Secondary equipment test method, device and equipment based on generative large model

Abstract

The invention relates to the technical field of substation secondary equipment testing, and discloses a secondary equipment testing method, device and equipment based on a generative large model. The method comprises the steps that test requirements of different signal specifications are obtained, test cases are generated through a test case generation model according to standard signals, and the standard signals are obtained based on standard signal model classification and represent the signal specifications followed by different transformer substations and / or different secondary devices; the standard signal model is configured to identify a common property between device signals of different signal specifications to form a standard signal; and carrying out collaborative testing by adopting the test case to form a test result. According to the method and the device, the common property between equipment signals of different signal specifications is identified through the standard signal model to form the standard signal, and then the test case of the collaborative test is generated by the standard signal, so that the test collaboration, interoperability and test efficiency between transformer substations and / or secondary equipment of different signal specifications are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of substation secondary equipment testing, and particularly to a method, device, and equipment for secondary equipment testing based on a generative large model. Background Art

[0002] Substation secondary equipment, such as relay protection devices, as a core component of the safe and stable operation of the power grid, its efficient inspection and testing in various links such as factory production, grid connection, construction, and commissioning are crucial. The intelligent substation based on the IEC61850 standard provides a foundation for the automation and informatization construction of the power system. There are already some automated testing equipment applied to the detection of substation secondary equipment in China, and most of these equipment have certain automated testing functions. Relay protection testers adopting the IEC61850 protocol have been put into use in some power systems, and these instruments can complete some automatic testing tasks through docking with the communication protocol of the equipment.

[0003] Power system real-time digital simulation is a rapidly developing field. Especially abroad, real-time simulation technology has been applied in multiple fields, particularly for the testing of power systems and substation secondary equipment. In China, although the digital dynamic real-time simulation system has achieved certain testing functions, its simulation scale and application scenarios still have limitations. The application of system-level testing technology has also been proposed at home and abroad. Especially in the full-station commissioning of intelligent substations, the relevant testing technologies and solutions are still in the exploration stage. The existing system-level testing mainly relies on multiple relatively independent testing equipment, such as relay protection testers, time synchronization detection tools, etc. These equipment usually only test a single device or subsystem and lack a comprehensive verification of the entire secondary system. For example, in the face of system-level testing between substations with different signal specifications or between different secondary equipment, generally, multiple relatively independent substation systems or independent secondary equipment are tested separately.

[0004] Although the current automatic testing of substation secondary equipment has improved the testing efficiency to a certain extent, it still faces problems such as low automation level and complex configuration, which seriously restrict the efficient commissioning and acceptance of intelligent substations and secondary equipment. Many units in China have developed automatic testing software or functional modules for digital relay protection testers and have been applied in multiple power supply enterprises, testing institutions, and scientific research institutes across the country.

[0005] However, since the current automated testing of secondary equipment in substations is still limited to single-function testing or subsystem testing, there is a lack of a global view of the entire secondary equipment system in the substation. In addition, there are a wide variety of current secondary equipment in substations (such as relay protection devices), and there are differences in the service models of relay protection devices produced by different manufacturers. There are different signal specifications between different substations or different secondary equipment within a substation. At the same time, the interoperability of different testing equipment and testing software is poor, and it is impossible to conduct collaborative testing and acceptance of multiple systems. Especially in the scenario of collaborative debugging between multiple substations with different signal specifications or multiple secondary equipment within a substation with different signal specifications, it seriously affects the test collaboration and efficiency of multiple systems.

[0006] Therefore, the prior art has problems of poor test collaboration and low test efficiency due to different signal specifications between different substations or different secondary equipment. Summary of the Invention

[0007] The present invention provides a method, device and equipment for testing secondary equipment based on a generative large model to solve the technical problems of poor test collaboration and low test efficiency between different substations using different signal specifications and / or between different secondary equipment in a substation.

[0008] The present invention provides a method for testing secondary equipment based on a generative large model, and the method includes: Obtain the test requirements of the secondary equipment in the substation; the test requirements at least include equipment test signals with different signal specifications between different substations and / or different secondary equipment in the substation; Through a pre-trained test case generation model, generate test cases corresponding to the test requirements according to the standard signals corresponding to the equipment test signals; the standard signals are obtained by classifying the equipment signals of the secondary equipment in the substation based on a standard signal model, and are used to represent the signal specifications uniformly followed between some or all different substations and / or between some or all different secondary equipment in the substation; the standard signal model is configured to identify the common attributes between equipment signals with different signal specifications to form the standard signal; Use the test cases to conduct collaborative testing on the different substations and / or different secondary equipment in the substation to generate test results corresponding to the test requirements.

[0009] The present invention identifies the common attributes of device signals with different signal specifications through a standard signal model, and forms a standard signal representing the signal specification that is uniformly followed; furthermore, test cases for collaborative testing between substations with different signal specifications and / or different secondary devices are generated using the standard signal, which can improve the test collaboration between substations with different signal specifications and / or secondary devices. In addition, using the standard signal can ignore the signal differences between different signal specifications, improve the interoperability between substations with different signal specifications and / or secondary devices, and thus improve the test efficiency.

[0010] Furthermore, the present invention also provides a secondary device test apparatus based on a generative large model, and the apparatus includes: A test requirement acquisition module, configured to acquire the test requirements of secondary devices in a substation; the test requirements at least include device test signals with different signal specifications between different substations and / or different secondary devices in a substation; A test case generation module, configured to generate test cases corresponding to the test requirements according to the standard signal corresponding to the device test signal through a pre-trained test case generation model; the standard signal is obtained by classifying the device signals of the secondary devices in the substation based on a standard signal model, and is used to represent the signal specification that is uniformly followed between some or all different substations and / or between some or all different secondary devices in a substation; the standard signal model is configured to identify the common attributes between device signals with different signal specifications to form the standard signal; A collaborative test module, configured to perform collaborative testing on the different substations and / or different secondary devices in a substation using the test cases to generate test results corresponding to the test requirements.

[0011] In the present invention, the common attributes of device signals with different signal specifications are identified through a standard signal model, and a standard signal representing the signal specification that is uniformly followed is formed; furthermore, the test case generation module generates test cases for collaborative testing between substations with different signal specifications and / or different secondary devices using the standard signal, which can improve the test collaboration between substations with different signal specifications and / or secondary devices. In addition, using the standard signal can ignore the signal differences between different signal specifications, improve the interoperability between substations with different signal specifications and / or secondary devices, and thus improve the test efficiency.

[0012] Furthermore, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the method for testing secondary devices based on a generative large model as provided above is implemented.

[0013] A secondary equipment test method, device and equipment based on a generative large model provided by an embodiment of the present invention have the following beneficial effects compared with the prior art: The present invention obtains device test signals with different signal specifications between different substations and / or different secondary equipment in a substation; through a pre-trained test case generation model, test cases corresponding to test requirements are generated according to the standard signals corresponding to the device test signals; the test cases are used to perform collaborative testing on different substations and / or different secondary equipment in a substation, and test results corresponding to the test requirements are generated. By using a standard signal model to identify the common attributes of device signals with different signal specifications, a standard signal representing a uniformly followed signal specification is formed; furthermore, test cases for collaborative testing between substations with different signal specifications and / or different secondary equipment are generated using the standard signal, which can improve the test collaboration between substations and / or secondary equipment with different signal specifications. In addition, using the standard signal can ignore the signal differences between different signal specifications, improve the interoperability between substations and / or secondary equipment with different signal specifications, and thus improve the test efficiency. Description of the Drawings

[0014] Figure 1 is a schematic flowchart of an embodiment of a secondary equipment test method based on a generative large model provided by the present invention; Figure 2 is a schematic structural diagram of an embodiment of a secondary equipment test device based on a generative large model provided by the present invention; Reference numerals: 201, test requirement acquisition module; 202, test case generation module; 203, collaborative test module. Detailed Embodiments

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] The present invention obtains device test signals with different signal specifications between different substations and / or different secondary devices in a substation; through a pre-trained test case generation model, test cases corresponding to test requirements are generated according to the standard signals corresponding to the device test signals; the test cases are used to co-test different substations and / or different secondary devices in a substation, and test results corresponding to the test requirements are generated. By using a standard signal model to identify the common attributes of device signals with different signal specifications, a standard signal representing the signal specifications uniformly followed is formed; furthermore, on the basis of obtaining device test signals with different signal specifications between different substations and / or different secondary devices in a substation, test cases for co-testing between substations and / or different secondary devices with different signal specifications are generated using the standard signal, which can improve the test coordination between substations and / or secondary devices with different signal specifications. In addition, using the standard signal can ignore the signal differences between different signal specifications, improve the interoperability between substations and / or secondary devices with different signal specifications, and thus improve the test efficiency.

[0017] See Figure 1 , Figure 1 FIG. is a schematic flowchart of an embodiment of a method for testing secondary equipment based on a generative large model provided by the present invention. As Figure 1 shown, the method includes S101 - S103, which are specifically as follows: S101: Obtain the test requirements of the secondary equipment in the substation; the test requirements at least include device test signals with different signal specifications between different substations and / or different secondary devices in the substation; S102: Through a pre-trained test case generation model, generate test cases corresponding to the test requirements according to the standard signals corresponding to the device test signals; the standard signal is obtained by classifying the device signals of the secondary equipment in the substation based on a standard signal model, and is used to represent the signal specifications uniformly followed between some or all different substations and / or between some or all different secondary devices in the substation; the standard signal model is configured to identify the common attributes between device signals with different signal specifications to form the standard signal; S103: Use the test cases to co-test the different substations and / or different secondary devices in the substation to generate test results corresponding to the test requirements.

[0018] In order to achieve co-testing of different substations and / or different secondary devices in a substation, first, the test requirements of the secondary equipment in the substation are obtained. Among them, the test requirements at least include device test signals with different signal specifications between different substations and / or different secondary devices in the substation. That is, the device test signals of different substations have different signal specifications, and the device test signals between different secondary devices in the substation also have different signal specifications.

[0019] Furthermore, a standard signal corresponding to the device test signal is determined through a pre-trained test case generation model. Among them, the standard signal is obtained by classifying the device signals of the secondary substation equipment based on a standard signal model, and is used to represent the signal specifications uniformly followed between some or all different substations and / or between some or all different secondary equipment in the substation. The standard signal model is configured to identify the common attributes between device signals with different signal specifications to form the standard signal. That is, the standard signal corresponding to the device test signals with different signal specifications between different substations can be identified through the standard signal model, and / or the standard signal corresponding to the device test signals with different signal specifications between some or all different secondary equipment in the substation can be identified.

[0020] Then, test cases corresponding to the test requirements can be generated according to the standard signal corresponding to the device test signal. Finally, the test cases are used to perform collaborative testing on the different substations and / or different secondary equipment in the substation to generate test results.

[0021] The present invention identifies the common attributes of device signals with different signal specifications through a standard signal model to form a standard signal representing the uniformly followed signal specifications; furthermore, test cases for collaborative testing between substations and / or different secondary equipment with different signal specifications are generated using the standard signal, which can improve the test collaboration between substations and / or secondary equipment with different signal specifications. In addition, using the standard signal can ignore the signal differences between different signal specifications, improve the interoperability between substations and / or secondary equipment with different signal specifications, and thus improve the test efficiency.

[0022] In a further embodiment, the construction process of the standard signal model includes: Obtain a first set of device signals of the secondary substation equipment; the first set of device signals includes standard signals corresponding to device signals with different signal specifications between different substations and / or different secondary equipment in the substation and data labels corresponding to each of the standard signals; Perform preprocessing on the first set of device signals; the preprocessing includes at least data cleaning; Divide the preprocessed first set of device signals into at least a training set and a validation set; Use the training set to train the constructed standard signal model to identify the common attributes of device signals with different signal specifications between different substations and / or different secondary equipment in the substation to form the standard signal, and use the validation set to verify the standard signal model until the iteration termination condition is met to obtain a trained standard signal model.

[0023] In this embodiment, through process standardization and intelligent business modeling, the entire process operation of testing secondary substation equipment (such as relay protection devices) is unified. Based on artificial intelligence and big data technologies, by constructing a standard signal model, process visualization and operation standardization are achieved. The model is dynamically optimized through simulation verification and adaptive adjustment mechanisms to improve the consistency and efficiency of testing, while meeting the testing requirements of multiple scenarios and multiple devices. The continuously optimized standard signal model helps reduce operation errors and improve the quality and efficiency of tests.

