Relay protection inspection system architecture and method based on cloud edge collaboration

Abstract

The invention relates to a relay protection inspection system architecture and method based on cloud edge collaboration. The objective of the invention is to solve the technical problems of low automation degree and insufficient test coverage rate of the existing relay protection test. According to the technical scheme, the system comprises a cloud end and a side end, the cloud end is deployed on a wave recording master station, the side end comprises a relay protection device, edge wave recording equipment and an automatic testing device, and the resource advantages of wave recording networking of secondary equipment are utilized to be responsible for collection, preprocessing and real-time testing of local data; the relay protection inspection method comprises the steps of data acquisition, state pre-evaluation of the relay protection device, automatic testing and formation of a test report, a positive interaction mode of information exchange is formed between the master station and the secondary equipment through cloud-side cooperation, optimal use of equipment resources of the whole network is facilitated, and the service life of the equipment resources of the whole network is prolonged. And the overall analysis and fault early warning functions of the master station are enhanced, and real-time pre-evaluation, automatic testing and intelligent operation and maintenance of the equipment state are realized in combination with cloud computing, edge computing, an improved BP neural network and an evidence theory.

Description

Technical Field

[0001] The present invention belongs to the technical field of relay protection for power systems, and specifically relates to a relay protection inspection system architecture and method based on cloud-edge collaboration. Background Art

[0002] The operation and maintenance of traditional relay protection equipment has the following problems:

[0003] 1. Manual inspections are inefficient and delayed, making it difficult to meet the rapidly growing scale of smart substation equipment.

[0004] 2. The existing automated test system has insufficient coverage, low fault location accuracy, and poor operability of diagnostic results;

[0005] 3. Test data is scattered, lacking unified management and analysis of data throughout its entire life cycle, and operation and maintenance decisions rely on experience.

[0006] Although existing technologies have introduced methods such as BP neural networks and edge computing, they still have defects such as slow model convergence, unintelligent test resource scheduling, and insufficient multi-source information fusion. Therefore, an efficient and intelligent relay protection test system architecture is needed. Summary of the Invention

[0007] The purpose of the present invention is to solve the above technical problems and provide a relay protection inspection system architecture and method based on cloud-edge collaboration.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] A relay protection inspection system architecture based on cloud-edge collaboration, including cloud and edge;

[0010] The cloud is deployed on the recording master station and uses data from different sources and dimensions to conduct in-depth feature correlation analysis of different defects in the data, and conducts comprehensive analysis and inference of abnormal situations to more accurately perceive the real-time operation status of the equipment;

[0011] The edge includes relay protection devices, edge recording equipment and automatic testing equipment, which takes advantage of the resource advantages of secondary equipment recording networking and is responsible for local data collection, preprocessing and real-time testing;

[0012] The cloud is electrically connected to the edge recording device, the edge recording device is electrically connected to the automatic testing device and the background monitoring system, the automatic testing device is electrically connected to the relay protection device, the relay protection device is electrically connected to the background monitoring system, and the cloud is electrically connected to the mobile terminal of the operation and maintenance personnel.

[0013] Furthermore, a BP neural network system is deployed in a cloud computing environment. The system continuously collects and feeds back errors generated by the system during the simulation training phase, uses these errors to adjust the weights of neurons, and thereby creates an artificial neural network system that simulates the initial problem. The BP neural network system includes an input layer, a hidden layer, and an output layer.

[0014] The input layer is used to receive device alarm information;

[0015] The hidden layer uses a dynamic number of neurons , and introduce the momentum factor α∈[0.1, 0.8], use the momentum factor to improve the convergence rate in the gradient slow region, and achieve a stable suppression effect in the convergence fast region;

[0016] Where N represents the number of faults that occur within an operation and maintenance cycle, and I represents the number of alarm types;

[0017] The output layer scores the health status of the device.

[0018] Furthermore, the scoring standard of the output layer is 1-10 points, and the larger the score, the better the equipment status, which is divided into three levels: 1-4 points are abnormal, 4-7 points are medium, and 7-10 points are good.

[0019] A relay protection inspection method based on cloud-edge collaboration is characterized by comprising the following steps:

[0020] Step 1) Data Collection: The relay protection device and edge recording equipment utilize the resource advantages of the secondary equipment recording network to perform normalized, multi-dimensional data collection and upload it to the cloud in real time;

[0021] Step 2) Pre-assessment of the relay protection device's status: The cloud performs an in-depth feature correlation analysis of different defects in the collected multi-dimensional data, and comprehensively analyzes and infers any abnormal conditions that occur, thereby pre-assessing the relay protection device's status.

