Power distribution gateway testing method, system, medium and terminal based on deep learning

Abstract

The present invention discloses a power distribution gateway testing method, system, medium and terminal based on deep learning, comprising: when using an operation and maintenance debugging tool to test the tested power distribution gateway, collecting data generated by the tested power distribution gateway during the test process as test data; performing cluster analysis on all test data, and classifying and processing each test data according to the cluster analysis results to form a test sample database of multiple manufacturers; using the MAAC algorithm to evaluate the local environment observation state after the operation and maintenance debugging tool performs each test action based on the test sample database of multiple manufacturers; using the MAAC algorithm to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results, and testing the tested power distribution gateway according to the test execution strategy. The present invention uses the MAAC algorithm to deeply mine the test sample databases of multiple manufacturers to generate test execution strategies suitable for power distribution gateways of different manufacturers and models.

Description

Technical Field

[0001] The present invention relates to the field of equipment testing, and in particular to a power distribution gateway testing method, system, medium and terminal based on deep learning. Background Art

[0002] In recent years, information technology, big data technology, and artificial intelligence technology have developed rapidly, and human society has gradually entered an intelligent era marked by the interconnection of everything. As a key technology for achieving the interconnection of everything, the Internet of Things (IoT) focuses on building an interconnected system through sensing devices, communications technologies, and the internet, enabling intelligent, panoramic interaction between people, machines, and things. The use of IoT technology is conducive to the revolutionary upgrading of industries and has been widely adopted across various sectors. In the power industry, the IoT, built by leveraging cutting-edge technologies such as "big cloud, Internet of Things, mobile Internet, and intelligent chain," has enabled the transformation of traditional power grids to digital ones, promoted the deep integration of the grid's physical and cyber systems, and facilitated the interconnection of "source, grid, load, and storage" equipment.

[0003] As the construction of the power distribution Internet of Things (IoT) progresses, a vast number of distribution gateways are connected to IoT platforms. However, current on-site operation and maintenance testing still primarily relies on single-unit testing, without networking. Field testing instruments have limited functionality, and field work relies heavily on personnel experience, resulting in inconsistent work quality. Fault diagnosis relies on supplier technical support, and diagnostic testing requires extensive experience. This results in low fault diagnosis efficiency, making it difficult to improve quality and efficiency, severely hindering the development of digital and intelligent distribution networks. Summary of the Invention

[0004] Embodiments of the present invention provide a distribution gateway testing method, system, medium and terminal based on deep learning. A deep learning algorithm is used to deeply mine the test sample database of multiple manufacturers containing observation data, and then generate a test execution strategy suitable for distribution gateways of different manufacturers and models to expand the applicable scenarios of the test execution strategy.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a power distribution gateway testing method based on deep learning, including:

[0006] When using the operation and maintenance debugging tool to test the tested power distribution gateway, collecting data generated by the tested power distribution gateway during the test process as test data;

[0007] Performing cluster analysis on all the test data, and classifying each test data according to the cluster analysis results to form a test sample database of multiple manufacturers;

[0008] Using the MAAC algorithm, based on the test sample databases of the multiple manufacturers, the local environment observation state of the operation and maintenance debugging tool after executing each test action is evaluated;

[0009] The MAAC algorithm is used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool performs each test action, and the tested distribution gateway is tested in accordance with the test execution strategy of the operation and maintenance debugging tool.

[0010] In the implementation of the embodiment of the present invention, when using the operation and maintenance debugging tool to test the tested distribution gateway, the data generated by the tested distribution gateway during the test process is collected as test data, and then all the test data are clustered and analyzed. According to the cluster analysis results, each test data is classified and processed, and data belonging to different manufacturers are intelligently identified to form a test sample database of multiple manufacturers, providing sample data support for the deep learning of the operation and maintenance debugging tool, solving the long-standing problem that test data cannot be effectively mined and extracted, effectively meeting the pertinence of field test cases brought by massive samples and the accuracy of diagnostic results, and facilitating the subsequent summary of common problems of different manufacturers. Then, the MAAC algorithm, a deep learning algorithm, is used to evaluate the local environmental observation status after the operation and maintenance debugging tool executes each test action based on the test sample database of multiple manufacturers. ,Then, the MAAC algorithm is used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool executes each test action, and the distribution gateway under test is tested in accordance with the test execution strategy of the operation and maintenance debugging tool, thereby forming a targeted test plan for manufacturers and common problems. Through the MAAC algorithm, the test sample databases of multiple manufacturers can be deeply mined, and then the local environment observation status after the operation and maintenance debugging tool executes each test action can be evaluated. Based on the analysis of the evaluation results, the test execution strategy of the operation and maintenance debugging tool is obtained, so that the test execution strategy can adapt to distribution gateways of different manufacturers and models, and then adapt to different test requirements and environmental changes, providing users with a wider range of test coverage, reducing unnecessary repeated tests, and improving test efficiency.