[0024] The standard signal model of secondary substation equipment needs to reflect in detail various states, actions, and communication signals of the device under different operating conditions. According to different stages of equipment operation and types of equipment failures, the standard signal model should cover the following main signals: (1) Signals in normal operating state System normal state signals: Such as "normal operation signal", indicating that the device is in a fault-free working state and the protection device has not operated. Working current and voltage signals: Monitor the current and voltage values during power grid operation to ensure they are within the normal working range.

[0025] (2) Protection action signals Tripping signal: When the relay protection device identifies a fault and triggers a tripping command, the tripping signal is activated, and the alarm system will give an alarm in a timely manner. Alarm signal: Such as the alarm state triggered by protection signals such as overcurrent, overvoltage, and undercurrent. Protection release signal: After the fault is eliminated, the relay protection device releases the trip and resumes operation, sending a release signal.

[0026] (3) Abnormal / fault state signals Fault location signal: For example, in the case of a short circuit fault or a ground fault, the system will generate a location signal to help locate the fault source. Alarm status signal: For example, in abnormal operating states such as overload and undervoltage, the system will trigger an alarm signal. Maintenance status signal: When the device enters the maintenance or test mode, the system generates a maintenance signal to block normal protection actions.

[0027] (4) Communication messages GOOSE (Generic Object Oriented Substation Event) signals: Used for real-time data exchange between relay protection devices, especially when coordinating actions between relay protection devices and switchgear. SV (Sampled Value) signals: Used to transmit digitized sampled values in the power system, such as numerical signals of current and voltage. MMS (Multimedia Message Service) signals: Used for remote control and data acquisition between devices, especially for remote configuration and management between the monitoring system and relay protection devices.

[0028] The standard signal model is used to uniformly define the signal format and content of relay protection devices and ensure signal interoperability between different devices and manufacturers. Artificial intelligence technologies, especially machine learning and / or deep learning algorithms, can be introduced to automatically learn the model files of secondary equipment in different substations by machines and automatically integrate the key information in the model files to achieve automated standard signal modeling.

[0029] Among them, the construction of the standard signal model mainly includes: signal acquisition, signal annotation, signal preprocessing, signal feature extraction, model training, model evaluation and optimization, standard signal model generation, etc.

[0030] Signal acquisition: Obtain the first device signal set of secondary equipment in the substation; the first device signal set includes standard signals corresponding to device signals with different signal specifications between different substations and / or different secondary equipment in the substation and data tags corresponding to each standard signal. First, a large amount of signal data needs to be collected from various relay protection devices, and these data may include the working status, action signals, communication signals, etc. of different secondary equipment.

[0031] Specifically, a distributed sensor network and data acquisition terminals can be used to comprehensively collect the first device signal set from secondary equipment (relay protection devices) in the substation. The first device signal set covers devices of different manufacturers, models, and operating years, as well as signals under various working conditions such as normal operation, fault, and maintenance, including working current, voltage signals, protection action signals (tripping, alarming, etc.), abnormal / fault status signals (fault location, alarm, etc.), and communication messages (GOOSE, SV, MMS signals). The multi-source data fusion technology is adopted to integrate data from different communication interfaces and protocols to ensure the integrity and consistency of the data. In addition, the first device signal set includes data tags corresponding to the standard signals of device signals with different signal specifications between different substations and / or different secondary equipment in the substation.

[0032] Exemplarily, in the first device signal set, the device signals of substation A1 and substation A2 have different signal specifications, and the device signals with different signal specifications of substation A1 and substation A2 have data tags corresponding to the standard signal A. Or, the device signals of secondary equipment a11 and secondary equipment a12 in substation A1 have different signal specifications, and the secondary equipment signals with different signal specifications of secondary equipment a11 and secondary equipment a12 have data tags corresponding to the standard signal a.

[0033] Signal annotation: Organize professionals to annotate the device signals in the first device signal set collected. According to the device operating status and signal characteristics, annotate the category of the standard signal to which the data belongs. For example, a signal during normal operation is annotated as "normal", and a tripping signal is annotated as "protection action - trip", etc. For complex fault signals, details such as the fault type and the fault occurrence time are annotated to provide accurate label data for subsequent model training.

[0034] Signal preprocessing (data cleaning): Preprocess the first device signal set; the preprocessing includes at least data cleaning. Through data cleaning and preprocessing by artificial intelligence technology, effective signals are identified and extracted, and redundant data is removed.

[0035] Specifically, an outlier detection algorithm based on statistics, such as the 3σ criterion, can be used to identify and remove abnormal device signals in the first device signal set that significantly deviate from the normal range. For the problem of missing data, according to the time series characteristics of the data, linear interpolation, polynomial interpolation, or the K-nearest neighbor interpolation algorithm based on machine learning is used for filling. At the same time, data smoothing algorithms, such as the moving average method and the median filtering method, are used to remove noise interference in the data and improve the data quality. Thus, the first device signal set after preprocessing (data cleaning) is obtained. The data is standardized to map data with different dimensions to the same numerical interval, such as [0,1] or [-1,1]. Common standardization methods include min-max standardization, Z-score standardization, etc., to improve the model training effect.

[0036] Signal feature extraction: Extract and select features from the device signals in the first device signal set after cleaning. For numerical device signals, calculate statistical features, such as mean, variance, standard deviation, maximum value, minimum value, etc.; for time-series device signals, methods such as Fourier transform and wavelet transform are used to extract frequency-domain features. The correlation analysis algorithm, such as the Pearson correlation coefficient, is used to screen out features closely related to signal modeling, reduce the data dimension, and improve the model training efficiency. For communication message data, features such as message length, message sending frequency, and specific field values are extracted. In addition, some features with physical meanings, such as power factor and impedance, can be constructed by combining the professional knowledge of the power system.

[0037] In addition, a feature selection algorithm is used to screen out the feature subset that contributes the most to model classification from the numerous extracted features. Common feature selection methods include the filter method (such as chi-square test, information gain), the wrapper method (such as recursive feature elimination), and the embedding method (such as feature selection based on decision trees). Through feature selection, not only can the data dimension be reduced, the model training time be reduced, but also the influence of redundant information between features on the model performance can be avoided, and the generalization ability of the model can be improved.

[0038] Model training: The preprocessed first device signal set is at least divided into a training set and a validation set; the constructed standard signal model is trained using the training set to identify the common attributes of device signals with different signal specifications between different substations and / or different secondary devices in a substation to form standard signals, and the validation set is used to validate the standard signal model until the iteration termination condition is met to obtain a trained standard signal model. Machine learning algorithms (such as clustering algorithms, decision trees, etc.) can help identify signal similarities between different secondary devices and classify and model the signals. Deep learning methods (such as convolutional neural networks, recurrent neural networks, etc.) can be used for more complex signal modeling. Through training on a large amount of historical signal data, a standard signal model for classification and recognition and standard signals that meet the standards can be automatically generated.

[0039] (I) Model training based on machine learning: Decision tree model: A standard signal model is constructed using decision tree algorithms (such as C4.5, CART). According to indicators such as information gain or Gini index of features, the optimal splitting attribute is selected, and the data set is gradually divided into different sub-nodes to form a decision tree structure. During the construction of the standard signal model, by setting appropriate pruning strategies (such as pre-pruning, post-pruning), overfitting of the decision tree is prevented. The decision tree model is trained using the training data set, and the parameters of the model, such as the maximum depth, minimum number of samples, etc., are adjusted to obtain the best classification performance.

[0040] Random forest model: A standard signal model of a random forest is constructed based on the decision tree model. Multiple sample subsets are randomly drawn from the training data set with replacement, and a decision tree is constructed for each sample subset, finally forming a forest containing multiple decision trees. During prediction, the prediction results of multiple decision trees are integrated, and the final classification result is determined by voting or averaging, etc. By introducing randomness, the random forest enhances the generalization ability of the model and reduces the risk of overfitting. During the training process, the parameters of the random forest, such as the number of trees, the size of the feature subset, etc., are adjusted to optimize the model performance.

[0041] Support Vector Machine (SVM) model: For linearly separable and linearly inseparable device signals, linear SVM and non-linear SVM (such as using kernel functions, such as radial basis function RBF, polynomial kernel function, etc.) are respectively used for classification to construct a standard signal model. By finding an optimal classification hyperplane and maximizing the margin between two types of data, the classification of the data is achieved. The SVM model is trained using the training data, and the parameters of the model, such as the penalty parameter C, kernel function parameters, etc., are adjusted to improve the classification accuracy of the standard signal model.

[0042] (II) Model training based on deep learning: Convolutional Neural Networks (CNN) model: For device signals with certain structures and local features (such as signal data converted into image form after preprocessing), the CNN model is used for classification. Designing the CNN model structure usually includes convolutional layers, pooling layers, and fully connected layers. The convolutional layer slides the convolutional kernel over the data for convolution to extract the local features of the data; the pooling layer is used to reduce the data dimension, reduce the computational amount, and at the same time retain important features; the fully connected layer comprehensively classifies the extracted features. The CNN model is trained using the backpropagation algorithm and stochastic gradient descent optimizers (such as Adagrad, Adadelta, Adam, etc.) to adjust the weights and biases of the model so that the model can accurately identify and classify the signal data.

[0043] Recurrent Neural Network (RNN) and its variant models: Considering the temporal characteristics of device signal data, RNN or its variants (such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU)) are used for modeling. Taking LSTM as an example, through the collaborative action of the input gate, forget gate, and output gate, it effectively processes the long-term dependence relationships in the signal. The temporal data of the device signal is input into the LSTM network at each time step. The network learns the changing features of the signal at each time step and passes the information to the next time step. Similarly, the LSTM model is trained using the backpropagation algorithm and a suitable optimizer so that it can accurately predict and classify the temporal changes of the signal. During the training process, the parameters of the LSTM model, such as the number of hidden layer units, the number of layers, etc., are adjusted to optimize the model performance.

[0044] The above standard signal models obtained through machine learning and / or deep learning training can identify the common attributes of device signals with different signal specifications among different substations and / or different secondary devices in a substation during the training process, thereby forming standard signals corresponding to device signals of different model specifications. The common attributes of device signals with different signal specifications can be, for example, signal content, signal format, and signal representation form, etc.

[0045] Model evaluation indicator selection: Use a variety of evaluation indicators to comprehensively evaluate the trained model. For the standard signal model of classification, commonly used evaluation indicators include accuracy, recall, F1-Score, precision, etc. Accuracy measures the proportion of samples correctly classified by the model; recall reflects the model's ability to identify positive samples; F1-Score takes into account accuracy and recall to more comprehensively evaluate model performance; precision indicates the proportion of samples that are actually positive among those predicted by the model. In addition, the confusion matrix can be used to intuitively display the classification of the model in each category.

[0046] Model optimization strategy: Optimize the standard signal model based on the evaluation results. If the standard signal model is overfitting, regularization techniques (such as L1 and L2 regularization, Dropout, etc.) are used to constrain the complexity of the model to prevent the model from over-learning the noise and details in the training data. For underfitting standard signal models, increase the complexity of the model, such as increasing the number of layers or neurons in the neural network, adjusting the model's hyperparameters (such as learning rate, number of iterations, etc.), or try to use a more complex model structure. At the same time, cross-validation techniques (such as K-fold cross-validation) are used to divide the training data set into multiple subsets, and one of the subsets is used as the validation set in turn, and the remaining subsets are used as the training set. The standard signal model is trained multiple times and the performance is evaluated to ensure that the evaluation results of the model are reliable and stable. By continuously optimizing the standard signal model, the accuracy and generalization ability of the standard signal model for signal classification modeling are improved.

[0047] The iteration termination condition may include one or more of the aforementioned accuracy, recall, F1 value and precision meeting a corresponding indicator threshold, such as an accuracy threshold or a precision threshold. The iteration termination condition may also include the number of training iterations of the standard signal model meeting a preset iteration number threshold.

[0048] Standard signal model generation: Encapsulate and serialize the model trained based on machine learning and / or deep learning to generate a standard signal model. The standard signal model should contain the structural information of the model (such as the number of layers and nodes of the neural network, the structure of the decision tree, etc.), the trained model parameters (weights, biases, etc.), and the metadata of the model (such as model name, training time, evaluation indicators, etc.). The format of the standard signal model can choose a common serialization format, such as Pickle (for machine learning models in Python), HDF5 (for deep learning models), etc., to facilitate the storage, transmission and subsequent use of the model. The generated standard signal model can be used as a standardized equipment signal classification modeling tool, which can be applied to the testing and diagnosis of substation secondary equipment to achieve accurate classification and identification of signals under different equipment and working conditions.