[0022] Step 3) Automated testing: The recording master station obtains and sends information about abnormal protection devices based on the pre-assessment results. Based on this information, the edge terminal develops a standard test template and transmits it to the automated testing device for automatic editing of the test process.

[0023] Step 4) Generate a test report: The recording master station receives the protection device action information uploaded by the edge terminal, generates a test report and sends it to the mobile terminal of the operation and maintenance personnel.

[0024] Furthermore, the specific steps of the state pre-assessment of the relay protection device in step 2) are as follows:

[0025] Step 2.1) Collect and organize historical data from relay protection devices: First, the cloud compiles historical operating conditions and fault records for each type of relay protection device. Then, the data is categorized by relay protection device type and the required data types are extracted. Finally, each data set is divided into multiple subcategories based on the relay protection device ID and sorted by time series, while excluding some invalid data.

[0026] Step 2.2) Data processing: The data obtained in step 2.1) is divided into two levels: alarm and fault. Faults that cause protection function failure are assigned a value from 1 to 10 based on the time since the last alarm information was issued. During the operation and maintenance cycle of each device, the longer the time since the last alarm information was issued, the better the fault status and the larger the corresponding assigned value. The cause matrix of historical fault information is constructed. The cause matrix is ​​expressed as:

[0027] (1)

[0028] in: Indicates the number of times the i-th alarm type occurs when the n-th fault occurs;

[0029] The cause matrix is ​​normalized as shown in formula (2), where the correction function As shown in formula (3):

[0030] (2)

[0031] (3)

[0032] Step 2.3) Training the BP neural network: Input the alarm data information obtained in step 2.2) into the input layer of the BP neural network system. The result vector presented by the sample is:

[0033] (4)

[0034] Import the cause matrix and result vector into the BP neural network, and use these result vectors as the expected value of the model output;

[0035] A single hidden layer neural network method is used: the number of neurons in the hidden layer is M, and M is determined by the data determined by formula (5);

[0036] (5)

[0037] Activation Function The S-shaped function is used, and its specific expression is shown in formula (6);

[0038] (6)

[0039] The value of belongs to the range [0,1], A and B are constants;

[0040] By inputting the cause matrix of relay protection, the data of the hidden layer can be obtained, as shown in formula (7);

[0041] (7)

[0042] (8)

[0043] in: is an intermediate variable, To pass parameters, is the offset, which transmits the data of the hidden layer to the output layer, as shown in formula (9);

[0044] (9)

[0045] (10)

[0046] in, is an intermediate variable, To pass parameters, is the offset; the error calculation of the neural network is as shown in formula (11);

[0047] (11)

[0048] Training error requirements , when the error does not meet the requirement, calculate the correction amount of the network parameters, and the correction amount is calculated as shown in formula (12);

[0049] (12)

[0050] Where, and is the learning factor, and is the gradient value, and the gradient is calculated as in formula (13);

[0051] (13)

[0052] In the formula, the function is the derivative of function f, and the weight parameter is corrected according to the correction amount. The weight parameter correction is shown in formula (14):

[0053]

[0054]

[0055]

[0056] (14)

[0057] Where, , , , are initialized to random numbers between 0 and 1, and the momentum factor α∈[0.1,0.8] is a constant. and The correction value is given for the n-1th group of data. The momentum factor is used to increase the convergence rate in the gradient slowing area and achieve a stable suppression effect in the gradient over-convergence area.

[0058] Step 2.4) Determine the structure and parameters of the neural network, train it, and perform validation: Use the cause matrix and result vector obtained in step 2.3) as input and output values ​​for training, and determine that the allowable error range is 4%-8%;

[0059] Step 2.5) Use the trained model to evaluate the running device.

[0060] Furthermore, the specific steps of the automated test in step 3) are as follows:

[0061] Step 3.1) The cloud obtains the test task information of the relay protection device and sends the basic information of the relay protection device to be tested to the edge recording device;

[0062] Step 3.2) The edge recorder collects the received information and develops a standard test template based on it. The template is then sent to the automatic test device according to the test items.

