Double-model protection behavior identification method and device for multi-source heterogeneous data of power grid, processor and computer readable storage medium thereof

Abstract

The invention relates to a double-model protection behavior identification method for multi-source heterogeneous data of a power grid, and the method comprises the steps: data standardization and preprocessing: converting the multi-source heterogeneous data required by the behavior identification of a transformer substation into an original standard sample set OriData suitable for the processing of an XGBoost model by using a relay protection semantic standardization model Standard Bert based on Bert; a power grid protection behavior identification algorithm based on an XGBoost model is used for training the original standard sample set OriData, and a power grid protection behavior identification model RelayActRecovery meeting the error requirement is obtained; and the semantic standardization model Standard Bert and the power grid protection behavior identification model RelayActRegeneration are utilized to complete double-model behavior identification and classification processing aiming at the protection equipment in the power grid. By the adoption of the double-model protection behavior identification method for the multi-source heterogeneous data of the power grid, semantic standardization processing can be conducted on uploading information of different plants and stations, different manufacturers and different devices, and therefore data used for artificial intelligence model analysis can be obtained.

Description

Technical Field

[0001] The present invention relates to the field of large model processing, in particular to the field of smart grids, and specifically refers to a dual-model protection behavior identification method, device, processor and computer-readable storage medium thereof for multi-source heterogeneous data of power grids. Background Art

[0002] With the increasing research and development efforts in smart grids in recent years, particularly the implementation of projects such as the State Grid Corporation of China's "SG186" and China Southern Power Grid's "Digital Southern Power Grid," the development direction of building a "unified, robust smart grid" has been further established. The introduction of mature and efficient artificial intelligence algorithms into the smart grid field has also gradually attracted the attention of scholars and research institutions.

[0003] Typically, the designated operation of a single device generates a defined sequence of events. However, in the massive amount of alarm information from the master station, the regularity of the time sequence may not be apparent due to factors such as inconsistent intervals between station operations, repeated transmissions of information between master and substations, and asynchronous information between master and standby devices. This makes manual monitoring and management extremely difficult. Currently, there is no effective behavior identification method. Therefore, designing an efficient and accurate method for identifying grid protection behavior has strong practical and theoretical significance. Leveraging artificial intelligence technology, we construct a dedicated protection model to match signals, infer and identify the maintenance status of protection equipment and on-site operational behavior, and replace the original decentralized monitoring signals with integrated signals, greatly reducing the pressure on monitoring panels and laying the foundation for on-site operation supervision.

[0004] For general high-dimensional data input, data dimensionality reduction is usually performed using named mapping. However, in the task of identifying power grid protection behavior, the master station receives a large amount of information with diverse types and different naming, making variable preprocessing difficult.

[0005] The current state of the art in related fields is that input data often contains time series information, so neural network-based RNN or LSTM models, which can process time series data, are often used. However, in the task of identifying power grid protection behavior, due to various factors such as inconsistent station operation intervals and asynchronous information between primary and standby units, extracting time series features becomes extremely difficult. Therefore, time series models cannot be used as the identification model for power grid protection behavior. Summary of the Invention

[0006] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a dual-model protection behavior identification method, device, processor and computer-readable storage medium thereof for multi-source heterogeneous data of a power grid.

[0007] To achieve the above objectives, the present invention provides a dual-model protection behavior identification method, device, processor, and computer-readable storage medium for multi-source heterogeneous data of a power grid as follows:

[0008] The dual-model protection behavior identification method for multi-source heterogeneous data of a power grid is characterized in that the method comprises the following steps:

[0009] (1) Data standardization and preprocessing: StandardBert, a semantic standardization model for relay protection based on Bert, is used to convert the multi-source heterogeneous data required for substation behavior identification into the original standard sample set OriData suitable for processing by the XGBoost model;

[0010] (2) Using the grid protection behavior recognition algorithm based on the XGBoost model to train the original standard sample set OriData, a grid protection behavior recognition model RelayActRecognition that meets the error requirements is obtained;

[0011] (3) Using the semantic standardization model StandardBert and the grid protection behavior recognition model RelayActRecognition, the dual-model behavior recognition and classification processing for the protection equipment in the grid is completed.

