Testing Method and System for Substation Measurement and Control Devices Based on Neural Networks

Abstract

This application relates to the field of power distribution system technology, and in particular to a testing method and system for substation measurement and control devices based on neural networks. The method includes: constructing a full-link signal topology map of the substation; extracting the correlation features of each signal node from various types of configuration files; inputting the full-link signal topology map and correlation features into a pre-trained graph neural network, which matches signal node pairs to obtain matching point pairs, and generating a test sequence consisting of simulated excitation signals and expected response information based on the matching point pairs; triggering the corresponding simulated excitation signal according to the test sequence, acquiring feedback signals in real time, comparing the feedback signals with the expected response information, and determining the test results. This application effectively reduces the manual burden of testing substation measurement and control devices and improves test accuracy.

Description

Technical Field

[0001] This application relates to the field of power distribution system technology, and in particular to a test method and system for substation measurement and control devices based on neural networks. Background Technology

[0002] With the widespread application of smart substations, substation secondary systems are gradually shifting from traditional hard-wired methods to digital communication based on process layer and station control layer networks. In this system, secondary equipment such as measurement and control devices, merging units, and intelligent terminals interact through fiber optic networks to jointly complete the acquisition, processing, and control of the operating status of primary equipment.

[0003] The merging unit is typically installed near the primary equipment to sample analog quantities such as current and voltage from the primary equipment and send the sampling results to the associated secondary equipment in the form of process-level digital messages. The intelligent terminal is typically used to receive control commands issued by the measurement and control device, drive primary equipment such as circuit breakers to perform corresponding actions, and collect switch status data and send it up via process-level communication. The measurement and control device, as the core equipment of the secondary system, is deployed in the control room to receive process-level signals from the merging unit and the intelligent terminal, process the signals to form remote signaling, telemetry, remote control and remote adjustment information, and interact with the upper-level system through station control layer communication.

[0004] During secondary operations in smart substations, the configuration files of the measurement and control devices typically need to be modified, downloaded, and verified before commissioning or after modification to ensure that the virtual loop relationships, communication relationships, and signal mapping relationships between the measurement and control devices, merging units, and smart terminals meet design requirements. Since the measurement and control devices involve multiple types of signals, including remote signaling, telemetry, remote control, and remote adjustment, these signals are mainly transmitted at the process layer via GOOSE and SMV messages, and interact at the station control layer via MMS messages. Their signal links often span multiple secondary devices, forming a complex end-to-end signal channel.

[0005] In current engineering practice, end-to-end testing of measurement and control devices mainly relies on manual methods. Specifically, testers typically manually trace the virtual connections between the measurement and control device, merging unit, and intelligent terminal based on the SCD file, and then perform segmented testing on process-level signals and station control-level signals. During the testing process, it is necessary to verify the configuration file content, communication parameters, and signal point tables one by one, and observe the response of the measurement and control device by manually applying analog or hard-contact signals. This method not only requires repeatedly switching test objects, but also relies heavily on the experience of the testers.

[0006] Currently, with the expansion of smart substations and the increasing complexity of secondary systems, the number of virtual loops associated with monitoring and control devices continues to rise, and the relationships between multi-source configuration files are becoming increasingly complex. Relying solely on manual full-link testing often requires repeatedly verifying signal mapping relationships across multiple nodes, involving numerous testing steps, a large workload, and is prone to configuration errors due to omissions or misunderstandings. Inconsistent configurations or incorrect signal links may lead to malfunctions or failures to operate after the device is put into operation, thereby affecting the safe and stable operation of the substation.

[0007] Furthermore, existing testing methods often rely on a single configuration file or a single signal type, lacking systematic verification methods for the consistency between multiple source configuration files. For example, the configuration files of the merging unit, intelligent terminal, and measurement and control device may differ in version, parameters, or signal definitions, but these differences are difficult to detect in a timely manner in the existing testing process.

[0008] In summary, the relevant end-to-end testing methods have a high risk of missing detections in practical applications, which affects the long-term stable operation of substations. Summary of the Invention

[0009] To reduce the occurrence of missed detections during end-to-end testing, this application provides a test method and system for substation measurement and control devices based on neural networks.

[0010] Firstly, this application provides a test method for substation measurement and control devices based on neural networks, employing the following technical solution:

[0011] A test method for substation measurement and control devices based on neural networks includes: constructing a full-link signal topology map of the substation, wherein the full-link signal topology map includes source end signal point nodes representing the starting point of signal flow and target end signal point nodes representing the ending point of signal flow.

[0012] Extract the associated features of each signal node from various types of configuration files;

[0013] The full-link signal topology and associated features are input into a pre-trained graph neural network. The graph neural network calculates the matching probability between source signal nodes and target signal nodes. Two signal nodes with a matching probability greater than a preset threshold are taken as matching point pairs. A test sequence consisting of simulated excitation signals and expected response information is generated based on the matching point pairs.

