection switch;

[0014] If the current protection range of the protection switch contains at least one other protection switch, the protection switch is determined to be a secondary protection switch; the setting calculation of the secondary protection switch is performed after the setting calculation of the primary protection switch.

[0015] Optionally, extracting a corresponding feature matrix from the operation data according to the protection type and protection level of the protection switch includes:

[0016] Extracting key features from the operation data according to the protection type of the protection switch; the key features include at least one of a voltage change vector, a current change vector, a power change vector and a switching state change amount of the power supply trunk and the power supply branch protected by the protection switch;

[0017] If the protection switch is a primary protection switch, the key features are used as row vectors to obtain a feature matrix;

[0018] If the protection switch is a secondary protection switch, the key features are used as row vectors to obtain a temporary matrix, and the setting values ​​of other protection switches within the protection range of the protection switch are concatenated with the temporary matrix to obtain a feature matrix.

[0019] Optionally, the step of concatenating the setting values ​​of other protection switches within the protection range of the protection switch with the temporary matrix to obtain a characteristic matrix includes:

[0020] Extract the setting values ​​of other protection switches within the protection range of the protection switch;

[0021] If the number of other protection switches within the protection range is equal to 1, the setting values ​​of other protection switches within the protection range are taken as column vectors and concatenated with the temporary matrix to obtain the characteristic matrix;

[0022] If the number of other protection switches within the protection range is greater than 1, the setting values ​​of other protection switches within the protection range are compared, and the maximum or minimum value of each type of setting value is taken to form a column vector group, which is concatenated with the temporary matrix to obtain the characteristic matrix.

[0023] Optionally, the protection types of the protection switch include substation switch protection, section switch protection, branch switch protection and boundary switch protection, and the training process of the deep learning model includes:

[0024] Determine the input type and loss function of each model based on preset parameters; the input type includes feature matrices of different sizes; the loss function determines the learning direction of the model;

[0025] For each model, according to the preset usage scenario and the input type of the model, the corresponding operation data is extracted from the historical operation data of the distribution network as the training set of the model; the usage scenario includes application to substation switches, application to section switches, application to branch switches and application to boundary switches;

[0026] Initialize the training parameters of the model and iteratively update the parameters of the model through the back-propagation algorithm to minimize the loss.

[0027] Optionally, the preset parameters include first-category parameters and second-category parameters; and determining the input type and loss function of each model according to the preset parameters includes:

[0028] The input type of the model is determined according to the first type of parameters; the first type of parameters are 1 or 2; the first type of parameters being equal to 1 indicates that there is no feature information of other protection switches in the input feature matrix of the model; the first type of parameters being equal to 2 indicates that the input feature matrix of the model contains feature information of at least one other protection switch;

[0029] The loss function of the model is determined based on the second type of parameters:

[0030] Loss total =w1*losS refuse +w2*loss false +w3*loss time

[0031] Among them, Loss total is the desired loss; loss refuse is the sensitivity loss; loss false is the selective loss; loss time is the speed loss; w1 is the weight coefficient of sensitivity loss; w2 is the weight coefficient of selectivity loss; w3 is the weight coefficient of speed loss; said w1, w2 and w3 are the second type parameters, and w1+w2+w3=1, the different proportional relationships among w1, w2 and w3 determine the different learning directions of the model.

[0032] Optionally, updating the setting value of each protection switch includes:

[0033] The setting value and update time of each protection switch are sent as update information to the corresponding protection switch, so that each protection switch completes the update of the setting value at the update time, ensuring that the setting value of each protection switch in the target distribution network is updated at the same time.

[0034] Optionally, the real-time monitoring of the operating status of the target distribution network and obtaining the operating data of the target distribution network and the target distribution network topology map includes:

[0035] Monitor the operating status of the target distribution network in real time to obtain the operating data of the target distribution network; the operating data includes voltage data, current data, power data of each power node of the target distribution network and length data of the line between each power node;

[0036] A target distribution network topology map is generated according to the positional relationship of each node of the target distribution network, and a power supply label and an outgoing line length label are added to each topological node.

[0037] Optionally, extracting a corresponding feature matrix from the operation data according to the protection type and protection level of the protection switch, and calling a corresponding deep learning model from a model library includes:

[0038] Determine the parameter template of the protection switch according to the protection type and protection level of the protection switch;

[0039] According to the parameter template, a corresponding feature matrix is ​​extracted from the operating data, and a corresponding deep learning model is called from a model library.

