Current transformer control method based on semi-supervised learning

Abstract

The present application relates to the field of artificial intelligence technology, and provides a current transformer control method based on semi-supervised learning. The method includes: an interactive power monitoring device extracts historical current signals, collects current transformer signals in real time to obtain a first current signal; marks historical signals, and constructs a polarity analysis model in combination with semi-supervised learning; inputs the first current signal to predict polarity, and obtains a first polarity verification result after verification; decomposes the verification result, traverses the error cause library to obtain the first polarity error cause; screens the correction scheme to obtain the first correction scheme, generates a first correction instruction after encoding; and transmits it to the motor to perform the correction operation. The present application solves the technical problems that the traditional current transformer polarity verification method is inefficient, easily affected by human factors, and cannot adapt to the expansion of the scale and complexity of the power system, and realizes the technical effect of intelligent diagnosis and correction of the current transformer polarity through a semi-supervised learning algorithm, thereby improving the stability and safety of the power system.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, specifically to the field of machine learning technology, and in particular to a current transformer control method based on semi-supervised learning. Background Art

[0002] With the continuous expansion of the scale and complexity of power systems, the accuracy of polarity verification of current transformers, as important measuring equipment in power systems, is crucial to the stable operation of the entire system. However, the traditional method of current transformer polarity verification mainly relies on manual operation and regular inspection. This method is not only inefficient, consumes a lot of manpower and time, but also easily leads to verification errors due to human factors, bringing potential risks to the safe operation of the power system. Summary of the invention

[0003] This application provides a current transformer control method based on semi-supervised learning, aiming to solve the technical problems that the traditional current transformer polarity verification method is inefficient, susceptible to human factors, and unable to adapt to the expansion of power system scale and increased complexity.

[0004] In view of the above problems, the present application provides a current transformer control method based on semi-supervised learning.

[0005] In the disclosed aspect of the present application, a current transformer control method based on semi-supervised learning is provided, the method comprising: an interactive power monitoring device, extracting a historical current signal, and collecting the input and output signals of the current transformer in real time to obtain a first current signal; data marking the historical current signal to obtain a first historical current signal and a first historical marked current signal, and constructing a current polarity analysis model in combination with semi-supervised learning, wherein the data marking is based on a division rule; inputting the first current signal into the current polarity analysis model, performing polarity prediction, obtaining a predicted polarity, performing polarity verification in combination with the actual polarity, and generating a first polarity verification result; decomposing the first polarity verification result, traversing a polarity error cause database based on the decomposition result, and obtaining a first polarity error cause; based on the first polarity error cause, screening a predetermined correction scheme set, extracting a first correction scheme, performing encoding conversion based on the first correction scheme, and generating a first correction instruction; transmitting the first correction instruction to the motor to execute a correction operation.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The above-mentioned current transformer control method based on semi-supervised learning extracts historical current signals through interactive power monitoring equipment, and collects the input and output signals of the current transformer in real time during operation to obtain the latest current signal. Subsequently, the extracted historical current signals are marked, and these marks are determined by the pre-established division rules. After that, a current polarity analysis model is constructed using the marked historical data and the semi-supervised learning algorithm. This model can predict the polarity of the current signal collected in real time and determine whether the direction of the current is correct. After the prediction is completed, it is compared and verified with the actual measured current polarity to obtain the polarity verification result. If the verification result shows that there is a polarity error, the error result is further analyzed and decomposed, and then the database of the cause of the error is searched to obtain the specific cause of the error. Based on the cause of the error, the most suitable solution is selected from the pre-set correction solution, and the solution is converted into a specific correction instruction. Then, the correction instruction is sent to the motor, and the motor performs the corresponding correction operation according to the instruction to correct the polarity error of the current transformer. The whole process realizes the automatic detection, analysis and correction of the polarity error of the current transformer, and improves the stability and safety of the power system.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a flow chart of a current transformer control method based on semi-supervised learning in one embodiment;

[0011] Figure 2 The figure is a flow chart of obtaining a current polarity analysis model of a current transformer control method based on semi-supervised learning in one embodiment. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a current transformer control method based on semi-supervised learning to solve the technical problems that the traditional current transformer polarity verification method is inefficient, easily affected by human factors, and cannot adapt to the expansion of power system scale and increase in complexity.