[0014] The corresponding simulation excitation signal is triggered according to the test sequence, the feedback signal is collected in real time, the feedback signal is compared with the expected response information, and the test result is judged.

[0015] By constructing a full-link signal topology diagram containing source-end and target-end signal nodes, extracting the association features of signal nodes from various types of configuration files, and introducing a pre-trained graph neural network to calculate the matching relationship between signal points, a test sequence consisting of simulated excitation signals and expected response information is automatically generated, thereby realizing the systematic testing of the entire link signal channel of the measurement and control device.

[0016] Compared to methods that rely on manual parsing of configuration files and applying test signals segment by segment, this method can simultaneously cover the signal links of both the process layer and the station control layer within a unified topology framework, avoiding missed detections caused by human interpretation bias. By selecting reasonable source and target signal point pairs based on matching probability, it ensures that the generated test sequence is highly consistent with the actual configuration logic, thereby enabling the simulated excitation signal to accurately reach the corresponding signal channel. Automated judgment of test results is achieved by using the expected response information as the criterion. In summary, this method improves the completeness and accuracy of end-to-end testing while reducing manual intervention and testing complexity.

[0017] Optionally, the loss function in the training process of the graph neural network is a joint loss function, which includes a cross-entropy loss term and a feature distance constraint term, wherein the feature distance constraint term is used to calculate the similarity between the source signal node and the target signal node in a specific level feature space.

[0018] In the training process of graph neural networks, a joint loss function including cross-entropy loss and feature distance constraint is introduced. This allows the model to not only focus on the correctness of the matching results when learning signal point matching relationships, but also to constrain the similarity between source and target signal point nodes in a specific level of feature space. This improves the robustness of matching signal point nodes under conditions of differences or noise in multi-source configuration files, enabling the trained graph neural network to maintain stable matching performance in different engineering configuration scenarios, and enhancing the generalization ability and engineering applicability of the end-to-end testing scheme.

[0019] Optionally, for any signal node, multiple association features are divided into multiple levels based on their importance to the connection relationship between the signal node and the association features; multiple guiding coefficients are preset according to the level of the association features.

[0020] The sum of the feature vector distances corresponding to the associated features at each level, weighted by the corresponding guiding coefficients, is used as the feature distance constraint term.

[0021] The importance of the connection relationship between signal nodes is graded according to the correlation features, and a guiding coefficient is preset for the correlation features of different levels. When calculating the feature distance constraint term, the distance of the feature vectors of each level is weighted and summed. This enables the graph neural network to distinguish the role of different correlation features in link identification during the training process. In this way, it can better fit the clear correlation between multiple source configuration files in actual engineering, and make the matching results output by the model more consistent with the real signal flow logic, thereby improving the accuracy and stability of the whole link signal matching.

[0022] Optionally, the secondary equipment includes the device under test and control, as well as the associated merging unit and intelligent terminal.

[0023] Secondary equipment is limited to the measurement and control device under test (DUT) and its associated merging units and intelligent terminals, ensuring that the full-link test objects cover key equipment types directly related to the DUT in the secondary system of a smart substation. Simultaneously, by analyzing the connection relationships between the DUT and other secondary equipment, irrelevant signal channels or redundant links can be eliminated, such as signal channels in the merging unit that are not associated with the DUT.

[0024] Optional, multiple configuration file types include at least: SCD file, CID file, ICD file, and CCD file.

[0025] The SCD file describes the logical connections at the entire station level, while the CID and CCD files reflect the configuration status after device instantiation. The ICD file provides the original model information of the device. By jointly parsing the above multi-source configuration files, the signal definitions, communication parameters, and virtual connection relationships can be fully understood before testing, thus avoiding the problem of incomplete information caused by testing based on only a single configuration file.

[0026] Optionally, before triggering the corresponding simulation excitation signal according to the test sequence, the method further includes: obtaining the operation configuration CRC code of the measurement and control device and the associated secondary equipment, verifying the consistency between the operation configuration CRC code and the corresponding CRC code configured in the SCD file, and determining that the verification is passed in response to the operation configuration CRC code being the same as the corresponding CRC code configured in the SCD file.

[0027] By comparing the CRC codes of the operation configuration of the measurement and control device and associated secondary equipment with the corresponding CRC codes in the SCD file, it is ensured that the actual operation configuration participating in the test is consistent with the design configuration, thereby reducing the possibility of incorrect test results due to misconfiguration.

[0028] Optionally, the matching probability between the source signal node and the target signal node is calculated by a graph neural network, including: iteratively fusing the feature vectors of the signal nodes through the graph neural network to generate signal feature vectors of the signal nodes; concatenating the signal feature vectors of the two signal nodes to obtain a joint feature vector; and inputting the joint feature vector into a multilayer perceptron for processing to obtain the matching probability.