[0040] Beneficial effects of the present invention:

[0041] The present invention proposes a distribution network relay protection setting optimization method based on multiple objectives, the method comprising: real-time monitoring of the operating state of the target distribution network, obtaining the operating data of the target distribution network and the target distribution network topology; the target distribution network is composed of a variety of power supply trunks and power supply branches, each power supply trunk and power supply branch is equipped with a corresponding protection switch; the function of the protection switch is to disconnect the faulty equipment from the target distribution network when the power supply trunk and power supply branch it protects fail; according to the target distribution network topology, determine the protection type and protection level of each protection switch; the protection type of the protection switch determines the calculation method of the setting value of the protection switch, and the protection level of the protection switch determines the order of the setting calculation of the protection switch; for each protection switch, according to the protection type and protection level of the protection switch, extract the corresponding feature matrix from the operating data, call the corresponding deep learning model from the model library, use the feature matrix as the input of the deep learning model, and obtain the setting value of the protection switch; update the setting value of each protection switch.

[0042] By calling the intelligent model to calculate the setting value, the setting calculation is made intelligent and automated, the errors caused by human factors are reduced, and the efficiency and accuracy of the setting calculation are improved; the protection type and protection level of each protection switch are determined through the real-time distribution network topology diagram, so that the calculation of the setting value can adapt to the changes in the distribution network structure and can be performed according to the characteristics of different protection switches, thereby improving the adaptability and flexibility of the setting calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention will be further described below in conjunction with the accompanying drawings.

[0044] Figure 1 A flow chart of a multi-objective-based distribution network relay protection setting optimization method is provided for an embodiment of the present invention;

[0045] Figure 2 A flowchart of deep learning model training is provided for an embodiment of the present invention;

[0046] Figure 3 A target distribution network protection switch distribution diagram is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0048] The embodiment of the present invention provides a multi-objective-based distribution network relay protection setting optimization method. Figure 1 , Figure 1 A flow chart of a multi-objective distribution network relay protection setting optimization method provided by an embodiment of the present invention. The method comprises the following steps:

[0049] S101, monitor the operating status of the target distribution network in real time, and obtain the operating data of the target distribution network and the topology map of the target distribution network.

[0050] S102: Determine the protection type and protection level of each protection switch according to the target distribution network topology.

[0051] S103, for each protection switch, extract a corresponding feature matrix from the operation data according to the protection type and protection level of the protection switch.

[0052] S104, calling the corresponding deep learning model from the model library, using the feature matrix as the input of the deep learning model, and obtaining the setting value of the protection switch.

[0053] S105, updating the setting value of each protection switch.

[0054] Among them, the target distribution network consists of a power supply trunk and power supply branches, and each power supply trunk and power supply branch is equipped with a corresponding protection switch; the function of the protection switch is to operate when a fault occurs in the power supply trunk and power supply branch it protects, and disconnect the faulty power supply line from the target distribution network.

[0055] A multi-objective distribution network relay protection setting optimization method provided by an embodiment of the present invention realizes the intelligence and automation of the setting calculation by calling an intelligent model to calculate the setting value, reduces the error caused by human factors, and improves the efficiency and accuracy of the setting calculation; the protection type and protection level of each protection switch are determined through a real-time distribution network topology diagram, so that the calculation of the setting value can adapt to the changes in the distribution network structure and can be performed according to the characteristics of different protection switches, thereby improving the adaptability and flexibility of the setting calculation.

[0056] In one implementation, the relay protection device on the protection switch includes a measuring part, a logic part and an execution part; the measuring part is used to measure the operating data such as voltage, current and power of the power supply trunk and the power supply branch; the logic part is used to compare the measured data with the set value, generate an action signal and send the action signal to the execution part; the execution part takes action according to the action signal, such as tripping, to disconnect the power supply trunk and the power supply branch from the target distribution network.

[0057] In one implementation, the protection type of the protection switch is determined by the power supply trunk and power supply branch protected by the protection switch. The protection type of the protection switch determines the calculation method of the setting value of the protection switch. According to the characteristics of different protection switches, a suitable deep learning model is selected to calculate the setting value, so that the calculation of the setting value is more accurate.

[0058] In one implementation, the protection level of the protection switch determines the order of the protection switch setting calculation. According to the requirements of the distribution network relay protection selectivity, the protection switch is graded to calculate the time setting value, thereby ensuring that the protection switch operates in a reasonable order.

[0059] In one implementation, the monitored operating data is stored in categories to facilitate targeted extraction when calculating set values.