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0015] Embodiment 1

[0016] like Figure 1 As shown, the present application provides a current transformer control method based on semi-supervised learning, the method comprising:

[0017] Step S100: Interacting with the power monitoring device, extracting the historical current signal, and collecting the input and output signals of the current transformer in real time to obtain the first current signal;

[0018] Current transformer is one of the important equipment in the power system, which is used to convert high voltage and large current into low voltage and small current for subsequent measurement, protection and control. During the use of current transformer, the correctness of its polarity directly affects the normal operation of the power system.

[0019] In an embodiment of the present application, the interactive power monitoring device is an important tool in the power system, which is responsible for real-time tracking and recording the performance of the current transformer. The system terminal first connects to the power monitoring device to extract the historical current signal data stored in the power monitoring device, which records the performance of the current transformer over the past period of time. Subsequently, the current input and output signals of the current transformer are collected in real time with the help of the power monitoring device to obtain the latest current signal data, namely the first current signal. These real-time data are an important basis for understanding the current working status of the current transformer.

[0020] Step S200: data marking the historical current signal to obtain a first historical current signal and a first historical marked current signal, and constructing a current polarity analysis model in combination with semi-supervised learning, wherein the data marking is performed based on a division rule;

[0021] In one embodiment, for the historical current signals obtained, the system terminal first preprocesses these data, including noise filtering, data normalization and other operations. Noise filtering is to remove interference components in the signal, which come from various external factors, such as electromagnetic interference, equipment failure, etc. Through noise filtering, the signal-to-noise ratio of the signal can be improved, making the data purer and more reliable. Data normalization is to convert data of different ranges and dimensions into data of the same range and dimension. In the current signal, due to the influence of various factors, the data collected at different times and different devices may have large differences. Through data normalization, these data can be unified to the same scale, which is convenient for subsequent data analysis and processing. After completing the preprocessing, the system terminal performs data marking on the processed historical current signal. This data marking process is completed based on pre-established division rules. Through these rules, the system terminal can clearly distinguish and identify different features and states in the historical current signal, and generate a first historical current signal and a first historical marked current signal. Among them, the first historical current signal refers to an unmarked historical current signal. The first historical marked current signal is a marked historical current signal. Subsequently, the system terminal uses the first historical current signal and the first historical marked current signal in combination with semi-supervised learning technology to construct a current polarity analysis model. This model can learn and identify the polarity characteristics in the current signal, thereby achieving accurate analysis and prediction of the polarity of the current transformer.

[0022] Further, the step S200 includes:

[0023] Connecting to the expert system, marking the sample current signal, and obtaining the marked sample current signal;

[0024] A classification neural network is trained based on the labeled sample current signal, and a decision boundary is extracted from the classification neural network to generate the division rule.

[0025] Preferably, in order to improve the accuracy and efficiency of data division, the system terminal adopts a method combining an expert system and a classification neural network. First, an expert system is connected and the collected sample current signals are labeled using the system. Based on rich domain knowledge and experience, the expert system can accurately identify and distinguish different characteristics and states of the current signal and assign corresponding labels to it. The labeled sample current signals generated by this process provide an important basis for subsequent analysis and model training. Subsequently, the system terminal divides the sample current signals labeled by the expert system into a training set and a test set. The training set is used to train the neural network, and the test set is used to evaluate the performance of the trained network. After that, the system terminal constructs a classification neural network structure according to actual needs and the characteristics of the current signal data, and initializes network parameters such as weights and bias terms. Then, the labeled sample current signal in the training set is used as input, and the corresponding label is used as the target output, and the network output is calculated by forward propagation. The error between the network output and the target output is calculated using the cross entropy loss. The back propagation algorithm is then used to calculate the gradient of the error to each parameter in the network. Then, the network parameters are updated using gradient descent to reduce the error. Repeat the above process until the number of training rounds reaches the upper limit. The performance of the trained neural network is then evaluated using the test set, and indicators such as classification accuracy and recall are calculated. If the calculated indicators meet the requirements of the corresponding threshold, the classification neural network is output. Further, in the feature space, the neural network defines one or more decision boundaries through the learned parameters, namely weights and biases. These boundaries divide the feature space into different regions, each corresponding to a category. Subsequently, the gradient of the decision function is calculated, and the specific location of the decision boundary is found based on the change in the calculated gradient. This is because the gradient changes significantly near the decision boundary. After determining the location of the decision boundary, the system terminal extracts the key information of the decision boundary, including the boundary equation, key points on the boundary, etc., as part of the division rule. Afterwards, based on the extracted key information of the decision boundary, combined with expert knowledge and domain experience, a set of clear division rules is generated. These rules can not only help the system terminal to more accurately mark new current signals, but also make the constructed current polarity analysis model more accurate.