[0063] Step 3.3) The automatic test device automatically edits the test process based on the acquired test item test template, and then sends the output simulated current and voltage to the relay protection device item by item;

[0064] Step 3.4) The background monitoring system of the smart substation uploads the real-time action information of the relay protection device to the edge recording device;

[0065] Step 3.5) The edge recording device performs preliminary processing on the data sent by the background monitoring system and uploads it to the cloud in real time.

[0066] Furthermore, the basic information in step 3.1) includes fixed values, test items, test methods and historical test data, and the preliminary processing in step 3.5) includes filtering, denoising and data compression.

[0067] Furthermore, the specific steps for forming the test report in step 4) are as follows:

[0068] Step 4.1) After the relay protection device is tested, the cloud obtains its topology information, test requirements, and basic information;

[0069] Step 4.2) The cloud breaks down the acquired information into test content and automatically generates a test list;

[0070] Step 4.3) Use DS evidence theory to integrate basic information about the relay protection device and multi-source information obtained from testing to provide a more accurate and comprehensive evaluation result;

[0071] Step 4.4) Automatically generate a test report, save it in the cloud, and send it to the operation and maintenance personnel's mobile terminal.

[0072] Furthermore, the multi-source information in step 4.3) includes analog quantity information, switch quantity information, protection action quantity information, related device data and a whole set of test data, and the test report in step 4.4) includes fault location, recommended measures and historical data comparison analysis.

[0073] Compared with the prior art, the present invention has the following beneficial effects:

[0074] 1. Through cloud-edge collaboration, the present invention forms a benign interactive mode of information exchange between the master station and secondary devices, which helps to optimize the use of equipment resources across the entire network and enhances the overall analysis and fault warning functions of the master station;

[0075] 2. The present invention uses an improved BP neural network to pre-evaluate the status of relay protection equipment and automatically detect relay protection equipment with abnormal pre-evaluation results. Finally, the test report generated by the cloud platform is stored in the cloud platform server and sent to relevant operation and maintenance personnel, facilitating comparative analysis of historical test data throughout the life cycle of the smart station and rapid operation and maintenance of the equipment;

[0076] 3. While optimizing the overall testing process, the present invention can greatly reduce the waste of material and human resources, effectively improve on-site debugging efficiency and test reliability, and has great significance for the promotion and application of future intelligent maintenance;

[0077] 4. The present invention combines cloud computing, edge computing, improved BP neural network and evidence theory to realize real-time pre-assessment of equipment status, automated testing and intelligent operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 Schematic diagram of the system architecture of the present invention;

[0079] Figure 2 Schematic diagram of the BP neural network of the present invention;

[0080] Figure 3 This is a schematic diagram of the relay protection device status pre-assessment of the present invention;

[0081] Figure 4 This is a schematic diagram of the automatic testing architecture of the present invention;

[0082] Figure 5 A schematic diagram of a process for automatically generating a test report of the present invention;

[0083] Figure 6 A training process graph of the neural network of the present invention;

[0084] Figure 7 This is a diagram showing the evaluation of the relay protection device of the present invention;

[0085] Figure 8 This is a test report diagram generated by the cloud platform of the present invention. DETAILED DESCRIPTION

[0086] The present invention will be further described below with reference to the embodiments and the accompanying drawings.

[0087] like Figure 1 As shown in the figure, a relay protection inspection system architecture based on cloud-edge collaboration includes the cloud and edge.

[0088] The cloud is deployed on the recording master station, using data from different sources and dimensions to conduct in-depth feature correlation analysis of different defects in the data, and comprehensively analyze and infer the abnormal situations that occur, so as to more accurately perceive the real-time operation status of the equipment and promote the formation of a collaborative operation mode between the master station and edge-side devices.

[0089] A BP neural network system is deployed in a cloud computing environment. The system continuously collects and feeds back the errors generated by the system during the simulation training phase, uses these errors to adjust the weights of neurons, and then creates an artificial neural network system that simulates the initial problem. The BP neural network system includes an input layer, a hidden layer, and an output layer, such as Figure 2 As shown;

[0090] The input layer is used to receive device alarm information;

[0091] The hidden layer uses a dynamic number of neurons , and introduce the momentum factor α∈[0.1, 0.8], use the momentum factor to improve the convergence rate in the gradient slow region, and achieve a stable suppression effect in the convergence fast region;

[0092] Where N represents the number of faults that occur within an operation and maintenance cycle, and I represents the number of alarm types;

[0093] The output layer scores the health status of the device on a scale of 1-10. The higher the score, the better the device status. The health status of the device is divided into three levels: 1-4 points are abnormal, 4-7 points are medium, and 7-10 points are good.