[0011] As a preferred solution, the MAAC algorithm is used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool performs each test action, and the tested power distribution gateway is tested according to the test execution strategy of the operation and maintenance debugging tool, specifically:

[0012] Inputting the evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action into the intelligent agent action network, so that the intelligent agent action network outputs the corresponding test instructions; wherein the intelligent agent action network is obtained by updating the network parameters of the initial action network using the gradient ascent method; the initial action network is pre-built using the MAAC algorithm;

[0013] Sending the test instruction to the operation and maintenance debugging tool through the manager;

[0014] When the operation and maintenance debugging tool receives the test instruction, the operation and maintenance debugging tool obtains the test execution strategy of the operation and maintenance debugging tool based on the test instruction, and tests the tested power distribution gateway according to the test execution strategy of the operation and maintenance debugging tool.

[0015] According to a preferred embodiment of the present invention, the local environment observation status assessment results of the operation and maintenance debugging tool are input into the intelligent agent action network, so that the intelligent agent action network outputs corresponding test instructions, and the intelligently generated test instructions are sent to the operation and maintenance debugging tool through the manager, so that the operation and maintenance debugging tool executes the test tasks based on the test instructions. This automated process reduces intermediate links and improves test efficiency and accuracy. In addition, an initial action network is pre-built based on the MAAC algorithm, and then the network parameters of the initial action network are updated using the gradient ascent method to obtain the intelligent agent action network. This allows the intelligent agent action network to learn from historical test data and optimize network parameters, thereby continuously improving the performance of the intelligent agent action network, further optimizing the test execution strategy, and improving the test effect.

[0016] As a preferred solution, the MAAC algorithm is used to evaluate the local environment observation status after the operation and maintenance debugging tool performs each test action based on the test sample database of the multiple manufacturers, specifically:

[0017] Inputting data contained in the test sample databases of the multiple manufacturers into an agent evaluation network, so that the agent evaluation network outputs an estimated value of a state-action evaluation function corresponding to each test action as an evaluation result of the local environment observation state after the operation and maintenance debugging tool executes each test action;

[0018] Among them, the test sample database of multiple manufacturers contains several test actions of multiple manufacturers, as well as the local environment observation state after executing each test action; the intelligent agent evaluation network is obtained by iteratively updating the parameters of the initial evaluation network with the goal of minimizing the loss function; the initial evaluation network is pre-built using the MAAC algorithm.

[0019] According to a preferred embodiment of the present invention, an initial evaluation network is pre-built based on the MAAC algorithm, so that the rating network is adaptive, thereby being able to cope with the testing requirements and environmental changes of different manufacturers, and then handle complex test scenarios and changing test environments. Then, with the goal of minimizing the loss function, the parameters of the initial evaluation network are continuously iteratively updated to obtain an intelligent agent evaluation network. This data-driven optimization method can learn from a large amount of historical test data and continuously improve the accuracy and reliability of the evaluation.

[0020] As a preferred solution, cluster analysis is performed on all the test data, and based on the cluster analysis results, each test data is classified to form a test sample database of multiple manufacturers, specifically:

[0021] Classifying all the test data into manufacturer data according to the pre-input manufacturer names to build an initial database of the multiple manufacturers;

[0022] Using the K-means algorithm, cluster analysis is performed based on the distance between each two test data, and according to the cluster analysis results, each of the test data is classified into the corresponding initial database;

[0023] After all the test data are classified, the current initial database of each manufacturer is used as the test sample database of each manufacturer.

[0024] According to the preferred embodiment of the present invention, after preliminary manufacturer data classification is performed on all test data according to multiple pre-input manufacturer names, a K-means algorithm is used to perform cluster analysis based on the distance between each two test data. Classification can be performed based on the intrinsic characteristics of the data to ensure the accuracy and efficiency of data classification. According to the clustering analysis results of the K-means algorithm, each test data is classified into a corresponding initial database, thereby constructing a test sample database of multiple manufacturers, which provides a structured and standardized data foundation for subsequent testing and analysis.

[0025] As a preferred solution, when the tested power distribution gateway is tested using the operation and maintenance debugging tool, data generated by the tested power distribution gateway during the test process is collected as test data, specifically:

[0026] When using the operation and maintenance debugging tool to execute the current test item to test the tested power distribution gateway, it is determined in real time whether the current test item is the last test item;

[0027] If so, the test-related information is filled into a preset test report template, a corresponding test report is generated, and a corresponding test conclusion is given; wherein the test-related information includes the test data and standard data corresponding to each test item, the absolute error between the test data and the standard data, and the preset threshold value corresponding to the absolute error;

[0028] If not, the index moves to the next test item, so that the operation and maintenance debugging tool continues to execute the next test item, tests the tested power distribution gateway and collects data generated by the tested power distribution gateway during the test process.

[0029] According to the preferred embodiment of the present invention, when the operation and maintenance debugging tool is used to execute the current test item to test the distribution gateway under test, it is determined in real time whether the current test item is the last test item. If the current test item is the last test item, the test-related information is filled into the preset test report template in real time to generate the corresponding test report, thereby improving the speed and accuracy of report generation. The standardized test process and report generation method ensure the consistency and repeatability of each test, which helps to perform the same test at different time points and in different environments, and obtain results in a consistent format for comparison. In addition, the test-related information includes the test data, standard data, absolute error and preset threshold corresponding to each test item. The detailed records ensure the comprehensiveness and traceability of the test results, which helps to facilitate subsequent analysis and optimization.