[0026] Further, if Figure 2 As shown, the step S200 includes

[0027] Training an initial current polarity analysis model based on the first historical marker current signal;

[0028] Predicting the first historical current signal using the initial current polarity analysis model to generate a pseudo label of the first historical current signal;

[0029] Optionally, the system terminal divides the first historically marked current signal into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to adjust hyperparameters and prevent overfitting, and the test set is used to evaluate the performance of the model. Subsequently, according to the complexity of the actual problem and the characteristics of the data, the initial current polarity analysis model architecture is constructed, and the parameters of the model, including weights and bias terms, are initialized. After that, the model is trained using the first historically marked current signal data in the training set. The model traverses the training data multiple times and iterates. In each iteration, the model receives a batch of current signals, calculates the prediction results, i.e., the predicted current polarity, through forward propagation, and calculates the error between the prediction results and the actual label through cross entropy loss. Then, the gradient of the error to the model parameters is calculated using the back propagation algorithm, and the model parameters are updated using, for example, gradient descent to reduce the error. Repeat the above iterative process until the performance of the model on the validation set reaches the preset standard. Finally, the performance of the trained model is evaluated using the test set. Calculate indicators such as the accuracy and precision of the model to understand the classification effect of the model on different categories. If the performance of the model is not ideal, the system terminal adjusts the model architecture and hyperparameters, and then retrains the model until the model performance meets expectations, and outputs it to generate an initial current polarity analysis model. Further, the system terminal uses this initial current polarity analysis model to predict the unlabeled first historical current signal. This prediction process allows the model to classify the first historical current signal based on the learned knowledge and give its predicted polarity label. The system terminal defines these predicted labels as pseudo-labels because they are not based on direct labels of the expert system, but are inferences made by the model based on its learned knowledge. This process not only demonstrates the learning ability of the model, but also provides a reference for subsequent data analysis and model optimization.

[0030] Calling the expert system to mark the first historical current signal, obtaining an actual label of the first historical current signal, comparing the actual label with the pseudo label, and calculating the accuracy of the pseudo label;

[0031] If the pseudo label accuracy is greater than or equal to the pseudo label accuracy threshold, the pseudo label is added as labeled data to the first historical labeled current signal, and the initial current polarity analysis model is trained twice to obtain a current polarity analysis model.

[0032] Optionally, the system terminal marks the first historical current signal through an expert system to obtain the actual labels of these labels. In order to evaluate the prediction ability of the model, the system terminal compares the pseudo-labels generated by the initial current polarity analysis model with the actual labels, and calculates the accuracy of the pseudo-labels. This accuracy is obtained by calculating the ratio of the number of labels that are the same as the pseudo-labels and the actual labels to the total number of labels, reflecting the degree of consistency between the model prediction results and the actual situation. If the accuracy of the pseudo-label reaches the preset pseudo-label accuracy threshold, the system terminal determines that the prediction ability of the initial current polarity analysis model meets the expected requirements and the pseudo-label has a certain reliability. Therefore, these pseudo-labels are added to the first historical labeled current signal as labeled data, further enriching the training data set of the model. Subsequently, the system terminal uses the updated first historical labeled current signal to conduct a secondary training of the initial current polarity analysis model. This training uses more data, including the original first historical labeled current signal and the newly added pseudo-label data, so that the model can learn more features and laws, and further improve its ability to analyze current polarity. Finally, the system terminal generates a current polarity analysis model based on the initial current polarity analysis model after secondary training. Compared with the initial model, this model has improvements in data volume and model performance, and can better adapt to the needs of practical applications.

[0033] Furthermore, the present application provides that when the pseudo label accuracy is less than the pseudo label accuracy threshold, the method further includes:

[0034] If the pseudo label accuracy is less than the pseudo label accuracy threshold, the initial current polarity analysis model is optimized, the first historical current signal is predicted again using the optimized model to generate a second pseudo label, and the second pseudo label accuracy is calculated. The model tuning includes adjusting model parameters and changing model structure.