[0012] Preferably, the step (1) comprises:

[0013] (1.1) Collect existing device point meter information data from different sources, classify the information sent by different devices from different manufacturers, and train the BERT-based relay protection semantic standardization model StandardBert;

[0014] (1.2) Using the semantic standardization large model StandardBert to process the simulation data, semantically standardize the non-standardized action, alarm, switch position change, and recording notification data to obtain standardized data StandardData;

[0015] (1.3) Count the displacement or upload frequency of each variable data, manually select the appropriate dimension for dimensionality reduction, and use the PCA-based data dimensionality reduction method to reduce the dimension of the standardized data StandardData to obtain the original standard sample set OriData consisting of standardized variable names, variable displacement upload frequencies, and behavior identification labels.

[0016] Preferably, the step (2) includes:

[0017] (2.1) XGBoost model initialization: Set the maximum number of iterations Tmax, learning rate η, maximum depth max_depth, subsample ratio, column sampling ratio, and regularization parameter of the XGBoost algorithm;

[0018] (2.2) Initial sample allocation weight: Set the initial weight ω(i) of each data sample in the original standard sample set OriData, i = 1, 2, ..., k, where k is the total number of samples in the original standard sample set OriData, and associate the initial weight ω(i) with each sample in the original standard sample set OriData to form a weighted sample set WeightedOriData;

[0019] (2.3) Training the XGBoost model: Using the original standard sample set OriData to train the XGBoost model, and in each iterative training, sampling is performed from the weighted sample set WeightedOriData according to the current initial weight ω(i) to form a new training sample set SampledData, thereby training a new decision tree-based weak learning model hj, and calculating the predicted value of the current weak learning model hj;

[0020] (2.4) Calculating the error εj of the weak learning model hj: In each iteration, the error εj of the weak learning model hj on the current sample set is calculated;

[0021] (2.5) Update sample weight: adjust the current sample weight ω according to the error εj j (i), and the current weight ω j (i) Perform normalization so that the sum of all sample weights is equal to 1;

[0022] (2.6) Repeat steps (2.3) to (2.5) until the maximum number of iterations Tmax is reached or the preset stopping condition is met;

[0023] (2.7) Combined weak learning model: The weak learning models hj obtained from all iterations are weightedly combined to form the final strong learning model H(x), and the grid protection behavior recognition model RelayActRecognition that can receive multi-source heterogeneous data input is obtained.

[0024] Preferably, the step (2.5) is specifically as follows:

[0025] According to the error εj, the current weight ω of the sample is adjusted using an exponential weighting method. j (i)

[0026] ω j+1 (i) = ω j (i)·exp(α·error(i))

[0027] Among them, ω j (i) represents the weight of the i-th sample in the j-th iteration, ωj+1 (i) is the weight of the sample after the j+1th iteration update, α is the importance coefficient of the weak learner, reflecting its classification performance, and error(i) is the prediction error of the weak learner on the i-th sample;

[0028] And all sample weights are normalized as follows:

[0029]

[0030] Make the sum of all sample weights equal to 1.

[0031] Preferably, the step (2.7) is specifically to construct the final strong learning model H(x) in the following manner:

[0032]

[0033] Among them, α j represents the weight of the j-th weak classifier (i.e., the tree model trained for the j-th iteration), h j (x) is the predicted output of the jth weak classifier for sample x.

[0034] Preferably, the step (3) includes:

[0035] (3.1) Using the protection information received in any period as the input of the grid protection behavior recognition model RelayActRecognition, and preprocessing the information using the semantic standardization model StandardBert to obtain standardized input data for the grid protection behavior recognition model RelayActRecognition;

[0036] (3.2) using the grid protection behavior recognition model RelayActRecognition to identify the behavior scenario of the pre-processed uploading information and obtain the recognition result;

[0037] (3.3) Add the identification results into the original standard sample set OriData to complete the identification and classification process.

[0038] The dual-model protection behavior identification device for multi-source heterogeneous data of power grid has the following main features:

[0039] a processor configured to execute computer-executable instructions;

[0040] The memory stores one or more computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned dual-model protection behavior identification method for multi-source heterogeneous data of the power grid are implemented.

[0041] The dual-model protection behavior identification processor for multi-source heterogeneous data of the power grid has the main feature that the processor is configured to execute computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned dual-model protection behavior identification method for multi-source heterogeneous data of the power grid are implemented.