[0049] In an embodiment of the present invention, the standard signal model can be constructed by collecting the first device signal set of the secondary devices in a substation and training using machine learning and / or deep learning, so as to identify the common attributes of device signals with different signal specifications to form standard signals, quickly identify the device signals of the secondary devices in the substation, classify them to form corresponding standard signals, improve the interoperability and test collaboration between device signals with different signal specifications, and improve the test efficiency.

[0050] In a further embodiment, the method further includes: Fusing at least two standard signal models constructed and trained by machine learning and / or deep learning to obtain a fused standard signal model.

[0051] Specifically, model fusion techniques are used to fuse multiple different standard signal models (such as decision trees, random forests, SVMs, CNNs, LSTMs, etc.) to improve the overall performance of the standard signal model. Common model fusion methods include voting methods (hard voting, soft voting), averaging methods, stacking methods, etc. The voting method is to let multiple models vote and select the category with the most votes as the final prediction result; the averaging method is applicable to regression models and obtains the final predicted value by averaging the prediction results of multiple standard signal models; the stacking method is to use a meta-model to fuse the prediction results of multiple base models. In practical applications, according to the performance and characteristics of different standard signal models, a suitable fusion method is selected to improve the accuracy and robustness of the standard signal model.

[0052] In this embodiment, multiple standard signal models are fused to obtain a fused standard signal model, which can improve the performance and robustness of the standard signal model.

[0053] In a further embodiment, the categories of the standard signal model include: the standard signal model in the device operation stage, the standard signal model of the device failure type, the standard signal model of the device signal nature, and the standard signal model of the device type; In S102, generating the test cases corresponding to the test requirements according to the standard signals corresponding to the device test signals includes: Generating the test cases corresponding to the test requirements according to at least one category of standard signals corresponding to the device test signals.

[0054] When constructing the standard signal model, the present invention can hierarchically classify the standard signal model from four key factors and standards: device operation stage, failure type, signal nature, and device type, to improve the systematicness, accuracy, and adaptability of the model.

[0055] Based on the equipment operation stage: In the prior art, the different stages of equipment operation are often not fully considered. In the present invention, the equipment operation stage is taken as a key factor, and the standard signal model is further divided into standard signal models for stages such as normal operation, protection action, abnormality / fault, maintenance or test. In the normal operation stage, the normal state signals of the system, working current and voltage signals are mainly monitored to ensure that the equipment is free of faults and the grid operation parameters are normal; in the protection action stage, signals such as tripping, alarm, and protection release are focused on to clarify the changes in the equipment state during the occurrence and handling of faults; in the abnormality / fault stage, equipment problems are detected and located in a timely manner through signals such as fault location and alarm status; in the maintenance or test stage, the maintenance status signal is used to shield normal protection actions to ensure the safe progress of maintenance and test work. This hierarchical method systematically covers the signal monitoring requirements of the entire life cycle of the equipment, provides comprehensive data support for testing, and is different from the dispersed monitoring mode of the prior art.

[0056] According to the equipment fault type: Hierarchical classification of the standard signal model according to the fault type is another major feature of the present invention. Fault location signals generated by different fault types such as short-circuit faults and grounding faults, as well as alarm signals caused by abnormal states such as overloading and undervoltage, are classified and managed. In this way, during the test, the signal changes of the equipment under different fault types can be accurately identified, and the equipment fault situation can be quickly judged. In the prior art, such detailed classification monitoring of fault types may be lacking, making it difficult to accurately locate problems in complex fault scenarios.

[0057] According to the nature of the equipment signals: In the present invention, according to the signal nature, the signals are divided into categories such as system normal state signals, protection action signals, abnormality / fault state signals, and communication messages (GOOSE signals, SV signals, MMS signals) to form standard signal models of different signal natures. Signals of different natures perform different functions in the power system. Hierarchical classification of signals of different natures helps to clearly define the uses and scopes of action of the signals. For example, GOOSE signals are used for real-time data exchange and coordinated actions between relay protection devices, SV signals transmit digital sampling values, and MMS signals enable remote control and data acquisition between devices. Through this hierarchical classification, for standard signal models of different signal natures during testing, targeted monitoring and analysis can be carried out for signals of different natures. In the prior art, signals may not be classified and managed so systematically, resulting in chaotic signal analysis during the test process.

[0058] Based on device type: When constructing the standard signal model, signals are modeled in combination with different device types (such as relay protection devices, measurement and control devices, intelligent terminals, merging units, etc.). Different devices have different functions in the power system, and the generated signals also have their own characteristics. For example, relay protection devices focus on protection action signals, while measurement and control devices focus on teleinformation and telemetry signals. The present invention realizes precise management of signals of different devices in this way to form standard signal models corresponding to different device types, enabling the test templates to better adapt to different devices and improving the pertinence and accuracy of testing, which is an aspect not fully emphasized in the prior art. The prior art may adopt a unified test mode, ignoring the differences between devices.

[0059] The present invention classifies and grades the standard signal model according to key factors such as the device operation stage, device fault type, device signal nature, and device type, improving the systematicness and adaptability of the standard signal model.

[0060] In a further embodiment, in step S102, generating the test cases corresponding to the test requirements according to the standard signals corresponding to the device test signals by the pre-trained test case generation model includes: Extracting the device test signals included in the test requirements by the test case generation model; Determining the standard signals corresponding to the device test signals based on the signal mapping model; the signal mapping model at least includes the mapping rules between the device test signals and the standard signals; Generating the initial test cases corresponding to the test requirements in a template configuration manner by the test case generation model according to the standard signals; the initial test cases at least include the standard signals corresponding to the device test signals, the test process, and the test results; Instantiating the standard signals in the initial test cases into the device test signals to form the instantiated test cases.

[0061] When generating test cases, first, the device test signals included in the test requirements are extracted by the test case generation model. Furthermore, the standard signals corresponding to the device test signals are determined based on the signal mapping model. Among them, the signal mapping model at least includes the mapping rules between the device test signals and the standard signals, and the standard signals corresponding to the device test signals can be obtained through the signal mapping model based on the mapping rules. Then, after obtaining the standard signals, the initial test cases corresponding to the test requirements are generated in a template configuration manner. Finally, the initial test cases are instantiated, and the standard signals are instantiated into the device test signals, and then the instantiated test cases are obtained.

[0062] In the embodiments of the present invention, determining the standard signal corresponding to the device test signal through the signal mapping model can improve the efficiency of determining the standard signal; furthermore, generating test cases in the form of template configuration based on the standard signal can improve the flexibility and scalability of the test cases.

[0063] In a further embodiment, generating the initial test cases corresponding to the test requirements in the form of template configuration according to the standard signal through the test case generation model includes: Through the test case generation model, generating the initial test cases corresponding to the test requirements in the form of template configuration according to the standard signals included in at least one of the prompt type, judgment type, pressure plate control type, and whole-group output fault type.

[0064] Through modular design, the present invention subdivides the test process in the test cases into combinable small modules to achieve rapid adaptation for multiple devices and multiple scenarios. Using an automated template generation tool, test scripts are generated based on standard signals, and at the same time, dynamic loading and adjustment of templates are supported to adapt to actual changes. The template system integrates the function of automatically generating reports. Through real-time data filling and version control, rapid output and unified management of test reports are realized, significantly improving the template configuration efficiency and flexibility.

[0065] Based on the standard signals defined by the standard signal model, initial test templates are edited to ensure that the generated initial test templates have the characteristics of flexible editability and generality. Initial test templates including one-key testing of protection monomers, one-key testing of the whole-group loop, etc. can be generated through template editing.

[0066] The so-called template configuration technology refers to generating test case templates through flexible editing, rather than solidifying test cases through test instruments, ensuring that the test templates can be applied to different test purposes and test scenarios, and can also be made suitable for differentiated requirements by simply editing the templates. Through complex global information modeling and test case configuration, a whole-station system-level joint automatic debugging solution including single-function debugging of relay protection devices, whole-group loop debugging of relay protection devices, remote control point matching, oscillograph acceptance, etc. can be formed.

[0067] One-key configuration and flexible adjustment: The generation of test templates can be automatically completed through the "one-key" configuration function. The tester only needs to provide the necessary test requirement parameters, and the system can generate the corresponding test templates. At the same time, the tester can adjust the template configuration through simple operations to meet the differentiated requirements on-site.

[0068] The specific method is to sort out the equipment test signals that need to be used according to the possible test items and test purposes in the test requirements, including MMS signals and GOOSE signals of various types of relay protection, MMS signals and GOOSE signals of measurement and control, GOOSE signals of smart terminals, GOOSE signals and SV signals of merging units, etc., and uniformly model them according to the equipment form. The above signal collection forms a standard signal database with the equipment as the object.

[0069] According to the test items or test tasks in the test requirements, sort out the test points. For example, based on a specific set of test reports, analyze the prerequisites that need to be met for each specific test sub-item, the signal state that needs to be applied, and the action results that need to be checked, and then complete the configuration of the corresponding initial test case through the standard signal in the standard signal database. The configuration of the initial test case supports the configuration of test sub-items in a sequential manner, and combines them into specific test tasks, and finally forms a complete initial test case. The initial test case is generated by template configuration based on the standard signal database. It is not based on specific SCD file information or fixed algorithms, which ensures the scalability and versatility of the initial test case.

[0070] The process of template configuration is to generate test templates (test cases) through four basic classes, including prompt class, judgment class, pressure plate control class and whole group output fault class. By performing logical operations on the state or value of MMS, GOOSE and other signals, the judgment results are formed, including: The steady-state judgment of remote signal requires that the state of a remote signal can be judged to be 1 or 0; The transient judgment of remote signal requires the ability to judge whether the state of a remote signal changes from 1 to 0 or from 0 to 1; The judgment of remote signaling requires that the state of some remote signals can be judged to be a steady state or transient state that meets the above two requirements; The relative time judgment of the telesignal change signal requires the ability to judge the result of the comparison between the relative time of the state change of a telesignal and the set time; The size comparison and judgment of telemetry requires the ability to determine the result of the comparison between the size of a telemetry signal and the set value.

[0071] The prompt class prompts the tester to participate in the operation and inspection of the external control conditions and inspection conditions required by the test project by sending prompt information through the human-computer interaction prompt window, so as to achieve the purpose of the overall debugging task requiring operation and inspection. The prompt class can not only solve the problem that some control and judgment operations cannot be implemented by the program and can only be completed with the participation of the tester, but also solve the situation that interruptions may occur in the debugging process. It is necessary to set the prompt class to prompt the tester to check before continuing the subsequent debugging work.

[0072] The pressure plate control category is a category in which the test system automatically completes the operation of the pressure plate and control word to avoid interference during the test. Since the same fault output may cause different protection components to operate, resulting in interference in the test results, it is necessary to add the control operation of the pressure plate control word between different test items during the normal one-button automatic test process, and only activate the target protection function and exit other protection functions to ensure the accuracy of the test items.

[0073] The whole group output fault class is a project that requires the test system to output sampled value SV or analog fault for a specific test task to drive the target protection, measurement and control equipment to act. The whole group output fault is preset for a specific debugging object. For example, line protection should at least include core protection function modules such as differential protection test, distance protection test, and zero-sequence overcurrent protection test. The test system can accurately simulate the fault simulation required for differential action, distance action, and zero-sequence overcurrent action by obtaining the fixed value information of the protection online and cooperating with the analog channel information of the instantiation mapping. It can also achieve normal operation condition simulation, fault simulation, and post-fault condition simulation by setting parameters such as pre-fault parameters, fault parameters, and post-fault parameters, thereby achieving the purpose of simulating specific fault types. The whole group output fault class also includes basic state sequence functions, which is convenient for testers to perform whole group tests according to the state sequence calculated by themselves during configuration. The flexible state sequence function can also be generated into fixed test cases through configuration, which greatly enriches the method of system-level whole group debugging.

[0074] The template configuration process is actually the process of configuring the standard signals in the standard signal database through the above four basic classes to form a specific preset configuration logic. This configuration (initial test case) includes the necessary content in the system testing process such as information prompts, condition checks, signal triggers and result judgments.

[0075] In a further embodiment, instantiating the standard signal in the initial test case into the device test signal to form an instantiated test case includes: Determine a standard signal corresponding to a device signal matching the device test signal from a pre-configured dictionary index using a similarity measurement method; the dictionary index at least includes a device signal, a standard signal, and a mapping relationship between the device signal and the standard signal; The standard signal corresponding to the device signal matching the device test signal in the initial test case is instantiated as the device test signal to form an instantiated test case.