[0030] As a preferred solution, when the tested power distribution gateway is tested using the operation and maintenance debugging tool, data generated by the tested power distribution gateway during the test process is collected as test data, specifically:

[0031] When using the operation and maintenance debugging tool to execute the current test project to test the tested power distribution gateway, real-time collection of data generated by the tested power distribution gateway during the test process as test data corresponding to the current test project;

[0032] Calculate the absolute error between the test data corresponding to the current test item and the standard data corresponding to the current test item, and determine whether the calculated absolute error is greater than a preset threshold;

[0033] If yes, the test conclusion of the current test item is determined to be unqualified;

[0034] If not, the test conclusion of the current test item is determined to be qualified.

[0035] A preferred solution for implementing the embodiment of the present invention is to calculate the absolute error between the test data corresponding to the current test item and the standard data corresponding to the current test item, and determine whether the calculated absolute error is greater than a preset threshold. If so, the test conclusion of the current test item is determined to be unqualified; otherwise, the test conclusion of the current test item is determined to be qualified. Through the above-mentioned real-time data collection and error judgment means, the test results can be fed back instantly, helping operators to identify unqualified test items in time, which helps to quickly locate and solve problems and reduce troubleshooting time.

[0036] In order to solve the same technical problem, an embodiment of the present invention further provides a power distribution gateway testing system based on deep learning, comprising:

[0037] A data acquisition module is used to collect data generated by the tested power distribution gateway during the test process as test data when the tested power distribution gateway is tested using the operation and maintenance debugging tool;

[0038] A cluster analysis module, configured to perform cluster analysis on all the test data, and classify each test data according to the cluster analysis results to form a test sample database of multiple manufacturers;

[0039] A state assessment module, configured to use the MAAC algorithm to assess the local environment observation state after the operation and maintenance debugging tool executes each test action based on the test sample database of the multiple manufacturers;

[0040] The strategy analysis module is used to adopt the MAAC algorithm to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool performs each test action, and test the distribution gateway under test in accordance with the test execution strategy of the operation and maintenance debugging tool.

[0041] As a preferred solution, the strategy analysis module specifically includes:

[0042] An instruction generation unit is configured to input evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action into the agent action network, so that the agent action network outputs corresponding test instructions; wherein the agent action network is obtained by updating the network parameters of the initial action network using the gradient ascent method; the initial action network is pre-built using the MAAC algorithm;

[0043] An instruction sending unit, configured to send the test instruction to the operation and maintenance debugging tool through a manager;

[0044] An instruction response unit is used to obtain the test execution strategy of the operation and maintenance debugging tool based on the test instruction when the operation and maintenance debugging tool receives the test instruction, and test the tested power distribution gateway in accordance with the test execution strategy of the operation and maintenance debugging tool.

[0045] In order to solve the same technical problem, the present invention also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the distribution gateway testing method based on deep learning.

[0046] In order to solve the same technical problem, the present invention also provides a terminal, including a processor, a memory and a computer program stored in the memory; wherein, the computer program can be executed by the processor to implement the power distribution gateway testing method based on deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 : A flowchart of a power distribution gateway testing method based on deep learning provided in Example 1 of the present invention;

[0048] Figure 2 : A structural diagram of a power distribution gateway testing system based on deep learning provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] Embodiment one:

[0051] Please refer to Figure 1 , a power distribution gateway testing method based on deep learning provided by an embodiment of the present invention, the method includes steps S1 to S4, each step is specifically as follows:

[0052] Step S1, when using the operation and maintenance debugging tool to test the tested power distribution gateway, collect data generated by the tested power distribution gateway during the test process as test data.

[0053] In this embodiment, to test the distribution gateway under test, the operation and maintenance debugging tool is first used to edit the test case and start the test, so that the test case data is sent to the tester driver module via the protocol. Next, the tester driver module controls the hardware module to output the test data to the corresponding end device based on the parsed test case data. The end device collects the received test data and sends the data to the distribution gateway under test via protocols such as RS485, IEC104, and MQTT. The distribution gateway under test then uploads the test data to the operation and maintenance debugging tool via the MQTT protocol. The operation and maintenance debugging tool compares and analyzes the test data with the standard data, automatically generating an electronic test report, and finally storing the electronic report in the test report library for access anytime and anywhere. After the test is started, the test software calls and executes the test script according to the pre-edited test case to start the first test item. After the current test item is executed, the test start index moves to the next item to be tested. When the test item switches from the start state to the pause state, the test device stops executing the test case and saves the current test section data. After clicking Start Execution, the test case can continue to execute, thus realizing automatic control of the test process. Among them, the test case is the test execution strategy.