[0094] The edge includes relay protection devices, edge recording equipment and automatic testing equipment, which takes advantage of the resource advantages of secondary equipment recording networking and is responsible for local data collection, preprocessing and real-time testing;

[0095] The cloud is electrically connected to the edge recording device, the edge recording device is electrically connected to the automatic testing device and the background monitoring system, the automatic testing device is electrically connected to the relay protection device, the relay protection device is electrically connected to the background monitoring system, and the cloud is electrically connected to the mobile terminal of the operation and maintenance personnel.

[0096] A relay protection inspection method based on cloud-edge collaboration includes the following steps:

[0097] Step 1) Data Collection: The relay protection device and edge recording equipment utilize the resource advantages of the secondary equipment recording network to perform normalized, multi-dimensional data collection and upload it to the cloud in real time;

[0098] Step 2) Pre-evaluation of the status of the relay protection device: The cloud performs an in-depth feature correlation analysis of different defects in the collected multi-dimensional data, and conducts a comprehensive analysis and inference of the abnormal conditions that occur, and pre-evaluates the status of the relay protection device, such as Figure 3 The specific steps are as follows:

[0099] Step 2.1) Collect and organize historical data from relay protection devices: First, the cloud compiles historical operating conditions and fault records for each type of relay protection device. Then, the data is categorized by relay protection device type and the required data types are extracted. Finally, each data set is divided into multiple subcategories based on the relay protection device ID and sorted by time series, while excluding some invalid data.

[0100] Step 2.2) Data processing: The data obtained in step 2.1) is divided into two levels: alarm and fault. Faults that cause protection function failure are assigned a value from 1 to 10 based on the time since the last alarm information was issued. During the operation and maintenance cycle of each device, the longer the time since the last alarm information was issued, the better the fault status and the larger the corresponding assigned value. The cause matrix of historical fault information is constructed. The cause matrix is ​​expressed as:

[0101] (1)

[0102] in: Indicates the number of times the i-th alarm type occurs when the n-th fault occurs;

[0103] The cause matrix is ​​normalized as shown in formula (2), where the correction function As shown in formula (3):

[0104] (2)

[0105] (3)

[0106] Table 1 Some historical alarm information of relay protection equipment

[0107] X1 sampling alarm X2 Fibre Channel alarms X3 board abnormal alarm X4 optical power exceeds the limit X5 long start … Xn chip abnormality 1 1 1 0 1 … 0 0 1 0 0 0 … 0 … … … … … … 0 0 3 0 0 0 … 1

[0108] Step 2.3) Training the BP neural network: Input the alarm data information obtained in step 2.2) into the input layer of the BP neural network system. The result vector presented by the sample is:

[0109] (4)

[0110] Import the cause matrix and result vector into the BP neural network, and use these result vectors as the expected value of the model output;

[0111] A single hidden layer neural network method is used: the number of neurons in the hidden layer is M, and M is determined by the data determined by formula (5);

[0112] (5)

[0113] Activation Function The S-shaped function is used, and its specific expression is shown in formula (6);

[0114] (6)

[0115] The value of belongs to the range [0,1], A and B are constants;

[0116] By inputting the cause matrix of relay protection, the data of the hidden layer can be obtained, as shown in formula (7);

[0117] (7)

[0118] (8)

[0119] in: is an intermediate variable, To pass parameters, is the offset, which transmits the data of the hidden layer to the output layer, as shown in formula (9);

[0120] (9)

[0121] (10)

[0122] in, is an intermediate variable, To pass parameters, is the offset; the error calculation of the neural network is as shown in formula (11);

[0123] (11)

[0124] Training error requirements , when the error does not meet the requirement, calculate the correction amount of the network parameters, and the correction amount is calculated as shown in formula (12);

[0125]

[0126]

[0127]

[0128] (12)

[0129] Where, and is the learning factor, and is the gradient value, and the gradient is calculated as in formula (13);

[0130]

[0131] (13)

[0132] In the formula, the function is the derivative of function f, and the weight parameter is corrected according to the correction amount. The weight parameter correction is shown in formula (14):

[0133]

[0134] (14)

[0135] Where, , , , are initialized to random numbers between 0 and 1, and the momentum factor α∈[0.1,0.8] is a constant. and The correction value is given for the n-1th group of data. The momentum factor is used to increase the convergence rate in the gradient slowing area and achieve a stable suppression effect in the gradient over-convergence area.