[0029] Optionally, the test results are judged by comparing the feedback signal with the expected response information, including: evaluating the feedback signal in multiple dimensions based on the difference between the feedback signal and the expected response information; if the response meets the preset requirements in both the multi-dimensional evaluation and the expected response information, the test is deemed to have passed.

[0030] By comprehensively analyzing the differences between the feedback signal and the expected response information, the robustness of the test result judgment can be improved.

[0031] Optionally, the feedback signal can be evaluated in multiple dimensions, including: assessing the timeliness of the feedback signal response based on the distance between the feedback signal acquisition time and the corresponding trigger signal trigger time; and assessing the response consistency score of the feedback signal based on the numerical difference between the feedback signal and the expected signal information.

[0032] By evaluating feedback signals from both time and numerical dimensions, the testing process can identify not only response delay issues but also numerical deviation issues, thereby enhancing the end-to-end testing capability to detect potential problems.

[0033] Secondly, this application provides a test system for substation measurement and control devices based on neural networks, which adopts the following technical solution:

[0034] The substation measurement and control device testing system based on neural networks includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned substation measurement and control device testing method based on neural networks.

[0035] The aforementioned test method for substation monitoring and control devices based on neural networks is used to generate a computer program, which is then stored in a memory for loading and execution by a processor. This allows for the creation of a system based on the memory and processor, making it convenient to use.

[0036] This application has the following technical effects:

[0037] By leveraging GNN deep analysis of multi-source configuration files such as SCD, CID, ICD, and CCD, and employing a joint loss function and a hierarchical correlation feature weighting mechanism, complex virtual loop signal topologies are automatically learned and accurately identified. This enables the automatic generation of test sequences and automated testing of measurement and control devices, reducing the inefficiency and high rate of missed detections associated with manual testing. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the testing method for a substation monitoring and control device based on a neural network, according to an embodiment of this application.

[0039] Figure 2 This is a schematic diagram showing the connection relationships between various secondary devices.

[0040] Figure 3 This is an example diagram showing the connection relationship between signal nodes. Detailed Implementation

[0041] This application discloses a test method for substation measurement and control devices based on neural networks. It involves deep analysis of the substation configuration description file (SCD) to extract the list of secondary equipment, logical connections, and configuration information within the substation, serving as the underlying data foundation for the entire end-to-end test. Combining multi-source configuration files, a graph neural network (GNN) is used to construct the end-to-end signal association topology. Through its powerful feature extraction and weight calculation capabilities, a test sequence covering "four remote" signals (remote signaling, telemetry, remote control, and remote adjustment) is automatically generated, replacing the inefficient manual matching process. This ensures that the logically ordered test sequence can be accurately executed in complex field environments.

[0042] Reference Figure 1 The test method for substation measurement and control devices based on neural networks includes steps S1-S4.

[0043] Step S1: Construct the full-link signal topology diagram of the substation, which includes source signal node representing the starting point of signal flow and target signal node representing the ending point of signal flow.

[0044] Centered on the device under test, obtain all logical relationships between the merging units and intelligent terminals that are logically related to it, and generate a preliminary test link topology.

[0045] In this embodiment, the SCD (Substation Configuration Description) file is first parsed. The Substation Configuration Description file is the core file for configuring a smart substation, containing modeling information, communication parameters, and logical connection relationships for all secondary equipment. Secondary equipment refers to low-voltage electrical equipment used for monitoring, controlling, measuring, regulating, and protecting the power system and primary equipment. They work in conjunction with primary equipment (such as generators, transformers, and circuit breakers) to ensure the safe, stable, and economical operation of the power system.

[0046] Specifically, by using XML parsing operators (DOM parsing, SAX parsing, etc.) to read XML format data in the SCD file, a list of secondary equipment in the substation is extracted. The secondary equipment includes, but is not limited to, merging units, intelligent terminals, and measurement and control devices.

[0047] While extracting the secondary equipment list, the logical topology of the secondary system with the measured and controlled device as the central node is extracted, clarifying the logical connection relationships of data interaction between the measured and controlled device, the merging unit, and the intelligent terminal at the process layer and station control layer. For example, this logical connection relationship is defined through internal virtual connections, i.e., message transmission paths determined by a subscription and publish mechanism.

[0048] In addition, the system will also extract the configuration information of each device. The configuration information includes basic information of secondary devices, such as manufacturer name, device version number, and manufacturing date; process layer configuration information, such as message identifiers and transmission cycles of GOOSE (Generic Object Oriented Substation Event) and SMV (Sampled Value); and station control layer configuration information, such as the service access point of MMS (Manufacturing Message Specification).

[0049] The end-to-end signal topology diagram includes signal nodes and connecting edges that connect the signal nodes. In this embodiment, the signal channels in each secondary device are used as signal nodes in the end-to-end signal topology diagram. The connecting edges between the signal nodes are determined in this embodiment based on the connection relationships between the secondary devices and the signal flow.