[0060] In one embodiment, step S102 includes:

[0061] Step 1: for each protection switch, determine the protection type of the protection switch according to the marking information of the protection switch in the target distribution network topology diagram.

[0062] Step 2: Determine the current protection range of the protection switch according to the position of the protection switch in the target distribution network topology map.

[0063] Step three: if there is no other protection switch in the current protection range of the protection switch, then the protection switch is determined to be a primary protection switch.

[0064] Step 4: If the current protection range of the protection switch contains at least one other protection switch, the protection switch is determined to be a secondary protection switch.

[0065] The marking information includes information about the power supply trunk and power supply branch protected by the protection switch.

[0066] In one implementation, the protection range of a protection switch means that when the protection switch is actuated, that is, when the protection switch disconnects an area from the power distribution network, all devices in the disconnected area are powered off, and this area is called the protection range of the protection switch. For example, when protection switch A is within the protection range of protection switch B, if A does not actuate but B actuates, the power supply line protected by A will still be powered off.

[0067] In one implementation, the protection type and location information of the protection switch are obtained from the target distribution network topology map to determine the protection range of the protection switch, thereby ensuring that when a fault occurs, various types and levels of protection switches can reliably operate or not operate, thereby achieving effective isolation of the fault and minimizing the spread of the fault.

[0068] In one implementation, the setting calculation of the secondary protection switch is performed after the setting calculation of the primary protection switch. In a setting value update cycle, the setting value of the primary protection switch is calculated in the first round; the setting value of the secondary protection switch is calculated in the second round, and the setting value of the primary protection switch calculated in the first round is required. The protection switch of the directly protected equipment is set first, and then the upper protection switch is set. This order ensures the logic and rationality of the setting calculation and improves the accuracy and reliability of the setting value.

[0069] In one embodiment, step S103 includes:

[0070] Step 1: For each protection switch, according to the protection type put into operation by the protection switch, key features are extracted from the operation data.

[0071] Step 2: If the protection switch is a primary protection switch, the key features are used as row vectors and concatenated to obtain a feature matrix.

[0072] Step three: if the protection switch is a secondary protection switch, the key features are used as row vectors to obtain a temporary matrix, and the setting values ​​of other protection switches within the protection range of the protection switch are concatenated with the temporary matrix to obtain a feature matrix.

[0073] Among them, the key features include at least one of the voltage change vector, current change vector, power change vector and opening and closing state change amount of the power supply trunk and power supply branch protected by the protection switch.

[0074] In one implementation, key features are extracted from the operating data based on the protection type put into use by the protection switch, which means that only the data most relevant to the type and function of the protection switch is selected, which helps to reduce unnecessary data processing and improve computing efficiency.

[0075] In one implementation, if the protection switch is a primary protection switch, the key features are used as row vectors and concatenated to obtain a feature matrix. For example, if the key features of a device are a voltage change vector and a current change vector, the feature matrix is ​​in the following form:

[0076]

[0077] Among them, V1 to V n Represents the change of the voltage of the protected line, I1 to I n Represents the change of the current in the protected line.

[0078] In one implementation, by considering the association and dependency with other protection switches, it is possible to adapt to different distribution network structures, which helps to improve the accuracy and reliability of setting calculations.

[0079] In one embodiment, if the protection switch is a secondary protection switch, the feature matrix extraction process includes:

[0080] Step 1: extract the setting values ​​of other protection switches within the protection range of the protection switch.

[0081] Step 2: If the number of other protection switches within the protection range is equal to 1, the setting values ​​of other protection switches within the protection range are taken as column vectors and concatenated with the temporary matrix to obtain a characteristic matrix.

[0082] Step 3: If the number of other protection switches within the protection range is greater than 1, the setting values ​​of other protection switches within the protection range are compared, and the maximum or minimum value of each type of setting value is taken to form a column vector group, which is concatenated with the temporary matrix to obtain a feature matrix.

[0083] In one implementation, if there are missing values ​​in the matrix after concatenation, zero padding is performed. For example, if the temporary matrix is ​​in the form of 2 rows and 10 columns, and the set value is in the form of 3 rows and 2 columns, then after concatenation, 0 is added to the missing value position of the last row of the concatenated matrix to obtain a complete matrix. For example, if the temporary matrix is ​​in the form of:

[0084]

[0085] Among them, V1 to V n Represents the voltage change of the protected line, I1 to I n Represents the current change of the protected line.