[0054] It should be noted that the operation and maintenance debugging tool includes a device cluster and a computing engine. The device cluster can perform data export, device control, and device management, while the computing engine can perform test case editing, test case execution, data collection and analysis, test report analysis, test log management, and deep learning management. Specifically, data export is the export and viewing of data saved during the test process; device control is the management and control of the device's test process; device management is the management of information such as the IP address and port number of the tested distribution gateway; test case editing is the writing of distribution gateway operation and maintenance debugging cases; test case execution is the control of the hardware device's output test volume based on the written test case; data collection and analysis is the pass / fail analysis of the data collected by the tested device, and the test process data is stored in a database to facilitate deep learning analysis; test report management is the automatic filling of test reports based on the data collection and analysis results and saved as a Word or PDF electronic version; test log management is the recording of the test process; deep learning management can mine and classify test data, establish a test sample database, and establish intelligent testing strategies based on deep learning. It can intelligently identify data from different manufacturers, summarize common problems of different manufacturers, and automatically generate targeted test cases.

[0055] As a preferred solution, step S1 may include steps S11 to S13, each of which is specifically as follows:

[0056] Step S11, when using the operation and maintenance debugging tool to execute the current test item to test the tested power distribution gateway, it is determined in real time whether the current test item is the last test item; if so, step S12 is executed; if not, step S13 is executed.

[0057] Step S12: Fill the test related information into the preset test report template, generate the corresponding test report and give the corresponding test conclusion.

[0058] The test-related information includes the test data corresponding to each test item, the standard data, the absolute error between the test data and the standard data, and the preset threshold corresponding to the absolute error.

[0059] In this embodiment, a pre-set test report template is populated with test data, standard data, errors, set thresholds, and other information to generate a test report in a user-defined format, along with the test conclusion. The test report supports both Word and PDF formats; the test conclusion includes whether each test item passed or failed, and the reasons for failure for any test item.

[0060] As an example, when telemetry fails, the test data, standard data, and error magnitude are provided, and the cause of the failure is determined to include one or more of the following: incorrect time point number (misalignment), incorrect phase sequence (one phase value is large, another phase value is zero), and inaccurate accuracy (greater than the error range). As another example, when telesignaling and remote control fail, the cause of the failure can be determined to include one or more of the following: incorrect point number (misalignment), control pulse interval (no command received), and incorrect command format (single or double point position).

[0061] Step S13: The index moves to the next test item, so that the operation and maintenance debugging tool continues to execute the next test item, tests the tested power distribution gateway, and collects data generated by the tested power distribution gateway during the test process.

[0062] As a preferred solution, step S1 may further include steps S14 to S17, each of which is specifically as follows:

[0063] Step S14, when the operation and maintenance debugging tool is used to execute the current test project to test the tested power distribution gateway, data generated by the tested power distribution gateway during the test process is collected in real time as test data corresponding to the current test project.

[0064] Step S15, calculate the absolute error between the test data corresponding to the current test item and the standard data corresponding to the current test item, and determine whether the calculated absolute error is greater than a preset threshold; if so, execute step S16; if not, execute step S17.

[0065] Step S16: determine that the test conclusion of the current test item is unqualified.

[0066] Step S17: determine whether the test result of the current test item is qualified.

[0067] Step S2: Perform cluster analysis on all test data, and classify each test data according to the cluster analysis result to form a test sample database of multiple manufacturers.

[0068] As a preferred solution, step S2 includes steps S21 to S23, and each step is specifically as follows:

[0069] In step S21 , all test data are classified into manufacturer data according to the pre-input manufacturer names to construct an initial database of multiple manufacturers.

[0070] In step S22 , a K-means algorithm is used to perform cluster analysis based on the distance between each pair of test data, and each test data is classified into a corresponding initial database according to the cluster analysis result.

[0071] It should be noted that the core idea of ​​the K-means algorithm is to select k data points from n data sets as the original cluster centers, and then determine which category they belong to by combining the distance between the remaining nk data points and these k data points; then obtain the average value of the data in each category, redefine the cluster center, and repeat the above steps until the objective function converges. Based on this, the operation and maintenance debugging tool can implement classified management of equipment data from different manufacturers, establish a sample library and test library based on cluster learning analysis, and use the protection trip control bit, zero-sequence fault detection control bit, automatic reclosing simulation control bit, secondary reclosing simulation control bit, protection trip time (milliseconds), zero-sequence protection trip time (milliseconds), automatic reclosing time (milliseconds), secondary reclosing time (milliseconds), post-acceleration time (milliseconds), first-stage overcurrent setting (A), first-stage zero-sequence overcurrent setting (A), normal voltage value (V), and residual voltage value (V) as reference data to establish a sample library of fault test data for each manufacturer, such as circuit breaker faults, short-circuit faults, ground faults, and polarity faults.

[0072] In this embodiment, a K-means algorithm is used, with the database of the nth manufacturer as the nth data set, and k data are screened from the n data sets as initial cluster centers. Cluster analysis is performed on the databases of different manufacturers to further improve the databases of different manufacturers, maximize the distance between different databases (classes), and ensure that the distance within a database (class) is minimized. Specifically, this step includes steps S221 to S224, each of which is as follows:

[0073] Step S221: randomly select k data from n data sets as initial cluster centers μ1, μ2, ..., μ k ∈R n .