[0050] Based on the operational logic functions of the telemetry and control device and the signal flow direction, signal nodes can be divided into four categories: telemetry signal nodes, remote signaling signal nodes, remote control signal nodes, and remote adjustment signal nodes. Each category can be further divided into two types: source signal nodes, representing the starting point of the signal flow, and target signal nodes, representing the ending point of the signal flow.

[0051] For example:

[0052] Telemetry signal nodes include:

[0053] Source-end signal point node: corresponds to the SMV message transmission channel used to transmit current or voltage data in the merging unit, such as the merging unit logic node. Instances of " A certain line "Phase current sampling" channel point.

[0054] Target signal point node: The telemetry logic channel point inside the telemetry and control device used to receive the above SMV messages and perform calculations.

[0055] Remote signaling nodes include:

[0056] Source signal point node: The intelligent terminal collects the status of primary equipment and converts it into... The signal channel for transmitting bits, such as the "circuit breaker closed signal" transmitting bit under the logic node of the smart terminal.

[0057] Target-end signal point node: A logical channel through which the telemetry and control device receives position change signals and maps them to the remote signaling status of the station control layer. For example: a logical node of the telemetry and control device. The "Circuit Breaker Position" receiving channel point is below.

[0058] Remote control signal nodes include:

[0059] Source signal point node: Corresponds to the GOOSE transmission signal channel where the measurement and control device issues control commands. Example: Logic node instance of the measurement and control device. The next signal bit is used to issue the closing command.

[0060] Target signal point node: corresponds to the GOOSE receive signal channel where the smart terminal receives the above instructions and drives the output relay.

[0061] Remote signal nodes include:

[0062] Source signal point node: corresponds to the adjustment command issued by the measurement and control device. Signal transmission channel. For example: a logic node example in a measurement and control device. The signal bit used to issue the gear adjustment command is an independent node.

[0063] Target signal point node: corresponds to the actuator (such as a smart terminal) receiving adjustment commands. Signal receiving channel.

[0064] It is important to note here that... , , The names are determined based on the conventional naming method of "logical node instances" in the Power System Configuration Language (SCL) and the IEC 61850 standard, and are mainly for the convenience of those skilled in the art to understand.

[0065] The edges between signal nodes in the full-link signal topology diagram are established based on the physical link or communication subscription relationship.

[0066] Reference Figure 3For example, when the circuit breaker in the field changes from the "closed" position to the "open" position, the smart terminal collects the hard contact signal and triggers a change in the data bits of the corresponding GOOSE message. Therefore, the remote signaling signal in the monitoring and control device used to characterize the opening and closing of the circuit breaker is associated with the data bits of the GOOSE message sent by the smart terminal to construct a link topology of "smart terminal - GOOSE message sending - monitoring and control device GOOSE message receiving".

[0067] Alternatively, when the current transformer senses a change in the primary circuit current, the merging unit converts the analog voltage and current signals into digital SMV and encapsulates them into an analog SMV message. The signal nodes in the measurement and control device used to characterize the current and voltage amplitudes are associated with the signal nodes in the analog SMV message sent by the merging unit, thus constructing a link topology of "merging unit - SMV message sending - measurement and control device SMV message receiving".

[0068] S2: Extract the associated features of each signal point node from various types of configuration files.

[0069] XML parsing operators are used to parse multi-source configuration files such as SCD, CID, ICD, and CCD to extract the association features of each signal channel node. Association features include, but are not limited to, device type and associated device extracted from SCD; signal type, channel description, and configuration validity extracted from CCD; device manufacturer and channel description extracted from ICD; and signal type, channel description, and configuration validity extracted from CID.

[0070] The CCD file represents the configured device description file, which describes the process layer configuration information of the measurement and control device. Specifically, it includes node information such as GOOSE message sending, GOOSE message receiving, SMV message sending, and SMV message receiving.

[0071] The CID file is an instance configuration description file that describes the station control layer configuration information of the telemetry and control device.

[0072] The ICD file is an instance device description file that describes the original model information of secondary devices (such as measurement and control devices). It contains the device's built-in logical nodes, data objects, and default GOOSE and SMV message configuration information.

[0073] By parsing the multi-source configuration file, a multi-dimensional initial feature vector is constructed for each signal channel node.

[0074] S3: Input the full-link signal topology map and associated features into the pre-trained graph neural network. The graph neural network calculates the matching probability between the source signal node and the target signal node. Two signal nodes with a matching probability greater than a preset threshold are taken as matching point pairs. Based on the matching point pairs, a test sequence consisting of simulated excitation signals and expected response information is generated.