[0035] Optionally, if the accuracy of the pseudo-label is lower than the preset pseudo-label accuracy threshold when comparing the pseudo-label with the actual label, this means that the prediction ability of the initial current polarity analysis model has not yet met the expected requirements. At this time, the system terminal tunes the model to improve its performance. Model tuning is an iterative process aimed at improving the prediction accuracy of the model. This process includes adjusting the parameters of the model, such as learning rate, batch size, number of iterations, etc., which have an important impact on the training speed and final performance of the model. Subsequently, if the expected requirements are still not met after adjusting the model parameters, it means that the selected model architecture is not suitable for the current task, and the system terminal will change the structure of the model, for example, adding or reducing hidden layers, changing activation functions, etc. After the model architecture is changed, the tuned model is used again to predict the first historical current signal to generate a second pseudo-label. After that, the second pseudo-label is compared with the actual label again, and the accuracy of the second pseudo-label is calculated. If the accuracy of the second pseudo-label reaches the preset threshold, it means that the model tuning is successful and the prediction ability of the model has been improved. At this time, the second pseudo-label can be added to the training set as labeled data to further train the model to improve its generalization ability. Otherwise, the above process is repeated until the expected requirements are met.

[0036] Step S300: inputting the first current signal into the current polarity analysis model, performing polarity prediction, obtaining predicted polarity, performing polarity verification in combination with actual polarity, and generating a first polarity verification result;

[0037] In one embodiment, the system terminal inputs the current signal to be analyzed, i.e., the first current signal, into a trained current polarity analysis model. The model uses the parameters and weights learned during its training process to calculate the output value through the forward propagation algorithm to obtain a predicted current polarity result, i.e., the predicted polarity. In order to verify the accuracy of the model prediction, the system terminal performs a polarity check on the predicted polarity result and the actual polarity. By comparing whether the predicted polarity and the actual polarity are completely consistent, a first polarity check result is generated. If all are consistent, it means that the current transformer has no polarity error; if there is an inconsistency, it means that the current transformer has a polarity error.

[0038] Step S400: Decomposing the first polarity check result, and traversing the polarity error cause database based on the decomposition result to obtain the first polarity error cause;

[0039] In one embodiment, after obtaining the first polarity check result, the system terminal identifies from the first polarity check result which current signals have predicted polarities that do not match the actual polarities, i.e., error signals. Subsequently, for each error signal, it is decomposed to obtain multiple fault feature sets. These fault feature sets correspond to the error signals one by one, including information such as the fault location and current changes of the error signals. Afterwards, based on these decomposed features, the system terminal traverses the polarity error cause database. This polarity error cause database contains various known factors that cause polarity errors, such as equipment overload, wiring error, electromagnetic interference, etc., each of which corresponds to a series of specific error features. During the traversal process, the system terminal matches the decomposed features of the error signal with the error cause in the database. This involves comparing the information such as the fault location and current changes of the error signal with the corresponding features described in the database, calculating the similarity between each error cause and the error signal, and outputting the one with the largest similarity as the first polarity error cause. This cause has important guiding significance for subsequent equipment maintenance or adjustment, and can help the system terminal quickly locate and solve the problem that causes the polarity error.

[0040] Further, the step S400 includes:

[0041] Decomposing the first polarity check result to obtain a fault feature set, wherein the fault feature set includes fault location information, current change information, voltage change information, and device status information;

[0042] A similarity match is performed between the fault feature set and the polarity error cause database, and a first polarity error cause is output, wherein the first polarity error cause is the cause with the greatest similarity in the polarity error cause database.

[0043] Preferably, after obtaining the first polarity check result, the system terminal decomposes the first polarity check result in order to identify specific fault features related to the polarity error. In this process, a series of fault features decomposed are sorted to form a fault feature set. This fault feature set contains fault location information, current change information, voltage change information, and device status information. After obtaining the fault feature set, the system terminal uses unique hot encoding to vectorize the fault feature set and the records traversed each time to generate a fault feature vector and a record feature vector. Subsequently, the cosine similarity formula is used to calculate the similarity between the fault feature vector and the record feature vector. All the calculated similarity values ​​are then arranged in descending order, and the cause record with the largest similarity is selected as the cause of the first polarity error. Through this process, the specific cause of the polarity error can be quickly located, providing strong support for subsequent fault repair and preventive measures.