[0074] Step S222: Obtain the distance between the remaining data and the k data, and refer to formula (1) to determine the distance between each remaining data x i Category C i .

[0075]

[0076] Where x j Represents any one of the k data extracted.

[0077] Step S223: For all categories j, refer to formula (2) and reselect the center μ of each category j .

[0078]

[0079] In step S224, if the standard measurement function converges, that is, the optimal solution for the sum of squared errors within the cluster is reached (making the K-means algorithm evaluation function fitness (A[1], A[2], ..., A[n]) have the minimum value), the program stops, otherwise it returns to step S222. The calculation formula of the K-means algorithm evaluation function is shown in formulas (3) and (4).

[0080]

[0081] Where, Dist(x i , C k ) represents x i data and the center point C k The distance between them. First determine the original value C k , and then calculate x i The value of, see formula (5), combined with x i The value of C k Optimize and finally obtain the best cluster center. Reduce the distance between data centers to achieve better convergence performance.

[0082]

[0083] In addition, the K-means algorithm evaluation function is improved by adding the weight factor W k After weighted processing, the sample is divided into the category closest to which sample. Among them, the weight factor W k For the expression of , please refer to formula (6).

[0084]

[0085] Where, ε k Represents standard deviation.

[0086] Assume the target is C k When k' is used, the function increment is expressed as Δε k Among them, Δε k For the expression of , please refer to formula (7).

[0087]

[0088] It should be noted that introducing the above algorithm into the system can realize the classification management of data, establish a sample library and test library based on cluster learning analysis, and use the protection tripping control bit, zero-sequence fault detection control bit, automatic reclosing simulation control bit, secondary reclosing simulation control bit, protection tripping time (milliseconds), zero-sequence protection tripping time (milliseconds), automatic reclosing time (milliseconds), secondary reclosing time (milliseconds), post-acceleration time (milliseconds), one-stage overcurrent setting (A), one-stage zero-sequence overcurrent setting (A), normal voltage value (V), and residual voltage value (V) as reference data to establish a sample library of fault test data such as circuit breaker faults, short circuit faults, grounding faults, and polarity faults from various manufacturers.

[0089] Step S23: After all test data are classified, the current initial database of each manufacturer is used as the test sample database of each manufacturer.

[0090] In this embodiment, after establishing a classification sample library of different manufacturers, during on-site operation and maintenance or retesting the same manufacturer, targeted testing can be performed based on common errors of different manufacturers, thereby reducing dependence on the manufacturer.

[0091] In step S3, the MAAC algorithm is used to evaluate the local environment observation status after the operation and maintenance debugging tool performs each test action based on the test sample database of multiple manufacturers.

[0092] As a preferred solution, step S3 includes step S31, which is specifically as follows:

[0093] In step S31, the data contained in the test sample databases of multiple manufacturers are input into the intelligent agent evaluation network, so that the intelligent agent evaluation network outputs the estimated value of the state-action evaluation function corresponding to each test action as the evaluation result of the local environment observation state after the operation and maintenance debugging tool executes each test action.

[0094] Among them, the test sample database of multiple manufacturers contains several test actions of multiple manufacturers, as well as the local environment observation status after executing each test action; the intelligent agent evaluation network is obtained by iteratively updating the parameters of the initial evaluation network with the goal of minimizing the loss function; the initial evaluation network is pre-built using the MAAC algorithm.

[0095] It should be noted that the estimated value of the state-action evaluation function corresponding to the test action is used to measure the quality of the agent's strategy for executing the test action. It can also evaluate common faults in distribution gateways and equipment from different manufacturers. The agent, as the manager of the intelligent operation and maintenance debugging tool, the Huang Jingshi operation and maintenance debugging tool test system, observes the changing state of the environment over time. Based on the new observation information, the agent conducts real-time testing for common faults and distribution gateways from different manufacturers, thus achieving targeted testing.

[0096] In this embodiment, the input of the agent evaluation network is the local environment observation state o and the tester test action a, and its output is the state-action evaluation function that measures the quality of the agent's corresponding action strategy. The estimated value of the target network parameter is the evaluation network parameter θ and the target evaluation network parameter The evaluation network of the agent includes the main evaluation network and the target evaluation network. The role of the target evaluation network is to improve the stability and convergence of the evaluation network. At the same time, in order to encourage the agent to actively explore the unknown space and avoid premature convergence to non-optimal strategies, the evaluation network also adds a policy entropy term, namely

[0097] Specifically, the expression of the loss function is shown in formula (8). The loss function is used to quantify the value of network evaluation. and expected cumulative reward y m The difference between them. Where, referring to formula (9), the expected cumulative reward y m It can be calculated using the Bellman equation combined with the policy entropy term. The evaluation network is iteratively updated by minimizing the loss function L(θ). Referring to Equation (10), the evaluation network parameters θ are iteratively updated by minimizing the loss function, resulting in an evaluation network that can more accurately evaluate the tester's test strategy. Specifically, during the centralized training phase, the rough quality of each tester's current test is measured based on the observed state and the tester's test items, and the accuracy and stability of the test are continuously improved.