[0075] During the training phase of a graph neural network, the core objective is to acquire a weight matrix that reflects the logic of power business through learning from massive amounts of samples. The weight matrix is ​​a parameter matrix used for feature transformation in a graph neural network; essentially, it is a nonlinear mapping operator. In the computation process, this weight matrix is ​​responsible for projecting the original input signal features onto a high-dimensional semantic space. By performing weighted transformations on features of different dimensions, it achieves accurate extraction and representation of signal point features.

[0076] During training, due to the varying discriminative power of features across different dimensions in link identification, this scheme categorizes the features in each dimension of the feature vectors of remote signaling, telemetry, remote control, and remote adjustment signals into four levels based on their importance to the connection of signal nodes, from highest to lowest. This hierarchical mechanism guides the graph neural network to automatically adjust the parameter distribution of the weight matrix during the training phase. This allows the model to prioritize topological searching based on core correlation features during the inference phase, supplemented by secondary features for logical verification, thereby achieving accurate identification of the entire signal link in heterogeneous environments.

[0077] Specifically, the multidimensional correlation features in the initial feature vector of the signal point node are divided into the following four levels to adapt to their respective physical and business characteristics:

[0078] For remote signal point nodes: the first-level association features are the device type and signal associated device extracted from the SCD file, which are used to describe the virtual connection matching between the smart terminal and the measurement and control device in the SCD file.

[0079] The second-level correlation features are the signal type, channel description, and configuration validity extracted from the CCD file. These are used to describe the GOOSE receiver virtual connection matching status under the GOOSESUB node of the telemetry and control device in the CCD file.

[0080] The third-level associated features are the device manufacturer and channel description extracted from the ICD file. These are used to describe... Measurement and control device in the document Logical nodes and The relationship between signal points with the same description under a logical node.

[0081] The fourth level of association features are the signal type, channel description, and configuration validity extracted from the CID file. These are used to describe the measurement and control device in the CID file. Nodes and The connection status of remote signaling signals under the node.

[0082] For telemetry signal point nodes:

[0083] The first-level association features are the device types and signal-associated devices extracted from the SCD file. These are used to describe the virtual connection matching between the merging unit and the measurement and control device in the SCD file, establishing the link foundation by locking the physical topology connection.

[0084] The second level of correlation features consists of signal type, channel description, and configuration validity extracted from the CCD file. Specifically, this may include the sampling channel sequence, synchronization sampling identifier, and signal validity flag, used to distinguish different AC sampling channels.

[0085] The third-level associated features are the equipment manufacturer and channel description extracted from the ICD file. Specifically, this may include the rated transformation ratio of the current transformer, sampling frequency, signal range upper and lower limits, and engineering unit information; used to describe the internal structure of the measurement and control device in the ICD file. Logical nodes and The association between logical nodes with the same text description signal points enables the functional alignment of the internal model of the device.

[0086] The fourth level of association features are the signal type, channel description, and configuration validity extracted from the CID file. These are used to describe the measurement and control device in the CID file. Nodes and The correlation of telemetry signals under the node serves as an auxiliary verification basis at the business level.

[0087] For remote control signal point nodes:

[0088] The first-level association features are the signal type, channel description, and configuration validity extracted from the CID file. These are used to describe the measurement and control devices in the CID file. Nodes and The logical correspondence between remote control signal points under a node ensures the uniqueness of the starting point of the control command.

[0089] The second-level associated features are the equipment manufacturer and channel description extracted from the ICD file. These are used to describe the internal structure of the measurement and control device within the ICD file. Nodes and Model association of signal points with the same description under a node.

[0090] The third-level correlation features are the signal type, channel description, and configuration validity extracted from the CCD file. These are used to describe the virtual connection matching status of the GOOSE sent by the monitoring and control device in the CCD file, and to lock the communication exit for issued commands.

[0091] The fourth-level association feature is the device type and signal-associated device in the SCD file. It is used to describe... The document details the matching of virtual connections between the monitoring and control device and the intelligent terminal, determining the physical execution device that ultimately executes the command.

[0092] For remotely tuned signal points:

[0093] The first-level association features are the signal type, channel description, and configuration validity extracted from the CID file. These are used to describe... Measurement and control device in the document Nodes and The association status of remote control signal points under the node.

[0094] The second-level associated features are the device manufacturer and channel description extracted from the ICD file. These are used to describe... The document contains the internal measurement and control device. Nodes and Model association of signal points with the same description under a node.

[0095] The third-level correlation features are the signal type, channel description, and configuration validity extracted from the CCD file. These are used to describe the measurement and control device within the CCD file. Send virtual connection matching information and lock the communication exit from which the command was issued.

[0096] The fourth-level association feature is the device type and signal-associated device in the SCD file. It is used to describe... The document details the matching of virtual connections between the monitoring and control device and the intelligent terminal, determining the physical execution device that ultimately executes the command.

[0097] Here, the remote control signal is focused on the opening and closing control of the circuit breaker, while the remote adjustment signal is used for the adjustment control of transformer taps, etc. The two are similar in configuration structure.