[0076] After using the standard signal model to classify device signals into standard signals, the next step is to instantiate and map these standard signals, that is, to map the abstract standard signals to the device test signals of specific engineering projects.

[0077] In a further embodiment, the determining of the standard signal corresponding to the device signal that matches the device test signal from a pre-configured dictionary index by using a similarity measurement method includes: Using a similarity measurement method based on a corpus and / or a knowledge base, determining candidate device signals from the dictionary index whose similarity to the device test signal exceeds a preset similarity threshold; Determining the device signal with the highest similarity to the device test signal among the candidate device signals, and obtaining, from the dictionary index, the standard signal corresponding to the device signal with the highest similarity as the standard signal corresponding to the device signal that matches the device test signal.

[0078] The mapping process is a matching process of text information, which can also be understood as a semantic similarity matching. Semantic similarity calculation mainly includes two categories, knowledge-based and corpus-based. The knowledge-based similarity calculation method quantifies the semantic association degree between two texts by using the information obtained from a knowledge base. WordNet is a method that solves the text similarity calculation problem through a dictionary. With the help of the WordNet idea, the automatic mapping of information can be realized by directly establishing the dictionary index of signals and the corresponding dictionary interpretation relationship.

[0079] The rules of the dictionary index can be defined based on the promulgated protection standardization specifications. For normalized device test signals, by combining simple morpheme expressions and weighting algorithms for each morpheme weight, the object with the highest similarity in the global scope is calculated as the associated mapping object; for non-normalized device test signals, by strengthening the unique feature information of the model number, parameters, and description, an iterative dictionary index can be established to solve the problem of automatic signal association mapping.

[0080] The establishment of the dictionary index mainly includes: index structure design, index initialization, index update and maintenance, etc.; Index structure design: Design an iterative dictionary index structure for storing the key features of signals and the corresponding mapping information (such as the corresponding standard signals, mapping rules, etc.). This index structure can be implemented using data structures such as hash tables, tree structures (such as binary search trees, B-trees), etc. Hash tables have the advantage of fast searching and are suitable for fast retrieval of large-scale data; tree structures are convenient for range queries and sorting operations. In the index, the key features of the signal are used as keys, and the mapping information of the signal is used as values, that is, they are stored in the form of key-value pairs.

[0081] Index Initialization and Population: Based on the key features of the extracted device signals, initialize the dictionary index and populate the relevant information of the signals into the index. During the population process, for each signal, calculate the hash value of its key features or its position in the tree structure, and store the mapping information of the signal at the corresponding position. For signals with similar features, an association relationship can be established in the index for subsequent quick search and matching.

[0082] Index Update and Maintenance: As new device signals continuously appear or signal features change, update and maintain the dictionary index in a timely manner. When a new device signal is added, calculate its key features and insert it into the appropriate position in the dictionary index. If the features of the device signal change, such as signal name modification, data type change, etc., update the relevant information in the index accordingly. Regularly optimize the index, such as cleaning invalid index entries and adjusting the index structure to improve query efficiency.

[0083] The signal mapping of the dictionary index mainly includes: multi-source data acquisition, data cleaning and denoising, signal classification and marking, feature extraction and analysis, signal mapping, mapping optimization, etc.

[0084] Multi-source Data Acquisition: Widely collect signal data from various relevant devices, systems, and document materials. This includes not only various signals generated in real time during device operation, such as protection action signals, alarm signals, measurement signals, etc., but also signal-related information mentioned in the device's technical manuals, design documents, maintenance records, etc. Use data acquisition tools and interfaces to ensure comprehensive acquisition of signal data from different sources and in different formats, providing sufficient materials for subsequent processing.

[0085] Data Cleaning and Denoising: Clean and denoise the collected data. Since the signal data sources are complex, it may contain a large amount of noise, incorrect data, and duplicate information. Through data cleaning algorithms, identify and eliminate obviously incorrect or abnormal data, such as data with signal values outside the reasonable range and incorrect data formats. At the same time, remove duplicate signal records, reduce data redundancy, improve data quality, and lay a foundation for accurate signal mapping.

[0086] Signal Classification and Marking: Classify the cleaned data according to the nature, function, and source of the signals. For example, classify the signals into major categories such as protection signals, measurement signals, control signals, etc., and further subdivide each major category. For example, protection signals can be further divided into overcurrent protection signals, differential protection signals, etc. Mark each classified signal and assign it a clear category identifier for subsequent quick identification and processing.

[0087] Feature Extraction and Analysis: Key feature identification: Identify the key features of the signal. These features include the signal name, description, data type, occurrence frequency, and association relationships with other signals. Important information such as device type and signal function may be included in the signal name and description; the data type determines the value range and processing method of the signal; the occurrence frequency can reflect the importance and stability of the signal; and the association relationships with other signals help to understand the role and interaction method of the signal in the system.

[0088] Feature quantization and encoding: Quantize and encode the identified key features so that they can be processed by computer algorithms. For text information such as signal names and descriptions, natural language processing techniques such as word vector models (e.g., Word2Vec, GloVe) can be used to convert the text into a numerical vector representation. For numerical features such as data type and occurrence frequency, perform normalization processing to map them to a specific numerical interval, such as [0,1], for easy comparison and analysis.

[0089] Selection of similarity measurement method: Determine the method for measuring the similarity of signal features. Commonly used similarity measurement methods include Euclidean distance, cosine similarity, and edit distance. Euclidean distance is used to measure the distance between numerical vectors, cosine similarity is used to measure the direction similarity between vectors, and edit distance is applicable to comparing the differences between text strings. According to the type and characteristics of the signal features, select an appropriate similarity measurement method to accurately calculate the similarity degree between signals.

[0090] Signal mapping: Processing of signals to be mapped: For the device test signals to be mapped, first perform feature extraction and quantization processing to obtain their key feature vectors. Then, according to the selected similarity measurement method, calculate the similarity score between this signal and the existing device signals in the dictionary index.

[0091] Screening of mapping objects: According to the similarity score, screen out the device signals with relatively high similarity to the device test signals to be mapped from the dictionary index as potential mapping objects. A similarity threshold can be set, and only the signals with similarity scores exceeding the threshold are regarded as potential mapping objects. Sort the screened potential mapping objects according to the similarity scores, and select several signals with the highest scores for further analysis.

[0092] Mapping Decision and Verification: Select the final mapping object from potential mapping objects according to certain decision rules. The decision rules can comprehensively consider factors such as similarity scores, the correlation relationship of signals, and historical mapping records. For example, if a potential mapping object has not only a high similarity with the signal to be mapped but also is frequently associated in historical mappings, then the probability of it being selected as the final mapping object is higher. Or directly use the candidate device signal with the highest similarity as the final mapping object. After determining the mapping object, verify the mapping result. It can be verified through actual system tests, expert reviews, etc. to check whether the mapping is accurate and reasonable. If the verification passes, the signal mapping is completed; if the verification fails, reselect the mapping object or adjust the mapping strategy.

[0093] Mapping Optimization: Mapping Result Evaluation: Conduct a comprehensive evaluation of the mapping result, including the accuracy, integrity, and consistency of the mapping. The accuracy evaluation mainly checks whether the mapped signal conforms to the actual requirements and can correctly reflect the function and meaning of the signal; the integrity evaluation focuses on whether all signals that need to be mapped have been properly processed and whether there are any omissions; the consistency evaluation ensures that the mapping rules and results between different signals are consistent to avoid contradictions and conflicts.

[0094] Feedback Adjustment Mechanism: Establish a feedback adjustment mechanism based on the evaluation results. If problems are found in the mapping result, such as mapping errors, incompleteness, or inconsistency, feedback the problems to the previous processing links, such as feature extraction, index establishment, or mapping decision-making. According to the feedback information, adjust the corresponding parameters, rules, or algorithms and remap the signals. For example, if it is found that the mapping of a certain signal is always incorrect, check whether its feature extraction is accurate and whether the similarity measurement method or mapping decision rule needs to be adjusted.

[0095] Knowledge Accumulation and Reuse: Accumulate successful mapping cases and experiences to form a knowledge database. In subsequent signal mapping processes, this knowledge can be reused to improve the efficiency and accuracy of mapping. For example, if an unnormalized signal obtains an accurate mapping result after multiple adjustments and verifications, record its relevant features, mapping rules, and verification processes. When encountering similar signals, these experiences can be directly referred to for mapping to reduce the repetitive processing process.

[0096] In a further embodiment, the construction process of the signal mapping model includes: Pre-train with general data to obtain a mapping pre-training model; the mapping pre-training model includes a machine learning model and / or a deep learning model; Obtain a second device signal set of substation secondary equipment; the second device signal set at least includes data labels representing the mapping relationship between device signals and standard signals; Preprocess the second device signal set; the preprocessing includes at least data cleaning; Use the preprocessed second device signal set to fine-tune the mapping pre-training model to obtain a trained signal mapping model.

[0097] The present invention realizes automatic mapping by means of artificial intelligence technology, which is completely different from the mapping methods relying on a large number of manual operations in the prior art. By using the natural language processing (NLP) technology in machine learning, the signal naming rules, signal types, etc. of different devices are converted into standard mapping relationships, and the mapping relationships between different devices and signals are automatically learned through training the model, reducing manual intervention. The neural network in deep learning adaptively adjusts the signal matching rules during the mapping process, trains according to historical data, and optimizes error detection and correction during the mapping process. Artificial intelligence technology plays a key role in this process, which can effectively improve the accuracy and flexibility of the mapping process.

[0098] Among them, the signal mapping model realizes signal mapping mainly including: data acquisition, mapping relationship annotation, data cleaning and preprocessing, selection of model architecture, pre-training and fine-tuning, feature extraction and model training, real-time signal mapping, mapping correction and optimization.

[0099] Data acquisition: Obtain the second device signal set of the substation secondary equipment.

[0100] Specifically, signal data is widely collected from various substation secondary equipment such as relay protection devices and measurement and control equipment. These data cover signals under various working conditions such as normal operation, fault, and maintenance of the equipment, including working current and voltage signals, protection action signals (such as tripping and alarming), abnormal / fault status signals (such as fault location and alarm), and various communication messages (GOOSE, SV, MMS signals). At the same time, detailed description information of each signal is recorded, including signal name, device type to which it belongs, signal meaning, data format, etc., to form a large-scale signal data set, that is, the second device signal set.

[0101] Mapping relationship annotation: The second device signal set includes at least data tags representing the mapping relationship between device signals and standard signals.

[0102] Specifically, organize power experts and NLP professionals to label the device signals in the collected second device signal set. The focus of labeling is to establish a mapping relationship between different device signals and standard signals. For example, label the "overcurrent trip signal" of a manufacturer's relay protection device as the "protection action-overcurrent trip" signal in the corresponding standard signal set, and record this mapping relationship in detail. For complex signals, multi-level labeling may be required, such as first labeling the major category to which the signal belongs (such as protection action signal), and then labeling the specific subcategory (such as overcurrent trip) to ensure the accuracy and consistency of the labeling.

[0103] Data cleaning preprocessing: preprocessing the second device signal set; the preprocessing at least includes data cleaning.

[0104] Specifically, data cleaning technology is used to remove noise and outliers in signal data. Typos and grammatical errors in signal description information are corrected and the data format is unified. Signal names are standardized, such as converting abbreviations and aliases into standard names. Text segmentation technology is used to split the signal description text into individual words or phrases for subsequent analysis. For example, the "busbar differential protection action signal" is segmented into "busbar", "differential protection", and "action signal". At the same time, a stop word list is used to remove meaningless function words (such as "的", "在", etc.) to reduce data redundancy and improve processing efficiency.

[0105] Select model architecture: The mapping pre-trained model includes a machine learning model and / or a deep learning model.

[0106] Specifically, according to the characteristics of the signal data and the requirements of the mapping task, select the appropriate NLP model architecture, that is, the mapping pre-trained model. For simple text classification and mapping tasks, you can consider using traditional machine learning models, such as Naive Bayes, support vector machines, etc. These models are relatively simple to calculate and fast to train. For complex semantic understanding and mapping tasks, deep learning models perform better, such as pre-trained models such as BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer) based on the Transformer architecture. Taking the BERT model as an example, it has strong semantic understanding capabilities in natural language processing tasks, can capture contextual information in the text, and more accurately understand the meaning of the signal description, thereby achieving more accurate mapping.