[0098]

[0099] Where M represents the total number of agents in the entire system; E[·] represents the expected value; and D represents the experience replay buffer, which is used to store the experience data generated by the interaction between the agent and the environment. r represents the reward value, Represents the newly observed state after executing the action; represents the target action network parameter of agent m; β represents the regularization system, which is used to control the importance of entropy; η Q To evaluate the learning rate of the network.

[0100] In step S4, the MAAC algorithm is used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool executes each test action, and the distribution gateway under test is tested in accordance with the test execution strategy of the operation and maintenance debugging tool.

[0101] As a preferred solution, step S4 includes steps S41 to S43, and the details of each step are as follows:

[0102] In step S41, the evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action are input into the intelligent agent action network, so that the intelligent agent action network outputs the corresponding test instructions.

[0103] Among them, the intelligent agent action network is obtained by updating the network parameters of the initial action network using the gradient ascent method; the initial action network is pre-built using the MAAC algorithm.

[0104] It should be noted that the input of the action network of agent m is the local observation state o m , whose output is the tester test instruction a m , the parameters to be optimized are the action network parameters and target action network parameters Similar to the agent evaluation network, the agent action network also introduces a target network and a policy entropy term.

[0105] In this embodiment, the MAAC algorithm uses the policy gradient method to train the action network. See formula (13). The agent updates its own action network parameters based on the gradient ascent method. A better test execution strategy for the operation and maintenance debugging tool manager is obtained. Specifically, during the centralized training phase, the operation and maintenance debugging tool manager continuously updates the test execution strategy of the operation and maintenance debugging tool based on the current environment observation status and the test items of the operation and maintenance debugging tool. The policy gradient expression can be found in Equations (11) and (12).

[0106]

[0107] Where, Indicates the selection of test execution strategy Reward expectations; is a scoring function that represents the gradient of the logarithm of the policy probability with respect to the parameter, and is used to quantify the impact of changes in network parameters on the policy; Expressed as the advantage function of agent m, used to measure the current test instruction a m The value of is superior to the average level, thus guiding the search for better actions; b(o,a M\m ) represents the expected Q-values ​​of all the actions of agents except agent m; is the policy entropy term; λ is the regularization coefficient; η μ is the learning rate of the action network.

[0108] After updating the agent evaluation network and the agent action network, the parameters of the target evaluation network and the target action network of the operation and maintenance debugging tool are also updated. For details, please refer to formulas (14) and (15).

[0109]

[0110] Where, represents the parameter of the target evaluation network; τ<<1 represents the soft update coefficient; Represents the parameters of the target action network. Repeat the above network parameter update process until the network parameters θ of the agent evaluation network and the network parameters of the agent action network are Reach the best.

[0111] Step S42: Send the test instruction to the operation and maintenance debugging tool through the manager.

[0112] In this embodiment, the intelligent agent action network is used to obtain the test execution strategy of the operation and maintenance debugging tool. During the training phase, the operation and maintenance debugging tool uses the current environmental observation status and the test items of the tester to train the test execution strategy. During the execution phase, the operation and maintenance debugging tool manager issues test instructions to each digital distribution network test based on the output of the action network, enabling targeted testing of common distribution gateway faults and equipment from different manufacturers.

[0113] In step S43, when the operation and maintenance debugging tool receives the test instruction, the operation and maintenance debugging tool obtains the test execution strategy of the operation and maintenance debugging tool based on the test instruction, and tests the power distribution gateway under test according to the test execution strategy of the operation and maintenance debugging tool.

[0114] In this embodiment, data from different manufacturers are classified according to the test data, and then the common errors of different manufacturers can be mined and analyzed based on the MAAC algorithm, a deep learning algorithm, to generate targeted test plans for manufacturers and common faults. In accordance with the targeted test plans for manufacturers and common faults, the current manufacturer is automatically tested, so that when similar errors are encountered in on-site operation and maintenance testing, the problem can be handled without relying on the equipment manufacturer. Moreover, the next time the same manufacturer sends the same equipment for inspection, a targeted test plan for the common errors of the manufacturer can be automatically generated according to the previous data mining results for automatic testing, thereby improving the test directionality and efficiency.

[0115] Please refer to Figure 2 , which is a schematic diagram of the structure of a power distribution gateway test system based on deep learning provided by an embodiment of the present invention. The system includes a data acquisition module M1, a cluster analysis module M2, a state assessment module M3, and a strategy analysis module M4. The details of each module are as follows:

[0116] The data acquisition module M1 is used to collect data generated by the tested power distribution gateway during the test process as test data when the tested power distribution gateway is tested using the operation and maintenance debugging tool;

[0117] Cluster analysis module M2, used to perform cluster analysis on all test data, and classify each test data according to the cluster analysis results to form a test sample database of multiple manufacturers;

[0118] The state assessment module M3 is used to evaluate the local environment observation state after the operation and maintenance debugging tool performs each test action based on the test sample database of multiple manufacturers using the MAAC algorithm;

[0119] The strategy analysis module M4 is used to adopt the MAAC algorithm to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool executes each test action, and test the distribution gateway under test according to the test execution strategy of the operation and maintenance debugging tool.