[0044] Furthermore, the present application provides that after outputting the cause of the first polarity error, the method further includes:

[0045] If the similarity of the first polarity error is less than a preset similarity threshold, constructing a current transformer physical model;

[0046] Inputting the fault feature set into the current transformer physical model, performing scene restoration, and obtaining a fault-similar scene;

[0047] Perform polarity error analysis on the fault-similar scenario to obtain the first polarity error cause, and add the first polarity error cause and the fault feature set to the polarity error cause database.

[0048] Optionally, when the similarity of the first polarity error is less than the preset similarity threshold, it means that there is no highly matching record in the existing polarity error cause database to explain the current fault. In this case, in order to diagnose the problem more accurately and update the database, the system terminal constructs a physical model of the current transformer. This model is built based on the actual working principle and real structure of the current transformer to simulate its behavior in the actual environment. Subsequently, the fault feature set extracted from the first polarity check result is input into the current transformer physical model, and the scene of the fault is reproduced by the model to obtain more information about the fault, so that various situations when the fault occurs can be more intuitively observed and analyzed. After that, the polarity error is diagnosed according to the data and phenomena in the fault-similar scene, and the polarity difference of the current transformer output signal under normal conditions and under fault conditions is compared. Combined with professional knowledge such as circuit analysis and electromagnetic field theory, the polarity difference is qualitatively determined to determine the specific cause of the polarity error, that is, the first polarity error cause. Finally, the system terminal adds the first polarity error cause and its corresponding fault feature set to the polarity error cause database. In this way, when a similar fault occurs again in the future, the system terminal can quickly find matching records from the database, improving the efficiency and accuracy of fault diagnosis.

[0049] Step S500: Based on the first polarity error cause, a predetermined correction scheme set is screened, a first correction scheme is extracted, and a code conversion is performed based on the first correction scheme to generate a first correction instruction;

[0050] In one embodiment, after obtaining the cause of the first polarity error, the system terminal loads a predetermined set of correction schemes, which contains correction schemes for various possible faults and errors. Each correction scheme is pre-designed to solve a specific type of problem. Subsequently, the system terminal selects the correction scheme that best matches the cause of the first polarity error from this set, namely the first correction scheme. The screening process is similar to the aforementioned process of obtaining the cause of the first polarity error. The similarity between the cause of the first polarity error and each correction scheme in the predetermined correction scheme set is calculated by cosine similarity, and compared with a preset correction threshold to select the correction scheme that meets the correction threshold. Afterwards, in order to convert the correction scheme into an instruction that can be executed by the motor, the system terminal performs code conversion. This process is to convert the text description, operation steps and other information in the correction scheme into binary codes that the motor can understand by consulting the instruction document of the motor. Through code conversion, the system terminal generates a first correction instruction to achieve rapid correction of the fault.

[0051] Further, the step S500 includes:

[0052] If no matching first correction scheme is screened out, a polarity error knowledge graph is constructed based on domain knowledge and historical correction records, wherein the polarity error knowledge graph includes error causes, influencing factors, correction schemes, and the relationship between them;

[0053] Inferring the cause of the first polarity error using the polarity error knowledge graph to obtain an inference correction solution;

[0054] Based on the current transformer physical model, the inference correction scheme is simulated and tested to generate a second correction scheme, and the second correction scheme and the first polarity error cause are added to the predetermined correction scheme set.