[0098] The aforementioned four-level association features are used to determine the weight matrix through a joint loss function constraint mechanism. This embodiment employs a joint loss function based on classification loss and feature distance constraints. The specific implementation formula is as follows:

[0099] ;

[0100] In this formula: The cross-entropy loss term is used to determine the overall accuracy of signal link matching, where These are real link labels manually assigned. The predicted matching probability by the model; This represents the source signal point node. In the eigenvectors, those belonging to the th Part of the hierarchical correlation characteristics. For example, for remote signaling signals, when At that time, it represents a node. The numerical vector of the dimension "Device type and topology connection"; Represents the target end signal point node In the eigenvectors, those belonging to the th Part of the hierarchical association features; For the first The second-norm distance constraint term of the level feature is used to calculate the similarity between the source signal node and the target signal node in a specific level feature space; For the corresponding to the first The guiding coefficients of the level features are used to artificially set the weight bias of different levels during the training phase; This is a balancing coefficient used to adjust the weight ratio between classification accuracy and feature consistency.

[0101] In this embodiment, the grading of association features is based on staff experience, and is determined according to the importance or confidence of the association features in the connection relationship. Therefore, the guiding coefficient is determined based on the grading of association features, and the guiding coefficient also corresponds to four gradings of association features, namely... ,and .

[0102] In one embodiment, the four guiding coefficients corresponding to different signal types can be the same, that is, the guiding coefficients corresponding to the first-level association features of remote signaling point nodes, telemetry point nodes, remote control signal point nodes, and remote adjustment signal point nodes are the same.

[0103] In another embodiment, multiple sets of guiding factors can be set separately according to different signal types. For example, for the correlation characteristics at each level in telemetry signal nodes and teleindication signal nodes, since the determination of telemetry and teleindication signals primarily depends on the physical device topology and virtual loop transmission path, their corresponding guiding coefficients are set as follows: Consequently, the model generates significant gradient pressure during training to compress the feature distance between these two layers, ensuring that the consistency of the physical link becomes a rigid threshold for judgment.

[0104] For the correlation features at each level in remote control signal nodes and remote adjustment signal nodes, for control signals, the safety requirement is that the model must lock the command origin with absolute precision. Therefore, their corresponding guidance coefficients are set as follows: .

[0105] Through the above training process, the resulting weight matrix solidifies the hierarchical evaluation criteria for different signal types. In practical applications, this matrix serves as a core calculation parameter, automatically performing hierarchical filtering and aggregation of input signal point features, thereby providing accurate data support for subsequent link identification and automatic generation of test sequences.

[0106] Through the above training process, the resulting weight matrix solidifies the hierarchical evaluation criteria for different signal types. In practical applications, this matrix serves as a core calculation parameter, automatically performing hierarchical filtering and aggregation of input signal point features. This allows for the extraction of signal feature vectors with unique business attributes from complex topological networks, providing accurate data support for subsequent link identification and automatic generation of test sequences.

[0107] Specifically, after inputting the full-link signal topology map and multi-source configuration files, the trained graph neural network aggregates the feature vectors corresponding to the signal nodes, and calculates the matching probability of the edges between the signal nodes after aggregation.

[0108] Specifically, the formula for the aggregation calculation of signal point nodes in a graph neural network can be expressed as:

[0109] ;

[0110] In the formula, Represents a node (such as the measured and controlled device) in the first Signal feature vector after layer iteration; This represents a nonlinear activation function, preferably a linear rectified unit (ReLU). and Representing nodes respectively and nodes The degree is used to normalize the features of neighboring nodes; Indicates the representative node With nodes Signal topology correlation values ​​between them; Indicates the first Signal feature matrix extracted by layer network; In a graph neural network, the first... The feature transformation weight matrix after the training of the layer network is completed; With nodes All neighboring nodes that are physically or logically connected A collection of (such as merging units or smart terminals associated with measurement and control devices).

[0111] After acquiring the signal feature vectors of the source and target signal nodes, the system performs the final link identification and determination using a multilayer perceptron (MLP). The specific process of MLP discrimination is described below:

[0112] Feature concatenation: The feature vectors of the source signal and the target signal are concatenated to construct a joint feature space. Subsequently, the hidden layer inside the MLP uses a weight matrix to cross-compare the "transmission semantics" of the source and the "reception semantics" of the target to identify the logical coupling between the two.

[0113] The calculation results are mapped to the matching probability using the following formula:

[0114] In the formula, Represents the source signal point node With the target end signal point node The probability of matching between them; Represents the source signal point node The signal feature vector after aggregation by the graph neural network; Represents the target end signal point node The signal feature vector after aggregation by the graph neural network; For feature splicing operators; Multilayer perceptron operator; Normalized exponential function.

[0115] when When the value is greater than or equal to a preset logical threshold, the system determines that the path is a logically matched associated edge.