[0086] The set value column vector group is as follows:

[0087]

[0088] Among them, V high is the maximum value of the overvoltage protection setting of other protection switches, V low It is the minimum value of the low voltage protection setting of other protection switches; I high is the maximum value of the overcurrent protection setting of other protection switches, I low It is the minimum value of the low current protection setting of other protection switches; T m It is the maximum value of the action time setting of other protection switches.

[0089] The concatenated feature matrix is ​​as follows:

[0090]

[0091] In one embodiment, the protection types of the protection switch include substation switch protection, section switch protection, branch switch protection and boundary switch protection. Figure 2 , the training process of the deep learning model includes:

[0092] S201, determine the input type and loss function of each model according to preset parameters.

[0093] S202: For each model, according to a preset usage scenario and the input type of the model, corresponding operation data is extracted from the historical operation data of the distribution network as a training set for the model.

[0094] S203, initializing the training parameters of the model, and iteratively updating the parameters of the model through a back propagation algorithm to minimize the loss.

[0095] The input types include feature matrices of different sizes; the loss function determines the learning direction of the model. The application scenarios include substation switches, section switches, branch switches, and boundary switches.

[0096] In one implementation method, the input type and loss function of each model are determined according to preset parameters, and model training can be performed according to the characteristics of different types of power supply trunks, power supply branches, and protection switches to obtain a variety of different deep learning models that are more in line with actual application needs.

[0097] In one embodiment, the preset parameters include first-category parameters and second-category parameters, and step S201 includes:

[0098] Step 1: Determine the input type of the model based on the first type of parameters.

[0099] Step 2: Determine the loss function of the model based on the second type of parameters:

[0100] Loss total =w1*loss refuse +w2*loss false +w3*loss time

[0101] Among them, Loss total is the desired loss; loss refuse is the sensitivity loss; loss false is the selective loss; loss time is the speed loss; w1 is the weight coefficient of sensitivity loss; w2 is the weight coefficient of selectivity loss; w3 is the weight coefficient of speed loss.

[0102] In one implementation, the first type parameter takes a value of 1 or 2. If the first type parameter is equal to 1, it means that there is no feature information of other protection switches in the input feature matrix of the model; if the first type parameter is equal to 2, it means that the input feature matrix of the model contains feature information of at least one other protection switch.

[0103] In one implementation, w1, w2 and w3 are second-type parameters, and w1+w2+w3 = 1. Different proportional relationships among w1, w2 and w3 determine different learning directions of the model.

[0104] In one implementation, the first type of parameters determines whether the model will be applied to the setting calculation of the primary protection switch or the secondary protection switch. If the first type of parameters is equal to 1, it means that the model will be used in the setting calculation of the primary protection switch after training, and the input matrix of the model does not contain information about other protection switches; if the first type of parameters is equal to 2, it means that the model will be used in the setting calculation of the secondary protection switch after training, and the input matrix of the model contains setting value information of other protection switches. The input type of the model is determined by the first type of parameters, that is, the size of the input matrix of the model is determined, thereby determining the size of the feature matrix when extracting features from the operating data.

[0105] In one implementation, the sensitivity loss loss refuse and selective loss false All are specified before model training, such as setting loss refuse =100, loss false =100; and the loss of quickness time It is calculated according to the time setting value given by the model during the training process. For example, if the time setting value given by the model is t seconds, the loss can be set time =t*10.

[0106] In one implementation, the sum of w1 and w2 can be set greater than w3, that is, the weights of sensitivity loss and selectivity are larger, so that the model learns in the direction of improving sensitivity and selectivity. It can be applied to substation switches and section switches to ensure that the protection switch can act quickly when the substation equipment or section line fails.

[0107] In one implementation, the sum of w1 and w3 can be set greater than w2, that is, the weight of sensitivity loss and the weight of speed loss are larger, so that the model learns in the direction of improving sensitivity and speed. It can be applied to branch switches and boundary switches to ensure that when a branch line or boundary line fails, the protection switch can act quickly.

[0108] In one implementation, different parameters are set to train different models, so that in actual applications, a suitable model can be selected for tuning calculations according to real-time conditions.

[0109] In one embodiment, step S105 includes:

[0110] Step 1: Send the setting value and update time of each protection switch as update information to the corresponding protection switch, so that each protection switch completes the update of the setting value at the update time, ensuring that the setting value of each protection switch in the target distribution network is updated at the same time.