[0055] Preferably, when a correction scheme that meets the current first polarity error cause cannot be screened, the system terminal constructs a polarity error knowledge graph based on domain knowledge, historical correction records and other information. This graph is a structured knowledge base that contains error causes, influencing factors, correction schemes and the associations between them. Specifically. The system terminal identifies entities related to polarity errors from domain knowledge and historical correction records, such as error causes, influencing factors, correction schemes, etc. Then the relationships between these entities are extracted to form a triple structure of "entity-relationship-entity", which constitutes the core of the knowledge graph. Subsequently, a polarity error knowledge graph is constructed based on the extracted triple structure. During the construction process, the architecture of the graph is defined according to actual needs, including the logical architecture and the technical architecture. In terms of the logical architecture, the knowledge graph is divided into a data layer and a mode layer. The data layer stores specific entity and relationship data, while the mode layer defines the structure and constraints of these data. In terms of the technical architecture, a graph database is selected to store and manage the data of the knowledge graph. When the logical architecture and the technical architecture are set, the system terminal constructs a polarity error knowledge graph. After the knowledge graph is constructed, the system terminal uses the knowledge graph to perform reasoning analysis on the cause of the first polarity error. By querying the relevant error causes, influencing factors and existing correction schemes in the graph, a potential correction scheme for the current error, namely the reasoning correction scheme, can be inferred. Afterwards, in order to determine whether the inferred correction scheme is truly effective, the system terminal performs a simulation test on the reasoning correction scheme based on the constructed current transformer physical model. By simulating the working conditions and parameters in the actual environment, observe whether the correction scheme can effectively solve the polarity error problem, and generate a second correction scheme. If the simulation test results show that the second correction scheme is effective, it can be added to the predetermined correction scheme set together with the first polarity error cause. In this way, when encountering similar polarity errors in the future, a suitable scheme can be directly selected from the correction scheme set for processing, thereby improving the efficiency and accuracy of fault handling. In summary, when a suitable correction scheme cannot be directly screened out, the system terminal assists in reasoning out potential correction schemes by constructing a polarity error knowledge graph, and verifies its effectiveness through simulation tests. This can not only solve the current fault problem, but also continuously improve the correction scheme set and improve the intelligent level of fault handling.

[0056] Step S600: transmitting the first calibration instruction to the motor to perform a calibration operation.

[0057] In one embodiment, the system terminal transmits the obtained first correction instruction to the motor so that the motor can perform the corresponding correction operation. Once the motor receives the correction instruction, the motor will perform corresponding operations according to the instructions in the instruction. These operations include adjusting the parameters inside the motor, setting the rotation angle of the motor, etc. Through these operations, the motor can correct the previous error state and return to a normal working state.

[0058] Furthermore, the present application provides real-time monitoring and remote control, and the method also includes:

[0059] Connecting to a remote terminal, transmitting the first current signal and the first polarity verification result to the remote terminal for visual display;

[0060] Based on the first polarity error cause, early warning information is generated, risk level identification is performed on the early warning information, and the risk level and the first correction instruction are transmitted to a remote terminal for management personnel to make decisions and execute.

[0061] Preferably, the system terminal establishes a connection with the remote terminal, and the purpose of doing so is to transmit the first current signal and the first polarity verification result to the remote terminal so that the operator can monitor and analyze the state of the current transformer in real time. Once these data are successfully transmitted to the remote terminal, the remote terminal calls a visualization tool to display the first current signal and the first polarity verification result in the form of graphics and charts. Such a visualization display can not only help operators quickly understand the real-time state of the current transformer, but also help them find potential faults or problems. Subsequently, the system terminal generates an early warning message based on the previously analyzed cause of the first polarity error. This early warning message will clearly indicate the type of error and possible consequences. In order to ensure that the management personnel can make corresponding decisions based on the severity of the early warning information, the system terminal conducts an in-depth analysis of the cause of the first polarity error, and evaluates the possible fault level and the degree of impact on the performance of the current transformer. Afterwards, according to the evaluation results, the early warning information is divided into different risk levels, including low risk, medium risk, high risk, etc. Once the risk level is determined, the system terminal transmits the risk level and the first correction instruction to the remote terminal. In this way, the operator can not only understand the current risk situation, but also obtain specific correction suggestions. Based on this information, operators can determine whether remote control is needed to ensure the safe and stable operation of the current transformer. In summary, this process involves multiple links such as remote monitoring, data transmission, visual display, early warning information generation, and risk level identification. Through these steps, real-time monitoring and fault early warning of current transformers can be achieved, providing operators with timely and accurate information support to help make accurate decisions.

[0062] In summary, the embodiments of the present application have at least the following technical effects:

[0063] The embodiment of the present application collects the input and output signals of the current transformer in real time through the power monitoring equipment, and performs data analysis in combination with the historical current signal. Subsequently, a current polarity analysis model is constructed using semi-supervised learning technology to predict and verify the polarity of the current signal collected in real time. Once a polarity error is found, the verification result is decomposed, and the matching cause is found in the polarity error cause database. After that, the corresponding correction operation is selected from the predetermined correction scheme according to the cause of the error. If a matching correction scheme cannot be found directly, a polarity error knowledge graph is constructed based on domain knowledge and historical correction records, a new correction scheme is obtained through graph reasoning, and its effectiveness is verified through physical model simulation testing. In addition, the entire control process supports real-time monitoring and remote control. By connecting to a remote terminal, the operator can view the current signal, polarity verification results, and warning information in real time, and make decisions and execute according to the risk level and correction instructions. These technical effects jointly solve the technical problems that traditional current transformer polarity verification methods are inefficient, susceptible to human factors, and unable to adapt to the expansion of power system scale and increasing complexity. They achieve the technical effect of intelligent diagnosis and correction of current transformer polarity through semi-supervised learning algorithms, thereby improving the stability and safety of power systems.