[0136] Step 2.4) Determine the structure and parameters of the neural network, train it, and perform validation: Use the cause matrix and result vector obtained in step 2.3) as input and output values ​​for training, and determine that the allowable error range is 4%-8%;

[0137] Step 2.5) Use the trained model to evaluate the running device.

[0138] Step 3) Automated testing: The recording master station obtains and sends information about protection devices with abnormal operating status based on the pre-assessment results. Based on this data, the edge terminal develops a standard test template and passes it to the automated testing device for automatic editing of the test process, such as Figure 4 The specific steps are as follows:

[0139] Step 3.1) The cloud obtains the test task information of the relay protection device and sends the basic information of the relay protection device to be tested (settings, test items, test methods, and historical test data) to the edge recording device;

[0140] Step 3.2) The edge recorder collects the received information and develops a standard test template based on it. The template is then sent to the automatic test device according to the test items.

[0141] Step 3.3) The automatic test device automatically edits the test process based on the acquired test item test template, and then sends the output simulated current and voltage to the relay protection device item by item;

[0142] Step 3.4) The background monitoring system of the smart substation uploads the real-time action information of the relay protection device to the edge recording device;

[0143] Step 3.5) The edge recording device performs preliminary processing (filtering, denoising, and data compression) on the data sent by the background monitoring system and uploads it to the cloud in real time.

[0144] Step 4) Generate a test report: The recording master station receives the protection device action information uploaded by the edge terminal, generates a test report and sends it to the mobile terminal of the operation and maintenance personnel, such as Figure 5 The specific steps are as follows:

[0145] Step 4.1) After the relay protection device is tested, the cloud obtains its topology information, test requirements, and basic information;

[0146] Step 4.2) The cloud breaks down the acquired information into test content and automatically generates a test list;

[0147] Step 4.3) Use DS evidence theory to integrate the basic information of the relay protection device and the multi-source information obtained from the test (analog information, switching information, protection operation information, related device data, and the entire set of test data) to provide a more accurate and comprehensive evaluation result;

[0148] Step 4.4) Automatically generate a test report (fault location, recommended measures, and historical data comparison and analysis), save it in the cloud, and send it to the operation and maintenance personnel's mobile terminal.

[0149] The method of the present invention has been applied to line protection equipment at 110 kV or higher voltage levels in the Shanxi power grid. Currently, the number of A3 line protection devices operating in substations has reached 563. The types of abnormal or alarm data generated by the device cover 20 different types of data, such as protection lockout, output abnormality, and CPU plug-in abnormality. The alarm and fault data generated by the A3 protection devices in recent years were sorted, and each row of data was assigned a value. Finally, the data table in Table 2 was generated, totaling 1500 rows. X1 to X20 were selected as the alarm categories. Each row represents the type and frequency of alarm information experienced by the A3 equipment before a fault occurred during the operation and maintenance cycle.

[0150] Table 2 Historical data of A3 type protection device

[0151] X1 memory alarm X2 opens abnormally X3 power supply alarm X4 management tasks … X18 protection lock X19CPU plug-in exception X20FLASH self-test error Y assignment 1 0 0 1 … 0 1 0 2 0 0 0 0 … 0 1 0 5 0 1 0 0 … 0 0 1 8 1 0 0 0 … 0 0 0 9 0 0 0 0 … 0 2 0 9 0 0 0 2 … 2 0 0 8 3 2 0 0 … 0 0 0 7 2 0 0 1 … 0 0 0 3 … … … … … … … … … 2 0 0 0 … 0 0 0 6 0 0 0 0 … 0 1 0 9 1 0 0 0 … 0 0 0 8

[0152] Each row of data is assigned a value based on the time between the device failure and the last alarm information. The longer the interval, the larger the assigned value (1 to 10). 1100 sets of data are used to train the BP neural network, 200 sets of data are used to verify the BP neural network, and the remaining 200 sets of data are used to test the model. Finally, a pre-assessment model of the protection device is generated. The training process of the BP neural network is as follows: Figure 6 As shown in the figure, it can be seen that the BP neural network reaches the optimal state when it is trained to the 45th generation, with a training error of 0.3674 and a training correlation coefficient R of 0.926.