[0111] In one implementation, the synchronous update of each protection switch is achieved by setting the update time, avoiding the risk of false operation that may be caused by the time difference in setting value update, helping to maintain the coordination and consistency of distribution network protection, and improving the stability and reliability of the protection switch.

[0112] In one embodiment, see Figure 3 , Figure 3 This is a target distribution network protection switch distribution diagram provided by an embodiment of the present invention. The embodiment of the present invention is only illustrated with one substation, one power supply trunk (divided into two sections), and two power supply branches, but the actual situation is not limited to this.

[0113] In one implementation, the boundary switch 1 protects electrical equipment, such as all electrical equipment in a residential community. Because it will not cut off the power of other protection switches when it is operated, it is a first-level protection switch; the branch switch 1 protects a section of the distribution branch line where it is located, but when it is operated, it will cut off the power of the boundary switch 1, so it is a second-level protection switch; when the section switch 1 is operated, it will cut off the power of the section switch 2, and it is a second-level protection switch; when the substation switch is operated, it will cut off the power of all protection switches on the power supply line, and it is a second-level protection switch; first calculate the setting value of the boundary switch 1, then calculate the setting values ​​of the branch switch 1 and the branch switch 2, then calculate the setting value of the section switch 2, then calculate the setting value of the section switch 1, and finally calculate the setting value of the substation switch.

[0114] In one embodiment, step S101 includes:

[0115] Step 1: monitor the operating status of the target distribution network in real time and obtain the operating data of the target distribution network.

[0116] Step 2: Generate a target distribution network topology map based on the positional relationship of each node in the target distribution network, and add power supply labels and outgoing line length labels to each topological node.

[0117] The operation data includes voltage data, current data, power data of each power node in the target power distribution network and length data of the lines between each power node.

[0118] In one implementation, considering the difference in power supply to the power node between the main power supply and the transfer power supply, adding power supply annotation can make the subsequent setting value calculation more in line with actual needs and more accurate.

[0119] In one implementation, the model to be called may be changed according to a change in the power type.

[0120] In one implementation, the line impedance can be obtained based on the outgoing line length, providing a data basis for the subsequent calculation of constant values ​​such as current and power.

[0121] In one embodiment, according to the protection type and protection level of the protection switch, extracting the corresponding feature matrix from the operation data, and calling the corresponding deep learning model from the model library includes:

[0122] Step 1: Determine a parameter template of the protection switch according to the protection type and protection level of the protection switch.

[0123] Step 2: According to the parameter template, extract the corresponding feature matrix from the running data, and call the corresponding deep learning model from the model library.

[0124] In one implementation, the parameter template provides a standard for extracting features from the operating data, which ensures that no matter which protection switch is processed, information can be extracted in a predetermined format and structure, thereby improving the data processing efficiency.

[0125] In one implementation, the parameter template is not only used for data extraction, but also matched with the deep learning model in the model library. This ensures that the extracted feature matrix matches the called model, thereby improving the accuracy and reliability of the model prediction.

[0126] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A multi-objective distribution network relay protection setting optimization method, characterized in that: The method comprises: Monitor the operating status of the target distribution network in real time, and obtain the operating data and topology of the target distribution network; the target distribution network consists of a power supply trunk and power supply branches, each of which is equipped with a corresponding protection switch; the function of the protection switch is to disconnect the faulty power supply line from the target distribution network when a fault occurs in the power supply trunk or power supply branch it protects; According to the target distribution network topology diagram, the protection type and protection level of each protection switch are determined; the protection type of the protection switch determines the calculation method of the setting value of the protection switch, and the protection level of the protection switch determines the order of the setting calculation of the protection switch; For each protection switch, according to the protection type and protection level of the protection switch, a corresponding feature matrix is ​​extracted from the operation data, and a corresponding deep learning model is called from the model library, and the feature matrix is ​​used as the input of the deep learning model to obtain the setting value of the protection switch; Update the setting value of each protection switch; The protection levels of the protection switch include level one and level two; according to the target distribution network topology, determining the protection type and protection level of each protection switch includes: For each protection switch, determine the protection type of the protection switch according to the marking information of the protection switch in the target distribution network topology map; the marking information includes information of the power supply trunk and / or power supply branch protected by the protection switch; Determine the current protection range of the protection switch according to the position of the protection switch in the target distribution network topology map; when the protection switch is actuated, the power supply lines in the current protection range of the protection switch are disconnected from the target distribution network; If there is no other protection switch in the current protection range of the protection switch, the protection switch is determined to be a primary protection switch; If the current protection range of the protection switch contains at least one other protection switch, the protection switch is determined to be a secondary protection switch; the setting calculation of the secondary protection switch is after the setting calculation of the primary protection switch; The extracting a corresponding feature matrix from the operation data according to the protection type and protection level of the protection switch comprises: Extracting key features from the operation data according to the protection type of the protection switch; the key features include at least one of a voltage change vector, a current change vector, a power change vector and a switching state change amount of the power supply trunk or power supply branch protected by the protection switch; If the protection switch is a primary protection switch, the key features are used as row vectors to obtain a feature matrix; If the protection switch is a secondary protection switch, the key features are used as row vectors to obtain a temporary matrix, and the setting values ​​of other protection switches within the protection range of the protection switch are concatenated with the temporary matrix to obtain a feature matrix.