[0120] As a preferred solution, the strategy analysis module M4 specifically includes an instruction generation unit 41, an instruction sending unit 42 and an instruction response unit 43. The details of each unit are as follows:

[0121] The instruction generation unit 41 is used to input the evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action into the intelligent agent action network, so that the intelligent agent action network outputs the corresponding test instructions; wherein the intelligent agent action network is obtained by updating the network parameters of the initial action network using the gradient ascent method; the initial action network is pre-built using the MAAC algorithm;

[0122] The instruction sending unit 42 is used to send the test instruction to the operation and maintenance debugging tool through the manager;

[0123] The instruction response unit 43 is used to obtain the test execution strategy of the operation and maintenance debugging tool based on the test instruction when the operation and maintenance debugging tool receives the test instruction, and test the tested power distribution gateway according to the test execution strategy of the operation and maintenance debugging tool.

[0124] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0125] An embodiment of the present invention also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a distribution gateway testing method based on deep learning as described in Example 1.

[0126] An embodiment of the present invention also provides a terminal, including a processor, a memory, and a computer program stored in the memory; wherein the computer program can be executed by the processor to implement a power distribution gateway testing method based on deep learning as described in Example 1.

[0127] Preferably, the computer program can be divided into one or more modules / units (e.g., computer program, computer program), one or more modules / units are stored in a memory and executed by a processor to implement the present invention. One or more modules / units can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal.

[0128] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor can also be any conventional processor. The processor is the control center of the terminal and uses various interfaces and lines to connect various parts of the terminal.

[0129] The memory mainly includes a program storage area and a data storage area. The program storage area can store the operating system, at least one application required for a function, etc., and the data storage area can store related data, etc. In addition, the memory can be a high-speed random access memory or a non-volatile memory such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, and a Flash Card, etc., or other volatile solid-state memory devices.

[0130] It should be noted that the above-mentioned terminal may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that the above-mentioned terminal is merely an example and does not constitute a limitation on the terminal. It may include more or fewer components, or a combination of certain components, or different components.

[0131] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0132] The present invention provides a power distribution gateway testing method, system, medium and terminal based on deep learning. When using an operation and maintenance debugging tool to test the tested power distribution gateway, the data generated by the tested power distribution gateway during the test process is collected as test data, and then all the test data are clustered and analyzed. According to the cluster analysis results, each test data is classified and processed, and data belonging to different manufacturers are intelligently identified to form a test sample database of multiple manufacturers, providing sample data support for the deep learning of the operation and maintenance debugging tool, solving the long-standing problem that test data cannot be effectively mined and extracted, effectively meeting the pertinence of field test cases brought by massive samples and the accuracy of diagnostic results, and facilitating the subsequent summary of common problems of different manufacturers. Then, the MAAC algorithm, a deep learning algorithm, is used to evaluate the operation and maintenance debugging tool's execution of various test actions based on the test sample database of multiple manufacturers. The local environment observation state after the operation and maintenance debugging tool executes each test action is then evaluated, and the MAAC algorithm is then used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action. The tested distribution gateway is tested in accordance with the test execution strategy of the operation and maintenance debugging tool, thereby forming a targeted test plan for manufacturers and common problems. Through the MAAC algorithm, the test sample databases of multiple manufacturers can be deeply mined, and then the local environment observation state after the operation and maintenance debugging tool executes each test action is evaluated. The test execution strategy of the operation and maintenance debugging tool is obtained based on the analysis of the evaluation results, so that the test execution strategy can adapt to distribution gateways of different manufacturers and models, and then adapt to different test requirements and environmental changes, provide users with a wider range of test coverage, reduce unnecessary repeated testing, and improve test efficiency.

[0133] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A power distribution gateway testing method based on deep learning, characterized in that: include: When using the operation and maintenance debugging tool to test the tested power distribution gateway, collecting data generated by the tested power distribution gateway during the test process as test data; Performing cluster analysis on all the test data, and classifying each test data according to the cluster analysis results to form a test sample database of multiple manufacturers; Using the MAAC algorithm, based on the test sample databases of the multiple manufacturers, the local environment observation state of the operation and maintenance debugging tool after executing each test action is evaluated; The MAAC algorithm is used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool performs each test action, and the tested distribution gateway is tested in accordance with the test execution strategy of the operation and maintenance debugging tool.

2. A power distribution gateway testing method based on deep learning according to claim 1, characterized in that: The MAAC algorithm is used to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool performs each test action, and the tested power distribution gateway is tested according to the test execution strategy of the operation and maintenance debugging tool, specifically: Inputting the evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action into the intelligent agent action network, so that the intelligent agent action network outputs the corresponding test instructions; wherein the intelligent agent action network is obtained by updating the network parameters of the initial action network using the gradient ascent method; the initial action network is pre-built using the MAAC algorithm; Sending the test instruction to the operation and maintenance debugging tool through the manager; When the operation and maintenance debugging tool receives the test instruction, the operation and maintenance debugging tool obtains the test execution strategy of the operation and maintenance debugging tool based on the test instruction, and tests the tested power distribution gateway according to the test execution strategy of the operation and maintenance debugging tool.