[0042] The main feature of the computer-readable storage medium is that a computer program is stored thereon, and the computer program can be executed by a processor to implement the steps of the above-mentioned dual-model protection behavior identification method for multi-source heterogeneous data of the power grid.

[0043] The dual-model protection behavior identification method, device, processor, and computer-readable storage medium for multi-source heterogeneous data in the power grid, as described in this invention, can identify behavioral scenarios such as protection terminal setting modification and periodic inspections. It also utilizes an auxiliary fault analysis program to intelligently identify faults based on specific behavioral scenarios, effectively reducing the false alarm rate of fault reports. Furthermore, the use of artificial intelligence algorithms can greatly improve computing efficiency and classification accuracy, significantly advancing the industry's current status and possessing significant industrial value. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of the dual-model protection behavior identification method for multi-source heterogeneous data of the power grid according to the present invention. DETAILED DESCRIPTION

[0045] In order to more clearly describe the technical content of the present invention, further description is given below in conjunction with specific embodiments.

[0046] Before describing in detail embodiments according to the present invention, it should be noted that, hereinafter, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, whereby a process, method, article, or apparatus comprising a list of elements includes not only those elements, but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0047] See also Figure 1 As shown, the dual-model protection behavior identification method for multi-source heterogeneous data of the power grid specifically includes the following steps:

[0048] Step 1: Data standardization and preprocessing: Use the Bert-based relay protection semantic standardization model StandardBert to convert the multi-source heterogeneous data required for substation behavior identification into the original standard sample set OriData suitable for XGBoost model processing.

[0049] This data is preprocessed and standardized, transforming the multi-source, heterogeneous power grid data into a unified, standardized format to ensure data consistency and operability. Using StandardBert, a BERT-based semantic standardization model for relay protection, non-normalized data from different manufacturers and devices is semantically standardized to generate a unified, standardized dataset. Feature extraction and dimensionality reduction are then used to reduce the data's dimensionality, retaining the most representative and influential key information. This improves data quality and effectiveness, ultimately enabling efficient and accurate identification of power grid protection behaviors.

[0050] As a preferred embodiment of the present invention, the data preprocessing process mainly includes:

[0051] Step 1.1: Collect existing data, classify different equipment information, and train the relay protection semantic standardization model StandardBert.

[0052] Collect historical data sent from different vendors and devices. This data may include various types of signals, such as device actions, alarms, switch position changes, and waveform recording notifications. Categorize the collected data and perform preliminary organization based on criteria such as device type and signal type. Train StandardBert, a standardized large-scale model for relay protection semantics based on BERT, to convert data from different vendors and devices into a unified semantic representation.

[0053] Step 1.2: Use the semantic standardization model StandardBert to process the simulation data, perform semantic standardization, and obtain standardized data StandardData.

[0054] The sorted historical data is input into the semantic standardization large model StandardBert, and semantic standardization is performed on non-standardized actions, alarms, switch changes, and recording notifications. Through the semantic standardization model, a standardized dataset StandardData with a unified format is generated. Ensure that the data after semantic standardization is consistent in format and content, which facilitates subsequent processing and analysis. Specifically, the acquired semantically standardized data is tokenized and a training set Json is formed. Tokenizer processing refers to encoding the original text through the built-in word segmenter of the pre-trained language model (such as Bert) to obtain an input format acceptable to the model (such as input_ids, attention_mask, token_type_ids) to capture semantic features. The Json format is a widely used data storage format in the current construction of training datasets due to its clear structure, easy reading and writing, and strong cross-platform compatibility. It is convenient for training frameworks (such as PyTorch and TensorFlow) to load data.

[0055] Step 1.3: Count the frequency of variable changes and use the PCA-based data dimensionality reduction method to reduce the dimensionality of StandardData to obtain the original standard sample set OriData.

[0056] Count the frequency of each variable's displacement or upload, and analyze the frequency of different signals in the data. Manually select the appropriate dimension for dimensionality reduction, and determine the number of core features to retain based on the actual application scenario and computing resources. Use a PCA (principal component analysis)-based method to reduce the dimensionality of StandardData, extract the main features, and reduce the data dimension. Generate the original standard sample set OriData, which consists of standardized variable names, variable displacement upload frequencies, and behavior identification labels for subsequent model training.