[0107] Feature Extraction and Model Training: For traditional machine learning models, feature engineering is required. Extract features from the signal description text, such as Bag of Words features, where each word that appears in the text is taken as a feature and its occurrence frequency is counted; TF-IDF (Term Frequency-Inverse Document Frequency) features, which not only consider the occurrence frequency of a word in the text but also the rarity of the word in the entire dataset, and can more accurately reflect the importance of the word. In addition, syntactic features, semantic features, etc. of the text can also be extracted. The extracted features are input into the machine learning model for training, and the hyperparameters of the model are adjusted (such as the smoothing parameter in the Naive Bayes model, the kernel function type and penalty parameter in the Support Vector Machine model, etc.). The optimal model parameters are selected through methods such as cross-validation to improve the accuracy and generalization ability of the model.

[0108] Pre-training and Fine-tuning: A mapping pre-training model is obtained by pre-training with general data, and the mapping pre-training model is fine-tuned using the preprocessed second device signal set to obtain a trained signal mapping model.

[0109] Specifically, if a pre-training model is selected, first, the pre-training model is pre-trained with a large amount of general data to learn the general features and semantic representations of the language, obtaining a mapping pre-training model. Then, the mapped pre-training model is fine-tuned using the labeled power device signal data, that is, the second device signal set that at least includes data labels characterizing the mapping relationship between the device signal and the standard signal. During the fine-tuning process, the device signal description text in the second device signal set is used as the input, and the corresponding standard signal annotation is used as the output. The parameters of the mapping pre-training model are adjusted through the backpropagation algorithm to make the mapping pre-training model adapt to the signal mapping task in the power field. For example, in the fine-tuning of the BERT model, the power signal description text is input into the BERT model. After being processed by multiple Transformer layers, the semantic representation of the text is obtained. Then, through a fully connected layer, the semantic representation is mapped to the standard signal, and the loss function (such as cross-entropy loss) between the prediction result and the annotation result is calculated. The model parameters are continuously adjusted through an optimizer (such as the Adam optimizer) to minimize the loss function, thereby improving the performance of the model in the signal mapping task, and finally obtaining a fine-tuned and trained signal mapping model.

[0110] Real-time signal mapping: In practical applications, when new device signal data is input, the signal description text is first preprocessed, such as operations like word segmentation and stop word removal. Then the preprocessed text is input into the trained NLP-based signal mapping model, and the signal mapping model outputs the corresponding standard signal. For example, when the "zero-sequence overcurrent alarm signal" of a certain device is input, the signal mapping model calculates and infers, and outputs the category of the standard signal "protection action - zero-sequence overcurrent alarm" in the standard signal database, realizing real-time signal mapping.

[0111] Mapping correction and optimization: Verify the mapping results output by the signal mapping model. The method of manual review can be adopted, and experts in the power field check the mapping results to ensure the accuracy of the mapping. If mapping errors are found, analyze the reasons for the errors, which may be insufficient training data of the signal mapping model, unclear signal descriptions, or limitations of the signal mapping model itself, etc. For errors caused by insufficient training data, collect more relevant signal data, re-label the mapping relationship, and retrain the signal mapping model; for the problem of unclear signal descriptions, communicate with the device manufacturer to obtain more accurate signal description information; for the limitations of the signal mapping model itself, try to adjust the architecture or parameters of the signal mapping model, or combine other technologies (such as rule engines) to correct the mapping results and improve the accuracy and reliability of the mapping.

[0112] With the development of the power system and the replacement of equipment, new signals and mapping relationships continuously emerge. Therefore, it is necessary to continuously collect new signal data, regularly retrain and update the signal mapping model. At the same time, pay attention to changes in power industry standards, timely adjust the standard signal database and mapping relationship annotation, and ensure that the signal mapping model can always accurately achieve standard mapping. In addition, continuously optimize the performance of the model, such as improving the processing speed of the model and reducing memory consumption, etc., to meet the needs of large-scale signal data processing.

[0113] Automated mapping and anomaly detection: Through the neural network in deep learning, the signal mapping model can adaptively adjust the signal matching rules during the mapping process, train based on historical data, and optimize error detection and correction during the mapping process. The neural network can learn more complex mapping relationships, automatically judge whether the signal is consistent with the target device, and perform anomaly alarm and correction when errors occur.

[0114] In a further embodiment, the method further includes: Dynamically adjusting the mapping rules of the signal mapping model through a reinforcement learning algorithm.

[0115] The reinforcement learning algorithm is adopted to make the signal mapping process more intelligent. According to the real-time data and status of the equipment at the engineering site, the signal mapping rules are dynamically adjusted to adapt to different operating environments, greatly improving the accuracy and flexibility of mapping.

[0116] Specifically, dynamically adjusting the mapping rules based on the reinforcement learning algorithm includes: defining the reinforcement learning environment, initializing the agent, the reinforcement learning training process, and dynamically adjusting the mapping rules.

[0117] I. Defining the reinforcement learning environment Definition of the state space: The state space needs to comprehensively reflect the relevant information of the current mapping task, including the signal characteristics to be mapped currently, the established mapping relationships, the historical mapping results, and the system operating status, etc. Signal characteristics cover signal names, types, the equipment to which they belong, etc.; the established mapping relationships contain the previous successful or failed mapping records; the historical mapping results record whether the mapping is accurate and the corresponding feedback; the system operating status includes factors affecting mapping decisions such as equipment load and network communication quality. These information constitute the state vector, serving as the basis for the decision-making of the reinforcement learning agent.

[0118] Definition of the action space: The action space is the set of mapping rule adjustment actions that the agent can take. It includes adding new mapping rules, such as establishing an association with the standard signal according to the new signal characteristics; modifying existing rules, adjusting signal matching conditions or target mappings; deleting unreasonable rules, removing the rules that cause incorrect mappings; adjusting the rule priorities, changing the execution order of different rules according to the actual situation to optimize the mapping decision-making process.

[0119] Design of the reward function: The reward function is the key to guiding the agent to learn the optimal strategy. A positive reward is given when the mapping is accurate. For example, when a complex signal is successfully mapped to the standard signal correctly, a corresponding reward is given according to the importance of the signal, and more important signals are rewarded more highly. A negative reward is given for incorrect mappings, and different penalty intensities are set according to the severity of the error, and more severe errors that affect the system operation are punished more severely. In addition, if the mapping process improves the system efficiency or stability, such as reducing the mapping time and lowering the system resource occupancy, an additional reward is also given, otherwise, a penalty is imposed, prompting the agent to improve the overall performance of the system while ensuring the mapping accuracy.

[0120] II. Initializing the agent Selecting the reinforcement learning algorithm: Algorithms such as the Deep Q-Network (DQN), policy gradient algorithms (such as A2C, A3C, PPO), or algorithms based on the Actor-Critic architecture (such as DDPG, TD3) can be selected. Taking DQN as an example, it approximates the Q-value function through a neural network, balancing exploration and exploitation, and is suitable for solving problems with discrete action spaces and has good performance in action selection problems such as mapping rule adjustment.

[0121] Initialize agent parameters: Construct a neural network model as the decision-making model of the agent. If DQN is used, the network structure includes an input layer, a hidden layer, and an output layer. The input layer receives the state vector, the hidden layer performs feature extraction and processing, and the output layer outputs the Q-values for each action. Randomly initialize the network weights and set hyperparameters such as the learning rate, discount factor, and size of the experience replay buffer. The learning rate controls the step size of weight updates, the discount factor determines the importance of future rewards, and the experience replay buffer is used to store and reuse historical experiences to prevent the agent from overly relying on the current experience and improve learning stability.

[0122] III. Training Process Environment interaction and experience collection: The agent selects an action from the action space according to the current state (mapping rule adjustment scheme), executes this action in the actual mapping task, observes the mapping result, and obtains the new state and reward. Record the state, action, reward, and new state in the experience replay buffer. As the training progresses, the buffer accumulates a large amount of historical experiences.

[0123] Experience replay and model update: Randomly sample a batch of experience data from the experience replay buffer and input it into the neural network for training. Calculate the Q-value estimates for taking each action in the current state and update the Q-values according to the Q-learning formula. Adjust the neural network weights through the backpropagation algorithm to make the Q-value estimates closer to the true values and improve the agent's decision-making ability.

[0124] Balance between exploration and exploitation: At the beginning of training, since the agent has limited understanding of the environment, a larger exploration probability (such as a larger value in the policy) is adopted to randomly select actions, explore different mapping rule adjustment methods, and obtain more experiences. As the training progresses, gradually reduce the exploration probability and increase the probability of using the existing experiences to select the optimal actions to improve the mapping efficiency and accuracy.

[0125] IV. Dynamically Adjust Mapping Rules Real-time mapping task execution: In the actual secondary equipment signal mapping scenario of a substation, the agent continuously monitors the changes in the state space. Once there is a new signal to be mapped, select an appropriate action according to the current state to adjust the mapping rule and complete the mapping task. For example, when it is detected that the new device signal does not match the existing rules, the agent decides to add new rules or adjust the existing rules according to the learned policy.

[0126] Performance evaluation and continuous optimization: Regularly evaluate the performance of the mapping system, and the metrics include mapping accuracy, system response time, resource utilization, etc. Judge whether it is necessary to further optimize the agent's policy according to the evaluation results. If the performance deteriorates, increase the training frequency or adjust the hyperparameters; if the performance improves, try to reduce the exploration probability to make the agent more dependent on the learned optimal policy to ensure that the mapping rules adapt to system changes and maintain an efficient and accurate mapping effect.

[0127] In a further embodiment, the process of constructing the test case generation model includes: Obtain a training and test case set for secondary equipment in a substation; the training and test case set includes training and test requirements and corresponding training and test cases; Use the training and test case set to fine-tune and train a generative pre-trained model for generating test cases to obtain the test case generation model.

[0128] To better meet the intelligent reasoning requirements during the testing process, the fine-tuning technology based on large language models (LLMs) is introduced. Through the training of domain knowledge, LLMs possess powerful natural language understanding and logical reasoning capabilities, and can automatically generate test templates that meet engineering requirements according to user needs.

[0129] Specifically, the process of the test case generation model generating test cases includes: I. LLM Adaptation and Fine-tuning Domain data collection: Obtain a training and test case set for secondary equipment in a substation; the training and test case set includes training and test requirements and corresponding training and test cases, Specifically, widely collect training and test cases from the historical records of substation secondary equipment tests, technical documents of various relay protection devices, industry standard specifications, and actual test cases to form a training and test case set. These training and test cases cover test requirements, test steps, expected results, etc. under different equipment types, test scenarios, and fault modes. For example, collect test cases for transformer protection devices produced by different manufacturers under fault scenarios such as overload and short circuit, as well as the corresponding test procedures and expected protection actions.

[0130] Model selection and loading: Select a large language model (LLM) with powerful natural language processing capabilities, such as the GPT series, Wenxin Yiyan, DeepSeek, or other open-source pre-trained models. Load the selected LLM pre-trained model into a local computing environment or a cloud platform to ensure that the model can run stably and receive input data for processing.

[0131] Fine-tuning process: Using the data collected in the substation field, i.e., a large number of training and test cases of secondary substation equipment, fine-tune the pre-trained LLM model. During the fine-tuning process, organize the training and test case data in a certain format. For example, use the test requirement description as the input text, and the corresponding test template fragment (test case) or optimization suggestion as the output label. Through the backpropagation algorithm, adjust the parameters of the pre-trained LLM model so that it can better understand and generate content related to the tests of secondary substation equipment. During the fine-tuning process, transfer learning techniques can be adopted to utilize the language knowledge and semantic understanding ability learned by the pre-trained model in general natural language processing tasks to accelerate the learning process in the substation test field. At the same time, set appropriate hyperparameters such as the learning rate and the number of training epochs to balance the training speed and performance of the model, avoid overfitting or underfitting, and finally obtain a test case generation model.

[0132] II. Test Task Analysis and Requirement Extraction Task Input Parsing: When a new test task is received, first, the test case template generation model parses the task description. The task description may be provided in the form of natural language text, including the purpose of the test, relevant information about the test object (such as equipment type, model, operating status, etc.), and specific requirements of the test (such as testing a certain fault scenario, verifying a certain functional module, etc.). The test case generation model uses natural language processing techniques, such as word segmentation, part-of-speech tagging, named entity recognition, etc., to decompose the task description into understandable semantic units and extract key information.