[0064] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0066] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A current transformer control method based on semi-supervised learning, characterized in that: The method comprises: Step S100: Interacting with the power monitoring device, extracting the historical current signal, and collecting the input and output signals of the current transformer in real time to obtain the first current signal; Step S200: data marking the historical current signal to obtain a first historical current signal and a first historical marked current signal, and constructing a current polarity analysis model in combination with semi-supervised learning, wherein the data marking is performed based on a division rule; Step S300: inputting the first current signal into the current polarity analysis model, performing polarity prediction, obtaining predicted polarity, performing polarity verification in combination with actual polarity, and generating a first polarity verification result; Step S400: Decomposing the first polarity check result, and traversing the polarity error cause database based on the decomposition result to obtain the first polarity error cause; Step S500: Based on the first polarity error cause, a predetermined correction scheme set is screened, a first correction scheme is extracted, and a code conversion is performed based on the first correction scheme to generate a first correction instruction; Step S600: transmitting the first correction instruction to the motor to perform a correction operation; Wherein, the step S200 includes: Connecting to the expert system, marking the sample current signal, and obtaining the marked sample current signal; Training a classification neural network based on the labeled sample current signal, extracting a decision boundary of the classification neural network, and generating the division rule; Wherein, the step S200 includes: Training an initial current polarity analysis model based on the first historical marker current signal; Predicting the first historical current signal using the initial current polarity analysis model to generate a pseudo label of the first historical current signal; Calling the expert system to mark the first historical current signal, obtaining an actual label of the first historical current signal, comparing the actual label with the pseudo label, and calculating the accuracy of the pseudo label; If the pseudo-label accuracy is greater than or equal to the pseudo-label accuracy threshold, the pseudo-label is added as labeled data to the first historical labeled current signal, and the initial current polarity analysis model is trained twice to obtain a current polarity analysis model; Wherein, the step S500 includes: If no matching first correction scheme is screened out, a polarity error knowledge graph is constructed based on domain knowledge and historical correction records, wherein the polarity error knowledge graph includes error causes, influencing factors, correction schemes, and the relationship between them; Inferring the cause of the first polarity error using the polarity error knowledge graph to obtain an inference correction solution; Based on the current transformer physical model, the inference correction scheme is simulated and tested to generate a second correction scheme, and the second correction scheme and the first polarity error cause are added to the predetermined correction scheme set.

2. The method according to claim 1, characterized in that: If the pseudo label accuracy is less than the pseudo label accuracy threshold, the initial current polarity analysis model is optimized, the first historical current signal is predicted again using the optimized model to generate a second pseudo label, and the second pseudo label accuracy is calculated. The model tuning includes adjusting model parameters and changing model structure.

3. The method according to claim 1, characterized in that: The step S400 includes: Decomposing the first polarity check result to obtain a fault feature set, wherein the fault feature set includes fault location information, current change information, voltage change information, and device status information; A similarity match is performed between the fault feature set and the polarity error cause database, and a first polarity error cause is output, wherein the first polarity error cause is the cause with the greatest similarity in the polarity error cause database.

4. The method according to claim 3, characterized in that: After outputting the first polarity error cause, the method includes: If the similarity of the first polarity error is less than a preset similarity threshold, constructing a current transformer physical model; Inputting the fault feature set into the current transformer physical model, performing scene restoration, and obtaining a fault-similar scene; Perform polarity error analysis on the fault-similar scenario to obtain the first polarity error cause, and add the first polarity error cause and the fault feature set to the polarity error cause database.

5. The method according to claim 1, characterized in that: Real-time monitoring and remote control are performed, the method comprising: Connecting to a remote terminal, transmitting the first current signal and the first polarity verification result to the remote terminal for visual display; Based on the first polarity error cause, early warning information is generated, risk level identification is performed on the early warning information, and the risk level and the first correction instruction are transmitted to a remote terminal for management personnel to make decisions and execute.

Images