[0057] Step 2: Use the power grid protection behavior recognition algorithm based on the XGBoost model to train the original standard sample set OriData to obtain the power grid protection behavior recognition model RelayActRecognition that meets the error requirements.

[0058] The XGBoost model, or eXtreme Gradient Boosting, is an efficient, flexible, and scalable machine learning algorithm based on a gradient boosting framework. It excels in computational efficiency and predictive performance by improving decision trees, and is capable of processing large-scale data sets and high-dimensional data. The XGBoost model supports custom objective functions and evaluation indicators, and has built-in mechanisms for handling missing values ​​and L1 and L2 regularization, which enhance the robustness and generalization capabilities of the model and avoid overfitting. In the present invention, the XGBoost model is selected as the grid protection behavior identification algorithm because it can effectively process high-dimensional, multi-source heterogeneous grid protection behavior data and capture the complex relationships and features in the data. The XGBoost model performs well in processing large amounts of data generated by the power grid system, and can complete training and prediction in a relatively short time, thereby improving the accuracy of grid protection behavior identification. Its automatic processing of missing values ​​simplifies the data preprocessing process, while distributed computing support ensures efficient processing in a cluster environment. Through standardized and preprocessed data, the XGBoost model can better realize the intelligent identification and accurate judgment of grid protection behavior, thereby improving the safety and reliability of grid operation.

[0059] As a preferred embodiment of the present invention, the step 2 specifically includes the following steps:

[0060] Step 2.1: Initialize the XGBoost model. Set the maximum number of iterations Tmax, learning rate η, maximum depth max_depth, subsample ratio, column sampling ratio, regularization parameters, etc. of the XGBoost algorithm, including:

[0061] Set the maximum number of iterations Tmax: Define the maximum number of iterations of the XGBoost algorithm, for example, Tmax = 100.

[0062] Set the learning rate η: Select an appropriate learning rate, such as η = 0.1, to control the step size of each iterative update.

[0063] Set the maximum depth max_depth: Set the maximum depth of the decision tree, for example max_depth = 6, to control the complexity of each tree.

[0064] Set the subsample ratio: define the data subsample ratio used to train each tree, for example, subsample = 0.8, to increase the generalization ability of the model.

[0065] Set the column sampling ratio colsample_bytree: Define the feature subset ratio used to train each tree, for example, colsample_bytree=0.8, to further enhance the generalization ability of the model.

[0066] Set regularization parameters: including the L1 regularization parameter alpha and the L2 regularization parameter lambda, for example, alpha = 0.1, lambda = 1.0, to prevent model overfitting.

[0067] Step 2.2: Initialize the sample allocation weights. The specific steps include:

[0068] Initialize sample weights: Assign an initial weight ω(i) to each sample, where the initial weight of each sample is equal, that is, ω(i) = 1 / k, where k is the total number of samples in the original sample set OriData.

[0069] Prepare training data: Associate sample weights with each sample in the original standard sample set OriData to form a weighted sample set WeightedOriData. Specifically, each sample in the initial sample set OriData is resampled according to the current weight ω(i) to generate WeightedOriData for training the new weak classifier.

[0070] Step 2.3: Train the XGBoost model. The specific steps include:

[0071] In each iteration, sampling is performed from the weighted sample set WeightedOriData according to the current sample weight ω(i) to form a new training sample set SampledData.

[0072] Use SampledData to train a new decision tree based weak learning model hj.

[0073] Calculate the predicted value of the current weak learning model hj.

[0074] Step 2.4: Calculate the error εj of the weak learning model hj. In each iteration, calculate the error εj of the weak learning model hj on the current sample set;

[0075] Step 2.5: Update sample weights. The specific steps include:

[0076] Adjust the sample weight ω using exponential weighting according to the error εj j (i)

[0077] ω j+1 (i) = ω j (i)·exp(α·error(i))

[0078] Among them, ω j (i) represents the weight of the i-th sample in the j-th iteration, ω j+1 (i) is the weight of the sample after the j+1th iteration update, α is the importance coefficient of the weak learner, reflecting its classification performance, and error(i) is the prediction error of the weak learner on the i-th sample;

[0079] Normalize the weights of all samples so that the sum of all sample weights is equal to 1:

[0080]

[0081] Step 2.6: Start iteration. Repeat steps 2.3 to 2.5 until the maximum number of iterations Tmax is reached or the preset stopping condition is met, such as the error εj no longer changes significantly or reaches a predetermined error threshold.