[0133] Requirement Understanding and Transformation: Based on the LLM test case generation model, according to the parsed task information and combined with the substation test knowledge learned during fine-tuning, understand the specific requirements of the test task. For example, if the task is to conduct a fault simulation test on the relay protection device of a certain transmission line, the test case generation model can identify that the test object is the relay protection device of the transmission line, and the test requirement is to simulate the fault scenario and verify the operation of the protection device. Then, the test case generation model transforms these requirements into an internal representation form for subsequent generation and optimization of test templates.

[0134] III. Test Template Generation and Optimization Initial Template Generation: The test case generation model generates a test case template based on its understanding of the test task requirements and the knowledge and patterns learned after fine-tuning. The content of the test case template includes the general framework of the test steps, the types of test signals to be applied and the parameter ranges, and the expected test results, etc. For the above test task of the relay protection device of the transmission line, the test case template may include test steps such as setting different types of fault current and voltage signals, observing the tripping time of the protection device and the output of the action signal, as well as the corresponding expected tripping time range and correct action signal.

[0135] Dynamic optimization process: Considering the complexity and variability of the on-site test environment, the test case generation model dynamically optimizes the test case template in the following ways. First, based on the real-time feedback information on-site, such as the real-time operating parameters of the device, the actual results of the previous test steps, etc., the test case template is adjusted. If it is found during the test that the actual operation time of the protection device is faster than expected, the test case generation model can analyze the possible reasons, such as device parameter differences, test signal interference, etc., and accordingly adjust the application time or parameters of the signal in the subsequent test steps. Second, the test case generation model can also optimize the template according to historical test data and experience. If historical data shows that a certain test step is prone to misjudgment under a certain specific device model and operating conditions, the test case generation model can add additional verification steps or adjust the combination method of test signals to improve the accuracy and reliability of the test. In addition, the test case generation model can also interact with testers, receive feedback and suggestions from testers, and further optimize the test template. Testers can point out unreasonable or difficult-to-operate parts in the template, and the test case generation model adjusts according to these feedbacks to make the test case template more in line with the actual test requirements.

[0136] IV. Template Verification and Application Simulation verification: Before actually applying the generated test case template, use simulation tools or historical data to conduct simulation verification on the test case template. Apply the test steps and expected results in the test case template to the simulation environment, simulate the test process, and check whether the test case template can accurately detect device faults and abnormal conditions, and whether the expected results match the actual situation. If problems are found during the simulation verification process, such as the test steps cannot be executed, the expected results do not match the actual situation, etc., feedback the problems to the test case generation model to optimize the test case template again.

[0137] Actual application and feedback: Apply the optimized test case template to the actual substation secondary equipment test. During the test, record the test data and results in real time, and feedback this information to the test case generation model. The test case generation model continuously optimizes and improves the test case template according to the actual test feedback. If it is found in the actual test that a certain test step shows significant differences on different devices, the test case generation model can further refine the test case template according to these differences to make it better adapt to the test requirements of different devices. At the same time, record the successful experiences and failure lessons of each test as new training data for further fine-tuning the test case generation model to improve its ability in test case template generation and optimization.

[0138] In a further embodiment, after using the test case to perform collaborative testing on the different substations and / or different secondary equipment in the substation to generate a test result corresponding to the test requirement, the method further includes: Obtain the real-time device signals of the secondary equipment of the substation; Input the real-time device signals into a pre-trained risk warning model, and predict the fault risk of the secondary equipment of the substation through the risk warning model; Execute a warning action corresponding to the fault risk.

[0139] The large artificial intelligence model also has the ability to predict fault risks. It can not only analyze the current test results, but also combine information such as the historical performance and fault records of the equipment to predict the possible future fault types or risks, so as to give early warnings in advance. This function is of great significance for preventive maintenance and ensuring equipment safety.

[0140] Specifically, the training process of the risk warning model and the process of predicting fault risks include: I. Data collection and collation Collect data from various data sources of the secondary equipment of the substation, including but not limited to the operation data of equipment such as relay protection devices, intelligent terminals, and merging units, that is, the device signals of the secondary equipment of the substation. These data cover real-time monitored electrical quantity data (such as the amplitude and phase of voltage and current), device status information (such as switch positions, the input and output states of protection devices), communication messages (GOOSE, SV, MMS signals), as well as historical fault records and equipment maintenance records. Clean the collected device signal data, remove duplicates, errors, and outliers, unify the data format, and integrate device signals of different devices and different types into a standardized database for convenient subsequent processing and analysis.

[0141] II. Feature engineering Analyze the collected device signal data and extract features related to faults or risks. For electrical quantity data, calculate features such as voltage fluctuation amplitude, current harmonic content, and power factor change rate; for device status information, pay attention to the number of switch operations and the action frequency of protection devices; extract features such as signal transmission delay and packet loss rate from communication messages. Use statistical analysis, correlation analysis, and other methods to screen out key features with high sensitivity and discrimination for fault or risk prediction, remove redundant features, reduce the data dimension, and improve the model training efficiency and prediction accuracy.

[0142] III. Model construction and training Select a suitable model for fault or risk early warning, such as decision trees, random forests, support vector machines based on machine learning, and recurrent neural networks (RNN) and their variants such as long short-term memory networks (LSTM) and gated recurrent units (GRU) in deep learning. Use historical data to train the risk early warning model, taking the feature data as input and the corresponding fault or risk labels (normal, minor fault risk, severe fault risk, etc.) as output. During the training process, by adjusting the parameters of the risk early warning model, selecting a suitable loss function and optimization algorithm (such as the Adam optimizer), minimize the error between the model prediction result and the actual label, so that the model learns the mapping relationship between data features and fault risks. Adopt cross-validation to evaluate the model performance to ensure that the risk early warning model has good generalization ability and avoid overfitting.

[0143] IV. Real-time Monitoring and Data Processing: Obtain the real-time device signals of the secondary equipment in the substation.

[0144] Specifically, during the operation of the substation, collect the real-time device signals of the secondary equipment in the substation.

[0145] V. Fault or Risk Prediction and Early Warning: Input the real-time device signals into the pre-trained risk early warning model, and predict the fault risk of the secondary equipment in the substation through the risk early warning model. According to the same preprocessing and feature extraction process as the training data, process the real-time device signals to obtain the feature vectors for model prediction. Input these feature vectors into the trained risk early warning model in real time. The model predicts the fault risk status of the secondary equipment in the substation according to the input real-time feature vectors, and then executes the early warning actions corresponding to the fault risks. If the prediction result of the risk early warning model is that there is a fault risk, different levels of early warnings are given according to the severity of the risk. For minor fault risks, such as predicting that the temperature of a certain device is gradually rising but has not reached the danger threshold, the operation and maintenance personnel may be prompted by a pop-up window in the substation monitoring system to pay attention to the device status, record relevant data and increase the monitoring frequency; for severe fault risks, such as the risk early warning model predicting that the protection device is about to malfunction, the risk early warning model immediately notifies the operation and maintenance personnel through multiple methods such as text messages, emails, and voice alarms, and at the same time issues a strong audible and visual alarm in the monitoring system and provides detailed early warning information, including possible fault types, affected ranges, and recommended measures.

[0146] VI. Early Warning Verification and Feedback After receiving the warning information, the operation and maintenance personnel will inspect and test the secondary equipment of the substation to verify the accuracy of the warning. If the warning is correct, the warning event and the processing results will be recorded and fed back to the risk warning model training process as new data to further optimize the risk warning model; if the warning is wrong, the cause of the error will be analyzed, which may be data anomalies, unreasonable parameter settings of the risk warning model, or inaccurate feature selection, etc., and targeted adjustments and improvements will be made to retrain the risk warning model to improve the warning accuracy of the model. The warning performance of the risk warning model will be evaluated regularly, and the accuracy rate, recall rate and other indicators of the warning will be statistically analyzed to continuously optimize the model to ensure that it can timely and accurately warn of failures or risks of substation equipment.

[0147] In addition, through the rule engine, the model can process complex test logic and make judgments based on conditions. Data-driven reasoning relies on the model's historical learning and real-time data input to achieve comprehensive analysis of multi-dimensional information. In this way, the system can not only complete routine test judgments, but also provide accurate fault diagnosis and operation suggestions in abnormal situations.

[0148] I. Rule Formulation and Sorting Organize power industry experts and senior testers to formulate rules applicable to substation secondary equipment testing based on the operating principles of power systems, technical specifications of relay protection devices (such as the provisions of the IEC61850 standard on signal transmission and equipment action logic), industry-related standards, and long-term accumulated field testing experience. Focusing on the test process, it is clear that before conducting a whole set of loop tests, it is necessary to ensure that all equipment has completed single-unit debugging and is in normal status; for signal judgment, the allowable error range of SV signal sampling values ​​and the upper limit of GOOSE signal transmission delay are specified; for equipment action logic, the action time and action sequence requirements of the protection device under different fault types are determined. These rules are classified and organized to form a structured rule set, covering different scenarios such as normal operation judgment, fault diagnosis, and abnormal warning.

[0149] 2. Digital Expression of Rules Use rule description languages ​​(such as Drools' DRL language) to convert the organized rules into a computer-recognizable form. Give each rule a unique identifier to clearly define the conditional part and action part of the rule. For example, for the rule "If the current value detected by the protection device is greater than the overcurrent protection setting value and lasts for more than 50 milliseconds, it is determined to be an overcurrent fault, a tripping command is issued, and the fault information is recorded", ensure that the digital expression of the rule is accurate and concise, which is convenient for the rule engine to parse and execute.

[0150] 3. Rule Engine Configuration and Integration Select a suitable rule engine (such as Drools, Jess, etc.) and configure it according to the architecture and requirements of the substation secondary equipment test system. Set the operating parameters of the rule engine, such as memory allocation, thread pool size, etc., to optimize its performance. Integrate the rule engine into the test system and establish interfaces with the test data acquisition module and the device control module. The test data acquisition module real-time obtains the operating data of secondary equipment (such as MMS, GOOSE, SV signals, device status information, etc.) and transfers it to the rule engine; according to the judgment result, the rule engine performs corresponding operations on the device through the device control module (such as sending control instructions, adjusting test parameters).

[0151] IV. Real-time Data Processing and Rule Matching When the test system is running, the rule engine receives the data transmitted by the test data acquisition module in real time. Preprocess the data, including data cleaning (removing noise data and outliers), format conversion (unifying data types and units). Match the preprocessed data with the rules in the rule library. Adopt an efficient matching algorithm (such as the Rete algorithm), which stores rule conditions by constructing a node network, reduces repeated calculations, and quickly locates the rules that match the current data. For example, when receiving the current data of a certain protection device, the rule engine quickly filters out the rules related to the current for matching judgment.

[0152] V. Condition Judgment and Decision Execution Once the data successfully matches the condition part of a certain rule, the rule engine immediately triggers the action part of the rule. If it is judged that the device is in normal operation state, the rule engine can send a normal signal to the monitoring system to maintain the continuation of the test process; if it is judged as an abnormal or faulty situation, such as detecting that the protection device fails to operate (meeting the rule condition of "a fault occurs and reaches the action time, and the protection device does not send an action signal"), the rule engine executes corresponding actions, such as generating a detailed fault report (including fault type, occurrence time, relevant data), sending an alarm message to the operation and maintenance personnel (by means of SMS, email or in-station message, etc.), and at the same time controlling the test system to pause the current test task and wait for the operation and maintenance personnel to handle it.

[0153] VI. Result Feedback and Rule Optimization After the rule engine executes the decision, it feeds back the result to other modules of the test system. For example, the test report generation module records the judgment result, and the device management module adjusts the device operation state according to the result. Regularly evaluate the operation results of the rule engine and collect feedback data in actual tests. If it is found that the rule judgment is inaccurate (such as misjudgment, missed judgment) or cannot handle newly emerging test scenarios, analyze the reasons in time. Organize experts to revise, supplement or optimize the rules, update the rule library, and redeploy it to the rule engine to ensure that the rule engine always adapts to the requirements of substation secondary equipment tests and improves the accuracy and reliability of the test system.

[0154] In addition, the core task of the large model is to conduct intelligent diagnosis and judgment on the real-time test process. Through the inflow of real-time data, the system can analyze the timing changes of signals according to the pre-established model and perform anomaly detection by comparing with historical patterns. Whenever a signal change or test condition change occurs, the model will immediately analyze it and output a diagnostic result. For example, when the signal of a certain protection device does not change as expected, the model will automatically identify this anomaly and speculate on possible fault causes.