[0153] This model was used to evaluate 100 A3 type devices currently in operation in substations. The evaluation results are as follows: Figure 7 As shown in the figure, the horizontal axis is the status evaluation result, which is divided into three levels: abnormal (1 to 4 points), medium (4 to 7 points), and good (7 to 10 points). The vertical axis is the number of corresponding devices. Among the 100 A3 type protection devices, 94 devices were evaluated as good, 2 devices were evaluated as medium, and 4 devices were evaluated as abnormal.

[0154] Automatically test the protection devices with abnormal evaluation results, and the test report generated by the cloud platform is as follows: Figure 8 As shown, relevant operation and maintenance personnel can use this as a reference to carry out reasonable and efficient operation and maintenance of related equipment.

Claims

1. A relay protection inspection system architecture based on cloud-edge collaboration, characterized by: Including cloud and edge; The cloud is deployed on the recording master station and uses data from different sources and dimensions to conduct in-depth feature correlation analysis of different defects in the data, and conducts comprehensive analysis and inference of abnormal situations to more accurately perceive the real-time operation status of the equipment; The edge includes relay protection devices, edge recording equipment and automatic testing equipment, which takes advantage of the resource advantages of secondary equipment recording networking and is responsible for local data collection, preprocessing and real-time testing; The cloud is electrically connected to the edge recording device, the edge recording device is electrically connected to the automatic testing device and the background monitoring system, the automatic testing device is electrically connected to the relay protection device, the relay protection device is electrically connected to the background monitoring system, and the cloud is electrically connected to the mobile terminal of the operation and maintenance personnel.

2. A cloud-edge collaboration-based relay protection inspection system architecture according to claim 1, characterized in that: Deploy a BP neural network system in a cloud computing environment. The system continuously collects and feeds back errors generated by the system during a simulation training phase, uses these errors to adjust the weights of neurons, and thereby creates an artificial neural network system that simulates the initial problem. The BP neural network system includes an input layer, a hidden layer, and an output layer. The input layer is used to receive device alarm information; The hidden layer uses a dynamic number of neurons , and introduce the momentum factor α∈[0.1, 0.8], use the momentum factor to improve the convergence rate in the gradient slow region, and achieve a stable suppression effect in the convergence fast region; Where N represents the number of faults that occur within an operation and maintenance cycle, and I represents the number of alarm types; The output layer scores the health status of the device.

3. The cloud-edge collaboration-based relay protection inspection system architecture according to claim 2 is characterized in that: The scoring standard of the output layer is 1-10 points. The larger the score, the better the equipment status. It is divided into three levels: 1-4 points are abnormal, 4-7 points are medium, and 7-10 points are good.

4. The relay protection inspection method using the cloud-edge collaboration-based relay protection inspection system architecture of claim 3 is characterized in that: The steps include: Step 1) Data Collection: The relay protection device and edge recording equipment utilize the resource advantages of the secondary equipment recording network to perform normalized, multi-dimensional data collection and upload it to the cloud in real time; Step 2) Pre-assessment of the relay protection device's status: The cloud performs an in-depth feature correlation analysis of different defects in the collected multi-dimensional data, and comprehensively analyzes and infers any abnormal conditions that occur, thereby pre-assessing the relay protection device's status. Step 3) Automated testing: The recording master station obtains and sends information about abnormal protection devices based on the pre-assessment results. Based on this information, the edge terminal develops a standard test template and transmits it to the automated testing device for automatic editing of the test process. Step 4) Generate a test report: The recording master station receives the protection device action information uploaded by the edge terminal, generates a test report and sends it to the mobile terminal of the operation and maintenance personnel.