3. A power distribution gateway testing method based on deep learning according to claim 1, characterized in that: The MAAC algorithm is used to evaluate the local environment observation status after the operation and maintenance debugging tool performs each test action based on the test sample database of the multiple manufacturers, specifically: Inputting data contained in the test sample databases of the multiple manufacturers into an agent evaluation network, so that the agent evaluation network outputs an estimated value of a state-action evaluation function corresponding to each test action as an evaluation result of the local environment observation state after the operation and maintenance debugging tool executes each test action; Among them, the test sample database of multiple manufacturers contains several test actions of multiple manufacturers, as well as the local environment observation state after executing each test action; the intelligent agent evaluation network is obtained by iteratively updating the parameters of the initial evaluation network with the goal of minimizing the loss function; the initial evaluation network is pre-built using the MAAC algorithm.

4. A power distribution gateway testing method based on deep learning according to claim 1, characterized in that: The cluster analysis is performed on all the test data, and the test data are classified according to the cluster analysis results to form a test sample database of multiple manufacturers, specifically: Classifying all the test data into manufacturer data according to the pre-input manufacturer names to build an initial database of the multiple manufacturers; Using the K-means algorithm, cluster analysis is performed based on the distance between each two test data, and according to the cluster analysis results, each of the test data is classified into the corresponding initial database; After all the test data are classified, the current initial database of each manufacturer is used as the test sample database of each manufacturer.

5. A power distribution gateway testing method based on deep learning according to claim 1, characterized in that: When the tested power distribution gateway is tested using the operation and maintenance debugging tool, data generated by the tested power distribution gateway during the test process is collected as test data, specifically: When using the operation and maintenance debugging tool to execute the current test item to test the tested power distribution gateway, it is determined in real time whether the current test item is the last test item; If so, the test-related information is filled into a preset test report template, a corresponding test report is generated, and a corresponding test conclusion is given; wherein the test-related information includes the test data and standard data corresponding to each test item, the absolute error between the test data and the standard data, and the preset threshold value corresponding to the absolute error; If not, the index moves to the next test item, so that the operation and maintenance debugging tool continues to execute the next test item, tests the tested power distribution gateway and collects data generated by the tested power distribution gateway during the test process.

6. A power distribution gateway testing method based on deep learning according to claim 1, characterized in that: When the tested power distribution gateway is tested using the operation and maintenance debugging tool, data generated by the tested power distribution gateway during the test process is collected as test data, specifically: When using the operation and maintenance debugging tool to execute the current test project to test the tested power distribution gateway, real-time collection of data generated by the tested power distribution gateway during the test process as test data corresponding to the current test project; Calculate the absolute error between the test data corresponding to the current test item and the standard data corresponding to the current test item, and determine whether the calculated absolute error is greater than a preset threshold; If yes, the test conclusion of the current test item is determined to be unqualified; If not, the test conclusion of the current test item is determined to be qualified.

7. A power distribution gateway testing system based on deep learning, characterized in that: include: A data acquisition module is used to collect data generated by the tested power distribution gateway during the test process as test data when the tested power distribution gateway is tested using the operation and maintenance debugging tool; A cluster analysis module, configured to perform cluster analysis on all the test data, and classify each test data according to the cluster analysis results to form a test sample database of multiple manufacturers; A state assessment module, configured to use the MAAC algorithm to assess the local environment observation state after the operation and maintenance debugging tool executes each test action based on the test sample database of the multiple manufacturers; The strategy analysis module is used to adopt the MAAC algorithm to analyze the test execution strategy of the operation and maintenance debugging tool based on the evaluation results of the local environment observation status after the operation and maintenance debugging tool performs each test action, and test the distribution gateway under test in accordance with the test execution strategy of the operation and maintenance debugging tool.

8. A power distribution gateway testing system based on deep learning as claimed in claim 7, characterized in that: The strategy analysis module specifically includes: An instruction generation unit is configured to input evaluation results of the local environment observation state after the operation and maintenance debugging tool executes each test action into the agent action network, so that the agent action network outputs corresponding test instructions; wherein the agent action network is obtained by updating the network parameters of the initial action network using the gradient ascent method; the initial action network is pre-built using the MAAC algorithm; An instruction sending unit, configured to send the test instruction to the operation and maintenance debugging tool through a manager; An instruction response unit is used to obtain the test execution strategy of the operation and maintenance debugging tool based on the test instruction when the operation and maintenance debugging tool receives the test instruction, and test the tested power distribution gateway in accordance with the test execution strategy of the operation and maintenance debugging tool.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a power distribution gateway testing method based on deep learning as described in any one of claims 1 to 6.

10. A terminal, characterized in that: It comprises a processor, a memory and a computer program stored in the memory; wherein, the computer program can be executed by the processor to implement a power distribution gateway testing method based on deep learning as described in any one of claims 1 to 6.

Images