[0155] I. Collection and Arrangement of Historical Test Data Comprehensively collect historical test data of secondary equipment in substations, covering data of different equipment types, test scenarios, and operation stages. Include the action records of relay protection devices, such as action time and action type; various signal data, like the real-time values and change situations of GOOSE, SV, and MMS signals; test input parameters, such as the magnitude and duration of simulated fault quantities; test environment information, such as temperature, humidity, and grid load. Obtain data from multiple channels, such as manual test records, automatic test system logs, and equipment operation and maintenance reports, and perform standardized processing to unify the data format, eliminate data noise and error values, and ensure data accuracy and consistency.

[0156] II. Data Classification and Labeling Classify historical data according to dimensions such as test equipment type, test items, and fault types. For example, classify data by equipment type into line protection device data, transformer protection device data, etc.; classify data by test items into single-function test data, integrated circuit test data, etc.; classify data by fault types into short-circuit fault data, over-current fault data, etc. Add labels to each type of data to clarify normal test results, various abnormal results, and their corresponding fault causes, such as "Normal - No Fault", "Abnormal - Protection Device Refusal to Operate - Hardware Fault", etc., and construct a label system that is convenient for retrieval and analysis.

[0157] III. Construction of Data Analysis Model Use machine learning algorithms, such as decision trees, random forests, support vector machines, etc., to model the classified and labeled data. Taking fault type prediction as an example, use the signal characteristics and test conditions in historical test data as inputs and the fault type label as the output to train the model to learn the association patterns between data. Utilize deep learning models, such as convolutional neural networks (CNNs) to process image-like signal data and long short-term memory networks (LSTMs) to analyze test data with temporal characteristics, to mine deep features and laws of the data and improve the model's prediction and analysis capabilities.

[0158] IV. Comparison and Analysis of Real-Time Test Data During the current test process, the test data of secondary equipment is collected in real time and compared with historical test data. Key indicators such as signal amplitude, change trend, and action time are compared. If the deviation between the real-time data and the historical normal data exceeds the preset threshold, the abnormal analysis process is started. The trained model is used to analyze the real-time data to predict the possible fault types and development trends. For example, analyze the timing change of the protection device action signal to predict whether there will be misoperation or refusal of protection.

[0159] V. Decision-making mechanism adjustment If the model predicts an anomaly, the decision-making mechanism is adjusted based on the handling experience and effect feedback of similar anomalies in historical data. For anomalies that occur frequently and for which effective handling measures have been identified, such as signal anomalies caused by a certain type of communication interference, targeted handling suggestions are directly given, such as checking the communication line connection and restarting the communication module. For newly emerging anomalies or those with poor handling effects, exploratory decisions are made by combining the characteristics of historical data and the experience of domain experts, such as trying different test parameter adjustments and conducting additional equipment inspection items, and continuously tracking the implementation effect of the decisions to update the historical data and the model.

[0160] VI. Decision implementation and effect evaluation Operate according to the adjusted decision-making mechanism, such as adjusting the test plan, performing maintenance on the equipment, etc., and record the operation process and results. Compare the changes in test data before and after the operation to evaluate the implementation effect of the decision. If the anomaly is resolved, the handling process and results of this time are added to the historical data as a successful case; if the anomaly has not improved or new problems occur, re-analyze the historical data and further optimize the decision-making mechanism until the problem is solved, continuously improving the accuracy and effectiveness of the decision-making mechanism.

[0161] In addition, the automatic test system software and the acceptance test host complete the functions of issuing test control commands and feedback of test results, communicate with the relay protection device under test based on the IEC61850 standard using the CMS / MMS communication protocol, implement the function of issuing control commands to the protection device and obtaining information such as device action reports, wave recording, and telecontrol signal changes, and the automatic test system software realizes test task scheduling control and result discrimination, ultimately realizing a fully automatic closed-loop test.

[0162] Based on the existing general test template, the result data of each test item in the test template is obtained through a flexible configuration method, and a report template in a specific format is generated. During the automatic test process, the test results are automatically filled into the test report in real time until a specific format report is automatically generated after one-key testing is completed; The configuration method of the test report is flexible and editable. First, based on the debugging report form, the header, test items, and test data table are edited and generated. The results in the table support dragging and selecting the data results in the debugging template. If the above conditions are met, the original debugging report form can be restored to the maximum extent, so that the configuration method generates the entire set of tests performed by the debugging template and does not affect the test report of the original manual test, ensuring high adaptability and availability when the test result data is reported.

[0163] In the process of generating a specific report, first of all, the format of the report is generally in table form. The rows and columns of the table can be generated on a completely blank basis. The text or the test results in the associated test template can be edited at will in the formed row and column cells. Ultimately, the format and content of the original test report can be restored. At the same time, the test results can be automatically filled in according to the judgment results during the test process to generate a debugging report with test results.

[0164] In addition, considering that in actual applications, disaster scenarios such as typhoons have brought severe challenges to substation secondary equipment such as substation relay protection devices. For example, after a typhoon passes and causes equipment abnormalities, the maintenance test of the substation secondary equipment faces a series of outstanding problems: the test process has a low degree of automation, resulting in low efficiency of maintenance tests and difficulty in responding quickly in emergency situations; the system has poor coordination, making it difficult to coordinate tests between different devices, affecting the overall test effect; the test strategy lacks adaptability and cannot be adjusted in time according to the complex equipment conditions after the typhoon disaster. At the same time, existing testing methods are difficult to meet the needs of multi-device collaborative testing, dynamic strategy optimization, and human-computer intelligent interaction in emergency power restoration scenarios, which seriously restricts the improvement of post-disaster repair efficiency and safety assurance capabilities. In order to further improve the application effect of the secondary equipment testing method based on the generative large model proposed in the present invention, the present invention can also address the difficulties in substation secondary equipment testing in disaster scenarios by adopting the following technical means: 1) By integrating multi-source maintenance data including real-time monitoring data of equipment operation, historical fault data, maintenance records and other information with the domain knowledge base that integrates power industry standards, professional technical knowledge and expert experience, a large model of intelligent maintenance test for substation relay protection is constructed based on the existing general or industry large model; the model integrates a large amount of data related to relay protection (including equipment operation data, fault cases, test standards, etc.) and conducts in-depth mining and analysis of these data, so that the model can better adapt to relay protection test scenarios and accurately understand and handle complex problems in relay protection tests; 2) Based on the integration of AIGC (Artificial Intelligence Generated Content) technology and graph neural network, we build intelligent test scheme generation technology, and use this technology to establish a deep connection between the equipment topology that reflects the connection and layout relationship of the internal equipment of the substation and the test logic that specifies the test process and test method, so as to realize the intelligent construction and dynamic adaptation of test schemes and test templates. It can intelligently generate efficient and adaptive test schemes and test templates according to different equipment status, fault types and test requirements; these templates not only cover comprehensive test steps and parameter settings, but also can be adjusted dynamically in real time according to actual conditions (such as changes in equipment status) to adapt to various complex and changeable test scenarios; 3) Based on the multi-objective test strategy optimization framework driven by reinforcement learning, a dynamic adjustment mechanism for test parameters driven by disaster scenarios is constructed. For example, in the emergency power supply scenario after a typhoon passes, the model is allowed to learn and optimize the test strategy in the continuous test process based on the reinforcement learning algorithm, so that the test parameters and processes can be dynamically adjusted according to the different impacts of disasters on the equipment (real-time status feedback of the equipment and test results). These experiences are accumulated to form a differentiated test strategy knowledge base, providing flexible and optimal maintenance test solutions for substation secondary equipment. In actual applications, according to different disaster scenarios, appropriate test strategies can be quickly obtained from the knowledge base to ensure that the test can efficiently and accurately detect potential problems of the equipment, improve the pertinence and effectiveness of the test, and thus improve the reliability of the equipment during emergency power supply. 4) Based on digital twin technology, substation equipment and test processes are virtually modeled, and a visual interactive interface based on digital twin is constructed. Operators can easily communicate and collaborate with the equipment. It can not only intelligently guide the test process according to the operator's instructions and equipment status, and provide real-time operation suggestions and risk warnings, but also reflect the actual status of the equipment and test in real time. Through interaction with operators, the test process and strategy can be continuously optimized to achieve efficient collaboration between humans and machines and coordinated coordination among multiple systems, thereby effectively improving the collaborative working ability and reliability of the secondary equipment system of the entire substation, providing pioneering auxiliary support for efficient and high-quality maintenance tests of secondary equipment for emergency power transmission after disasters, and significantly improving the recovery capability and operational safety of the entire power system after disasters.

[0165] Based on the above technical scheme, in special scenarios where disasters such as typhoons cause substation equipment failure, tripping and power outages, the application value of the secondary equipment testing method based on the generative large model provided by the present invention is further enhanced, and it can effectively meet the needs of power grid safety protection equipment such as relay protection to complete maintenance tests efficiently and with high quality before power supply is restored.

[0166] See also Figure 2 ,Figure 2 This is a schematic structural diagram of an embodiment of a secondary equipment test device based on a generative large model provided by the present invention. As Figure 2 shown, the device includes: A test requirement acquisition module 201, configured to acquire test requirements of substation secondary equipment; the test requirements at least include equipment test signals with different signal specifications between different substations and / or different secondary equipment in a substation.

[0167] A test case generation module 202, configured to generate test cases corresponding to the test requirements according to the standard signals corresponding to the equipment test signals through a pre-trained test case generation model; the standard signals are obtained by classifying the equipment signals of substation secondary equipment based on a standard signal model, and are used to represent the signal specifications uniformly followed between some or all different substations and / or between some or all different secondary equipment in a substation; the standard signal model is configured to identify common attributes between equipment signals with different signal specifications to form the standard signals.

[0168] A collaborative test module 203, configured to perform collaborative tests on the different substations and / or different secondary equipment in a substation using the test cases to generate test results corresponding to the test requirements.

[0169] In the present invention, the test case generation module 202 identifies the common attributes of equipment signals with different signal specifications through the standard signal model to form standard signals representing uniformly followed signal specifications; furthermore, the test case generation module 202 uses the standard signals to generate test cases for collaborative tests between substations and / or different secondary equipment with different signal specifications, which can improve the test collaboration between substations and / or secondary equipment with different signal specifications. In addition, using standard signals can ignore the signal differences between different signal specifications, improve the interoperability between substations and / or secondary equipment with different signal specifications, and thus improve the test efficiency.

[0170] In a further embodiment, the construction process of the standard signal model includes: Obtaining a first set of equipment signals of the substation secondary equipment; the first set of equipment signals includes standard signals corresponding to equipment signals with different signal specifications between different substations and / or different secondary equipment in a substation and data labels corresponding to each of the standard signals; Performing preprocessing on the first set of equipment signals; the preprocessing at least includes data cleaning; Dividing the preprocessed first set of equipment signals into at least a training set and a validation set; Use the standard signal model constructed by training with the training set to identify the common attributes of device signals with different signal specifications between different substations and / or different secondary devices in a substation to form the standard signal, and use the verification set to verify the standard signal model until the iteration termination condition is met to obtain a trained standard signal model.

[0171] In a further embodiment, the apparatus further includes: A fusion module, configured to fuse at least two standard signal models constructed and trained through machine learning and / or deep learning to obtain a fused standard signal model.

[0172] In a further embodiment, the categories of the standard signal models include: a standard signal model for the device operation stage, a standard signal model for the device failure type, a standard signal model for the device signal nature, and a standard signal model for the device type; A test case generation module 202, including: A test case generation unit, configured to generate test cases corresponding to the test requirements according to at least one category of standard signals corresponding to the device test signals.

[0173] In a further embodiment, the test case generation module 202 includes: A signal extraction unit, configured to extract the device test signals included in the test requirements according to the test case generation model.

[0174] A standard signal mapping unit, configured to determine the standard signal corresponding to the device test signal based on a signal mapping model; the signal mapping model at least includes a mapping rule between the device test signal and the standard signal.

[0175] A test case generation unit, configured to generate initial test cases corresponding to the test requirements in a template configuration manner according to the standard signal through the test case generation model; the initial test cases at least include the standard signal corresponding to the device test signal, the test process, and the test result.

[0176] An instantiation unit, configured to instantiate the standard signal in the initial test cases into the device test signal to form instantiated test cases.