5. The relay protection inspection method based on cloud-edge collaboration according to claim 4 is characterized in that: The specific steps of step 2) of pre-assessing the status of the relay protection device are as follows: Step 2.1) Collect and organize historical data from relay protection devices: First, the cloud compiles historical operating conditions and fault records for each type of relay protection device. Then, the data is categorized by relay protection device type and the required data types are extracted. Finally, each data set is divided into multiple subcategories based on the relay protection device ID and sorted by time series, while excluding some invalid data. Step 2.2) Data processing: The data obtained in step 2.1) is divided into two levels: alarm and fault. Faults that cause protection function failure are assigned a value from 1 to 10 based on the time since the last alarm information was issued. During the operation and maintenance cycle of each device, the longer the time since the last alarm information was issued, the better the fault status and the larger the corresponding assigned value. The cause matrix of historical fault information is constructed. The cause matrix is ​​expressed as: （1） in: Indicates the number of times the i-th alarm type occurs when the n-th fault occurs; The cause matrix is ​​normalized as shown in formula (2), where the correction function As shown in formula (3): （2） （3） Step 2.3) Training the BP neural network: Input the alarm data information obtained in step 2.2) into the input layer of the BP neural network system. The result vector presented by the sample is: （4） Import the cause matrix and result vector into the BP neural network, and use these result vectors as the expected value of the model output; A single hidden layer neural network method is used: the number of neurons in the hidden layer is M, and M is determined by the data determined by formula (5); （5） Activation Function The S-shaped function is used, and its specific expression is shown in formula (6); （6） The value of belongs to the range [0,1], A and B are constants; By inputting the cause matrix of relay protection, the data of the hidden layer can be obtained, as shown in formula (7); （7） （8） in: is an intermediate variable, To pass parameters, is the offset, which transmits the data of the hidden layer to the output layer, as shown in formula (9); （9） （10） in, is an intermediate variable, To pass parameters, is the offset; the error calculation of the neural network is as shown in formula (11); （11） Training error requirements , when the error does not meet the requirement, calculate the correction amount of the network parameters, and the correction amount is calculated as shown in formula (12); （12） Where, and is the learning factor, and is the gradient value, and the gradient is calculated as in formula (13); （13） In the formula, the function is the derivative of function f, and the weight parameter is corrected according to the correction amount. The weight parameter correction is shown in formula (14): （14） Where, , , , are initialized to random numbers between 0 and 1, and the momentum factor α∈[0.1,0.8] is a constant. and The correction value is given for the n-1th group of data. The momentum factor is used to increase the convergence rate in the gradient slowing area and achieve a stable suppression effect in the convergence fast area. Step 2.4) Determine the structure and parameters of the neural network, train it, and perform validation: Use the cause matrix and result vector obtained in step 2.3) as input and output values ​​for training, and determine that the allowable error range is 4%-8%; Step 2.5) Use the trained model to evaluate the running device.

6. The relay protection inspection method based on cloud-edge collaboration according to claim 4 is characterized in that: The specific steps of the automated test in step 3) are as follows: Step 3.1) The cloud obtains the test task information of the relay protection device and sends the basic information of the relay protection device to be tested to the edge recording device; Step 3.2) The edge recorder collects the received information and develops a standard test template based on it. The template is then sent to the automatic test device according to the test items. Step 3.3) The automatic test device automatically edits the test process based on the acquired test item test template, and then sends the output simulated current and voltage to the relay protection device item by item; Step 3.4) The background monitoring system of the smart substation uploads the real-time action information of the relay protection device to the edge recording device; Step 3.5) The edge recording device performs preliminary processing on the data sent by the background monitoring system and uploads it to the cloud in real time.

7. The relay protection inspection method based on cloud-edge collaboration according to claim 6 is characterized in that: The basic information in step 3.1) includes fixed values, test items, test methods and historical test data, and the preliminary processing in step 3.5) includes filtering, denoising and data compression.

8. The relay protection inspection method based on cloud-edge collaboration according to claim 4 is characterized in that: The specific steps for forming the test report in step 4) are as follows: Step 4.1) After the relay protection device is tested, the cloud obtains its topology information, test requirements, and basic information; Step 4.2) The cloud breaks down the acquired information into test content and automatically generates a test list; Step 4.3) Use DS evidence theory to integrate basic information about the relay protection device and multi-source information obtained from testing to provide a more accurate and comprehensive evaluation result; Step 4.4) Automatically generate a test report, save it in the cloud, and send it to the operation and maintenance personnel's mobile terminal.

9. The relay protection inspection method based on cloud-edge collaboration according to claim 8 is characterized in that: The multi-source information in step 4.3) includes analog quantity information, switch quantity information, protection action quantity information, related device data and a whole set of test data. The test report in step 4.4) includes fault location, recommended measures and historical data comparison and analysis.

Images