[0116] A test sequence consisting of an "excitation signal" and an "expected response" is generated based on the successfully matched signal point nodes.

[0117] Excitation signal sequence: Simulated excitation signal;

[0118] Expected response information: After the simulation excitation signal simulates the action, the station control layer monitoring backend or remote terminal should return the theoretically correct value or status according to the logical relationship defined in the configuration file.

[0119] Based on the application scenario of this application, the test sequences can also be divided into four categories. These four test sequences cover four types of signal test items in the core business of intelligent substation measurement and control devices: remote signaling test sequence, telemetry signal test sequence, remote control signal test sequence, and remote adjustment signal test sequence.

[0120] The remote signal test sequence includes an excitation signal that triggers the closing of the hard contact of the smart terminal, and a corresponding remote signal change message sent by the telemetry and control device as the expected response.

[0121] The telemetry signal test sequence includes the analog current or voltage signal output by the signal excitation module, as well as the corresponding SMV message data fed back by the measurement and control device as the expected response.

[0122] The remote control signal test sequence includes the control command excitation issued by the measurement and control device, and the corresponding GOOSE displacement signal fed back by the smart terminal as the expected response.

[0123] The remote adjustment signal test sequence includes the adjustment command excitation issued by the measurement and control device, and the corresponding feedback change signal fed back by the intelligent terminal as the expected response.

[0124] S4: Trigger the corresponding simulation excitation signal according to the test sequence, collect the feedback signal in real time, compare the feedback signal with the expected response information, and judge the test result.

[0125] In one embodiment, the configuration file of the secondary device can be checked for consistency before the actual test is performed.

[0126] Specifically, a test host is deployed on the measurement and control device side, and test slave units are deployed on the merging unit and intelligent terminal side. A communication link is established between the host and slave units through a backup optical fiber, and microsecond-level communication synchronization is achieved based on the Precision Time Protocol (PTP).

[0127] The system reads the real-time cyclic redundancy check (CRC) code of the secondary equipment online and compares it with the preset configuration information in the SCD file to prevent misoperation caused by the discrepancy between the CID / CCD file in the field and the SCD file in the design stage.

[0128] Specifically, the steps for consistency verification include:

[0129] 1. Read the GOOSE message sent by the merging unit (MU) in real time, extract the configuration version CRC code carried in the message, and compare it with the predefined CRC code of the merging unit configuration in the SCD file to verify whether the operation configuration of the field merging unit is consistent with the design file.

[0130] 2. Read the GOOSE messages sent in real time by the smart terminal (IT) and parse out the current configuration CRC code. Then, the system compares the measured CRC code with the corresponding smart terminal's configuration CRC code recorded in the SCD file online to ensure that its virtual loop configuration has not been illegally tampered with or deviated from its version.

[0131] 3. Read the CCD file inside the measured and controlled device. The computing module performs real-time CRC calculation on the process layer configuration information (including GOOSE subscription / publish SMV subscription nodes, etc.) in the CCD file. The calculated real-time value is then compared with the CRC code of the corresponding process layer configuration of the measured and controlled device in the SCD file for consistency.

[0132] 4. Read the CID file inside the telemetry and control device. The system performs real-time CRC check calculations on the station control layer MMS information and model parameters described in the CID file. Finally, the calculation result is compared with the station control layer configuration CRC code reserved in the SCD file, thereby completing the consistency check of the entire link configuration from the process layer to the station control layer.

[0133] In this embodiment, the verification threshold or comparison range is preferably full-word matching. If the CRC code of each secondary device calculated in real time is completely consistent with the preset value in the design stage, the consistency verification is deemed to have passed, and the system is allowed to enter the automatic testing stage; if they are inconsistent, it is determined to be a configuration version conflict, and the system automatically locks the test exit.

[0134] After the consistency check is passed, the automatic testing phase begins. The stimulus signals are triggered sequentially according to the generated test sequence.

[0135] For example, for remote signaling signals, the test slave triggers a hard contact signal to simulate a circuit breaker position change; for telemetry signals, the signal excitation module outputs analog current and voltage signals, preferably within a certain range. to or to When the current range is selected within the nominal value to When the sampling linearity is between 1 and 2, the best sampling linearity can be obtained.

[0136] Simultaneously with the issuance of the excitation signal, the system initiates the feedback monitoring logic, which collects the corresponding feedback signal in real time. Based on the difference between the feedback signal and the expected response information, the feedback signal is evaluated from multiple dimensions. If the response meets the preset requirements in both the multi-dimensional evaluation and the response, the test is deemed to have passed.

[0137] In this embodiment, the evaluation dimensions are mainly two: evaluating the timeliness of the feedback signal response from a time perspective.

[0138] Specifically, the time difference between the issuance of the excitation signal and the receipt of the feedback signal is calculated, and this time difference is used as the response delay. The calculation formula can be expressed as follows:

[0139] ;

[0140] in, Indicates the response latency of the current test; Indicates the timing of the feedback signal acquisition; This indicates the trigger time of the excitation signal sequence.