[0177] In a further embodiment, the construction process of the signal mapping model includes: Perform pre-training with general data to obtain a mapping pre-training model; the mapping pre-training model includes a machine learning model and / or a deep learning model; Obtain a second set of device signals of the secondary equipment of the substation; the second set of device signals includes at least data tags characterizing the mapping relationship between device signals and standard signals; Preprocess the second set of device signals; the preprocessing includes at least data cleaning; Use the preprocessed second set of device signals to fine-tune the mapping pre-training model to obtain a trained signal mapping model.

[0178] In a further embodiment, the device further includes: A mapping rule adjustment module for dynamically adjusting the mapping rules of the signal mapping model through a reinforcement learning algorithm.

[0179] In a further embodiment, the test case generation unit includes: A test case generation subunit for generating an initial test case corresponding to the test requirement in a template configuration manner according to at least one of the standard signals included in the prompt item, judgment item, pressure plate control item, and overall output fault item through the test case generation model.

[0180] In a further embodiment, the instantiation unit includes: A standard signal indexing subunit for determining a standard signal corresponding to a device signal matching the device test signal from a pre-configured dictionary index by using a similarity measurement method; the dictionary index includes at least device signals, standard signals, and the mapping relationship between device signals and standard signals.

[0181] An instantiation subunit for instantiating the standard signal corresponding to the device signal matching the device test signal in the initial test case into the device test signal to form an instantiated test case.

[0182] In a further embodiment, the standard signal indexing subunit includes: A similarity indexing subunit for determining candidate device signals with a similarity exceeding a preset similarity threshold with the device test signal from the dictionary index by using a similarity measurement method based on a corpus and / or knowledge base.

[0183] A standard signal determination subunit for determining the device signal with the highest similarity with the device test signal among the candidate device signals, and obtaining the standard signal corresponding to the device signal with the highest similarity from the dictionary index as the standard signal corresponding to the device signal matching the device test signal.

[0184] In a further embodiment, the construction process of the test case generation model includes: Obtain a set of training and test cases for the secondary equipment of the substation; the set of training and test cases includes training and test requirements and corresponding training and test cases. Fine-tune and train a generative pre-training model for generating test cases using the set of training and test cases to obtain the test case generation model.

[0185] In a further embodiment, the device further includes: A risk warning module, configured to obtain real-time device signals of the secondary equipment of the substation, input the real-time device signals into a pre-trained risk warning model, predict the fault risk of the secondary equipment of the substation through the risk warning model, and perform warning actions corresponding to the fault risk.

[0186] For the steps executed by each module or unit (and sub-units, etc.) in the secondary equipment test device based on the generative large model and the functional effects that can be achieved, please specifically refer to the relevant descriptions of each step in the corresponding method embodiment, and will not be elaborated here. Each module in the above device can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0187] Furthermore, an embodiment of the present invention also provides an electronic device. Among them, the above electronic device includes a server and / or a terminal. The terminal and the server can both independently execute the secondary equipment test method based on the generative large model provided in the embodiment of the present invention, and the terminal and the server can also cooperate to execute the secondary equipment test method based on the generative large model provided in the embodiment of the present invention. The terminal communicates with the server through a network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately, integrated on the server, or placed on the cloud or other servers.

[0188] The server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, a smart glasses, etc.

[0189] Among them, the server may specifically include: at least one processor, at least one memory, a power supply, a communication interface, an input / output interface, and a communication bus. In this embodiment, the power supply is used to provide working voltage for each hardware device on the server; the communication interface can create a data transmission channel between the server and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of the present invention, and no specific limitation is imposed on it here; the input / output interface is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and no specific limitation is imposed here. Generally, the terminal in this embodiment includes: a processor and a memory. In some embodiments, the terminal may further include a display screen, an input / output interface, a communication interface, sensors, a power supply, and a communication bus.

[0190] Among them, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor can be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for the rendering and drawing of the content to be displayed on the display screen. In some embodiments, the processor may further include an AI (Artificial Intelligence) processor, and the AI processor is used to process computational operations related to machine learning.

[0191] Among them, the memory is used to store computer programs, which are loaded and executed by the processor to implement the relevant steps in the request processing disclosed in any of the foregoing embodiments. As a carrier for storing resources, the memory can be a read-only memory, a random access memory, a magnetic disk, an optical disk, etc. The resources stored thereon include an operating system, computer programs, data, etc. The storage method can be temporary storage or permanent storage. Among them, the operating system is used to manage and control each hardware device and computer program on the server to enable the processor to perform operations and processing on the data in the memory. It can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer programs that can be used to complete the request processing method disclosed in any of the foregoing embodiments, the computer programs can further include computer programs that can be used to complete other specific tasks. In addition to the data that can include data such as update information of the application program, the data can also include data such as developer information of the application program.

[0192] In summary, the present invention provides a secondary device test method, device and equipment based on a generative large model. By obtaining device test signals with different signal specifications between different substations and / or different secondary devices in a substation; through a pre-trained test case generation model, generating test cases corresponding to test requirements according to the standard signals corresponding to the device test signals; using the test cases to perform collaborative testing on different substations and / or different secondary devices in a substation, and generating test results corresponding to the test requirements. By identifying the common attributes of device signals with different signal specifications through a standard signal model, forming a standard signal representing the uniformly followed signal specification; and then using the standard signal to generate test cases for collaborative testing between substations with different signal specifications and / or different secondary devices, the test collaboration between substations and / or secondary devices with different signal specifications can be improved. In addition, using the standard signal can ignore the signal differences between different signal specifications, improve the interoperability between substations and / or secondary devices with different signal specifications, and thus improve the test efficiency.

[0193] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards.

[0194] It can be understood that the technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combinations of these technical features do not conflict, they should be considered as the scope described in this specification.

[0195] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the appended claims.

Claims

1. A secondary equipment test method based on a generative large model, characterized in that: The method comprises: Acquire test requirements for secondary equipment of a substation; the test requirements at least include equipment test signals with different signal specifications between different substations and / or different secondary equipment in a substation; Generate a test case corresponding to the test requirement according to the standard signal corresponding to the device test signal by using a pre-trained test case generation model; the standard signal is obtained by classifying the device signal of the substation secondary device based on the standard signal model, and is used to characterize the signal specification uniformly followed between some or all different substations and / or some or all different secondary devices in the substation; the standard signal model is configured to identify the common attributes between device signals of different signal specifications to form the standard signal; The test case is used to perform collaborative testing on the different substations and / or different secondary devices in the substation to generate a test result corresponding to the test requirement.

2. The secondary equipment test method based on a generative large model as claimed in claim 1, characterized in that: The process of constructing the standard signal model includes: Acquire a first device signal set of the substation secondary equipment; the first device signal set includes standard signals corresponding to device signals with different signal specifications between different substations and / or different secondary equipment in the substation and data tags corresponding to each of the standard signals; Preprocessing the first device signal set; the preprocessing at least includes data cleaning; Dividing the preprocessed first device signal set into at least a training set and a validation set; The standard signal model constructed by training with the training set is used to identify the common attributes of equipment signals with different signal specifications between different substations and / or different secondary equipment in the substation to form the standard signal, and the standard signal model is verified with the verification set until the iteration termination condition is met to obtain the trained standard signal model.

3. The secondary equipment test method based on the generative large model as claimed in claim 2, characterized in that: The method further comprises: At least two standard signal models constructed and trained by machine learning and / or deep learning are fused to obtain a fused standard signal model.

4. The secondary equipment test method based on a generative large model as described in claim 1, 2 or 3, characterized in that: The categories of the standard signal models include: standard signal models for equipment operation phase, standard signal models for equipment failure types, standard signal models for equipment signal properties, and standard signal models for equipment types; Generating a test case corresponding to the test requirement according to a standard signal corresponding to the device test signal includes: A test case corresponding to the test requirement is generated according to at least one category of standard signals corresponding to the device test signal.

5. The secondary equipment test method based on generative large model as claimed in claim 1, characterized in that: The test case generation model trained in advance is used to generate a test case corresponding to the test requirement according to a standard signal corresponding to the device test signal, including: Extracting the device test signal included in the test requirement according to the test case generation model; Determine a standard signal corresponding to the device test signal based on a signal mapping model; the signal mapping model at least includes a mapping rule between the device test signal and the standard signal; Generate an initial test case corresponding to the test requirement in a template configuration manner according to the standard signal through the test case generation model; the initial test case at least includes the standard signal corresponding to the device test signal, the test process and the test result; The standard signal in the initial test case is instantiated as the device test signal to form an instantiated test case.

6. The secondary equipment test method based on generative large model as claimed in claim 5, characterized in that: The construction process of the signal mapping model includes: Pre-training with general data to obtain a mapping pre-training model; the mapping pre-training model includes a machine learning model and / or a deep learning model; Acquire a second device signal set of the substation secondary device; the second device signal set at least includes a data tag representing a mapping relationship between a device signal and a standard signal; Preprocessing the second device signal set; the preprocessing at least includes data cleaning; The preprocessed second device signal set is used to fine-tune the mapping pre-training model to obtain a trained signal mapping model.

7. The secondary equipment test method based on generative large model as claimed in claim 6, characterized in that: The method further comprises: The mapping rules of the signal mapping model are dynamically adjusted through a reinforcement learning algorithm.

8. The secondary equipment test method based on generative large model as claimed in claim 5, characterized in that: The generating model of the test case to generate the initial test case corresponding to the test requirement in a template configuration manner according to the standard signal includes: The test case generation model is used to generate initial test cases corresponding to the test requirements in a template configuration manner according to standard signals included in at least one of the prompt class, judgment class, pressure plate control class and the entire group of output fault classes.

9. The secondary equipment test method based on generative large model as claimed in claim 5, characterized in that: The instantiating the standard signal in the initial test case into the device test signal to form an instantiated test case includes: Determine a standard signal corresponding to a device signal matching the device test signal from a pre-configured dictionary index using a similarity measurement method; the dictionary index at least includes a device signal, a standard signal, and a mapping relationship between the device signal and the standard signal; The standard signal corresponding to the device signal matching the device test signal in the initial test case is instantiated as the device test signal to form an instantiated test case.

10. The secondary equipment test method based on generative large model according to claim 9, characterized in that: The method of using a similarity measurement method to determine a standard signal corresponding to a device signal matching the device test signal from a pre-configured dictionary index includes: Using a similarity measurement method based on a corpus and / or a knowledge base, determining from the dictionary index a candidate device signal whose similarity with the device test signal exceeds a preset similarity threshold; The device signal with the highest similarity to the device test signal among the candidate device signals is determined, and the standard signal corresponding to the device signal with the highest similarity is obtained from the dictionary index as the standard signal corresponding to the device signal matching the device test signal.

11. The secondary equipment test method based on generative large model according to claim 1, characterized in that: The construction process of the test case generation model includes: Obtaining a set of training test cases for the substation secondary equipment; the set of training test cases includes training test requirements and corresponding training test cases; The training test case set is used to perform fine-tuning training on a generative pre-trained model for generating test cases to obtain the test case generation model.

12. The secondary equipment test method based on generative large model according to claim 1, characterized in that: After using the test case to perform collaborative testing on the different substations and / or different secondary devices in the substations to generate a test result corresponding to the test requirement, the method further includes: Acquire real-time device signals of the substation secondary equipment, input the real-time device signals into a pre-trained risk warning model, predict the failure risk of the substation secondary equipment through the risk warning model, and execute warning actions corresponding to the failure risk.

13. A secondary equipment test device based on a generative large model, characterized in that: The device comprises: A test requirement acquisition module is used to acquire test requirements of secondary equipment in a substation; the test requirements at least include equipment test signals with different signal specifications between different substations and / or different secondary equipment in a substation; A test case generation module, used to generate a test case corresponding to the test requirement according to a standard signal corresponding to the device test signal through a pre-trained test case generation model; the standard signal is obtained by classifying the device signal of the substation secondary device based on the standard signal model, and is used to characterize the signal specification uniformly followed between some or all different substations and / or some or all different secondary devices in the substation; the standard signal model is configured to identify the common attributes between device signals of different signal specifications to form the standard signal; The collaborative testing module is used to use the test case to perform collaborative testing on the different substations and / or different secondary devices in the substations to generate test results corresponding to the test requirements.

14. The secondary equipment test device based on the generative large model according to claim 13, characterized in that: The device also includes: The risk warning module is used to obtain the real-time device signal of the substation secondary equipment, input the real-time device signal into a pre-trained risk warning model, predict the failure risk of the substation secondary equipment through the risk warning model, and execute the warning action corresponding to the failure risk.

15. An electronic device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the secondary equipment testing method based on the generative large model as claimed in any one of claims 1 to 12 is implemented.

Images