2. A multi-objective distribution network relay protection setting optimization method according to claim 1, characterized in that: The characteristic matrix obtained by splicing the setting values ​​of other protection switches within the protection range of the protection switch with the temporary matrix includes: Extract the setting values ​​of other protection switches within the protection range of the protection switch; If the number of other protection switches within the protection range is equal to 1, the setting values ​​of other protection switches within the protection range are taken as column vectors and concatenated with the temporary matrix to obtain the characteristic matrix; If the number of other protection switches within the protection range is greater than 1, the setting values ​​of other protection switches within the protection range are compared, and the maximum or minimum value of each type of setting value is taken to form a column vector group, which is concatenated with the temporary matrix to obtain the characteristic matrix.

3. A multi-objective distribution network relay protection setting optimization method according to claim 2, characterized in that: The protection types of the protection switch include substation switch protection, section switch protection, branch switch protection and boundary switch protection. The training process of the deep learning model includes: Determine the input type and loss function of each model based on preset parameters; the input type includes feature matrices of different sizes; the loss function determines the learning direction of the model; For each model, according to the preset usage scenario and the input type of the model, the corresponding operation data is extracted from the historical operation data of the distribution network as the training set of the model; the usage scenario includes application to substation switches, application to section switches, application to branch switches and application to boundary switches; Initialize the training parameters of the model and iteratively update the parameters of the model through the back-propagation algorithm to minimize the loss.

4. A multi-objective distribution network relay protection setting optimization method according to claim 3, characterized in that: The preset parameters include first-category parameters and second-category parameters; and determining the input type and loss function of each model according to the preset parameters includes: The input type of the model is determined according to the first type of parameters; the first type of parameters are 1 or 2; the first type of parameters being equal to 1 indicates that there is no feature information of other protection switches in the input feature matrix of the model; the first type of parameters being equal to 2 indicates that the input feature matrix of the model contains feature information of at least one other protection switch; The loss function of the model is determined based on the second type of parameters: Among them, Loss total is the desired loss; loss refuse is the sensitivity loss; loss false is the selective loss; loss time is the speed loss; w1 is the weight coefficient of sensitivity loss; w2 is the weight coefficient of selectivity loss; w3 is the weight coefficient of speed loss; said w1, w2 and w3 are the second type parameters, and w1+w2+w3=1, the different proportional relationships among w1, w2 and w3 determine the different learning directions of the model.

5. A multi-objective distribution network relay protection setting optimization method according to claim 1, characterized in that: The updating of the setting value of each protection switch includes: The setting value and update time of each protection switch are sent as update information to the corresponding protection switch, so that each protection switch completes the update of the setting value at the update time, ensuring that the setting value of each protection switch in the target distribution network is updated at the same time.

6. A multi-objective distribution network relay protection setting optimization method according to claim 1, characterized in that: The real-time monitoring of the operating status of the target distribution network and obtaining the operating data of the target distribution network and the target distribution network topology diagram include: Monitor the operating status of the target distribution network in real time to obtain the operating data of the target distribution network; the operating data includes voltage data, current data, power data of each power node of the target distribution network and length data of the line between each power node; A target distribution network topology map is generated according to the positional relationship of each node of the target distribution network, and a power supply label and an outgoing line length label are added to each topological node.

7. A multi-objective distribution network relay protection setting optimization method according to claim 1, characterized in that: The step of extracting a corresponding feature matrix from the operation data according to the protection type and protection level of the protection switch, and calling a corresponding deep learning model from a model library includes: Determine the parameter template of the protection switch according to the protection type and protection level of the protection switch; According to the parameter template, a corresponding feature matrix is ​​extracted from the operating data, and a corresponding deep learning model is called from a model library.

Images