[0141] The normalized result of the reciprocal of the response delay is used as the response timeliness. When the response timeliness is greater than a preset response threshold (for example, the remote signaling displacement response threshold is preferably...), the response timeliness is determined. If the response time is within the acceptable range, it is deemed acceptable. If the response delay exceeds this threshold, it indicates communication network congestion or a timeout in the logic processing of the monitoring and control device.

[0142] The response consistency score of the feedback signal is evaluated from the numerical dimension of the feedback signal.

[0143] The steps for obtaining the response consistency score include: calculating the numerical difference between the feedback signal and the expected response information; the formula for calculating the numerical difference can be expressed as:

[0144] ;

[0145] in, Indicates numerical differences; This represents the actual measured value in the feedback signal; This represents the expected response value in the test sequence.

[0146] The normalized result of the numerical difference is used as the response consistency score. When the response consistency score is within the allowed preset error limit, the whole link signal mapping is determined to be correct.

[0147] The test is considered passed when the response time is acceptable and the signal mapping across the entire link is correct.

[0148] By executing each of the remote signaling, telemetry, remote control, and remote adjustment signals, the automatic verification of the entire measurement and control device link is completed and a test report is generated.

[0149] This application also discloses a test system for substation measurement and control devices based on neural networks, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the test method for substation measurement and control devices based on neural networks according to this application is implemented.

[0150] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0151] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A test method for substation monitoring and control devices based on neural networks, characterized in that, include: Construct a full-link signal topology diagram of the substation, which includes source signal node representing the starting point of signal flow and target signal node representing the ending point of signal flow. Extract the associated features of each signal node from various types of configuration files; The full-link signal topology and associated features are input into a pre-trained graph neural network. The graph neural network calculates the matching probability between source signal nodes and target signal nodes. Two signal nodes with a matching probability greater than a preset threshold are taken as matching point pairs. A test sequence consisting of simulated excitation signals and expected response information is generated based on the matching point pairs. The corresponding simulation excitation signal is triggered according to the test sequence, the feedback signal is collected in real time, the feedback signal is compared with the expected response information, and the test result is judged. The loss function in the training process of a graph neural network is a joint loss function, which includes a cross-entropy loss term and a feature distance constraint term. The feature distance constraint term is used to calculate the similarity between the source signal node and the target signal node in a specific level feature space. For any given signal node, multiple association features are divided into multiple levels based on their importance to the connection relationship between the signal node and the association features; multiple guiding coefficients are preset according to the level of the association features. The sum of the feature vector distances corresponding to the associated features at each level, weighted by the corresponding guiding coefficients, is used as the feature distance constraint term.

2. The test method for substation monitoring and control devices based on neural networks according to claim 1, characterized in that, Secondary equipment includes the device under test and control, as well as the associated merging unit and intelligent terminal.

3. The test method for substation monitoring and control devices based on neural networks according to claim 1, characterized in that, Configuration files of various types include at least: SCD files, CID files, ICD files, and CCD files.

4. The test method for substation monitoring and control devices based on neural networks according to claim 1, characterized in that, Before triggering the corresponding simulation excitation signal according to the test sequence, the process further includes: obtaining the operation configuration CRC code of the measurement and control device and the associated secondary equipment, verifying the consistency between the operation configuration CRC code and the corresponding CRC code configured in the SCD file, and determining that the verification is passed in response to the operation configuration CRC code being the same as the corresponding CRC code configured in the SCD file.

5. The test method for substation monitoring and control devices based on neural networks according to claim 1, characterized in that, The matching probability between source-end signal node and target-end signal node is calculated by a graph neural network, including: iteratively fusing the feature vectors of the signal node through the graph neural network to generate the signal feature vector of the signal node; concatenating the signal feature vectors of two signal nodes to obtain a joint feature vector; and inputting the joint feature vector into a multilayer perceptron for processing to obtain the matching probability.

6. The test method for substation monitoring and control devices based on neural networks according to claim 1, characterized in that, The test results are determined by comparing the feedback signal with the expected response information, including: evaluating the feedback signal from multiple dimensions based on the differences between the feedback signal and the expected response information; if the response meets the preset requirements in both the multi-dimensional evaluation and the expected response information, the test is deemed to have passed.

7. The test method for substation monitoring and control devices based on neural networks according to claim 6, characterized in that, The feedback signal is evaluated in multiple dimensions, including: assessing the timeliness of the feedback signal response based on the distance between the feedback signal acquisition time and the corresponding trigger signal trigger time; The response consistency score of the feedback signal is evaluated based on the numerical difference between the feedback signal and the expected signal information.

8. A test system for substation measurement and control devices based on neural networks, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the test method for a substation monitoring and control device based on a neural network according to any one of claims 1-7.

Images