[0082] Step 2.7: Combine weak learning models. All iteratively obtained weak learning models hj are weighted and combined to form the final strong learning model H(x). This results in the RelayActRecognition grid protection behavior recognition model based on the XGBoost algorithm that can accept multi-source heterogeneous data input. This is done using a linear weighting approach:

[0083]

[0084] Among them, α jrepresents the weight of the j-th weak classifier (i.e., the tree model trained for the j-th iteration), which is usually determined based on the error of each model. j (x) is the predicted output of the jth weak classifier for sample x.

[0085] Step 3: Use the semantic standardization model StandardBert and the grid protection behavior recognition model RelayActRecognition to complete the behavior recognition and classification of the protection equipment in the grid.

[0086] By leveraging the semantic standardization model StandardBert and the grid protection behavior identification model RelayActRecognition, accurate identification and classification of protection device behavior in the power grid are achieved. First, the StandardBert model preprocesses protection device information sent over any time period, standardizing data from different vendors and devices to generate consistent input data. This step ensures data consistency and standardization, making subsequent processing more efficient. Next, the RelayActRecognition model performs behavioral scenario recognition on the standardized data, accurately identifying the specific behavior of protection devices through efficient algorithmic analysis. The recognition results are added to the original sample set OriData, which not only facilitates continuous optimization and improvement of the model but also provides reliable data support for grid operation and management. This process improves the accuracy and efficiency of grid protection behavior identification, reduces the burden of manual monitoring, and enhances the safety and reliability of grid operation. This combination of two models can better address the complex and changing behavior of grid protection devices and ensure the stable operation of the power grid system.

[0087] As a preferred embodiment of the present invention, the step 3 specifically includes the following steps:

[0088] Step 3.1: Use the protection information sent during any period as input and pre-process it using the semantic standardization model StandardBert to obtain standardized input data for the grid protection behavior recognition model RelayActRecognition. This means acquiring and recognizing the information sent in unknown scenarios. Once correctly recognized, it is collected into the original standard sample set to facilitate subsequent training and improve recognition accuracy.

[0089] Step 3.2: Using the data preprocessed by the semantic standardization model as input, the grid protection behavior recognition model RelayActRecognition is used to perform behavior scenario recognition and obtain the recognition results;

[0090] Step 3.3: Add the recognition results to the original standard sample set OriData.

[0091] After the training and learning of the above three steps, a dual-model protection behavior identification method for multi-source heterogeneous data of the power grid is formed, thereby realizing the intelligent identification of protection behavior scenarios.

[0092] In practical application, this technical solution has the following technical effects:

[0093] The dual-model protection behavior identification method for multi-source heterogeneous data in power grids, provided by this invention, utilizes artificial intelligence algorithms for on-site protection behavior identification. This method addresses the difficulty of inputting massive amounts of information and inconsistent naming by utilizing state-of-the-art Transformer-based BERT large-scale model technology. Combined with input data before and after naming standardization, this method trains a semantically standardized large-scale model for relay protection that can receive and process massive amounts of actions, alarms, switch position changes, and recording notifications, while also performing semantic standardization. This large-scale model can semantically standardize information transmitted from different plants, stations, manufacturers, and devices, thereby generating data suitable for AI model analysis.

[0094] Artificial intelligence algorithms are used to identify on-site protection behaviors. The solution to the unavailability of time series information and time series models is to consider sending the variable frequency after semantic standardization as input. Compared with the accumulated statistical number as input, using frequency as input means that the input range of the model does not need to be fixed, avoiding the problem of different durations of different behavioral scenarios.

[0095] In addition, the XGBoost algorithm used in this technical solution has higher computational efficiency and classification accuracy than commonly used machine learning algorithms such as random forest and gradient boosting tree. It is a more advanced and high-performance algorithm in the field of artificial intelligence.

[0096] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0097] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device.

[0098] Those skilled in the art will understand that all or part of the steps of the method for implementing the above-mentioned embodiment can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0099] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0100] Throughout this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "embodiment" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0101] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

[0102] The dual-model protection behavior identification method, device, processor, and computer-readable storage medium for multi-source heterogeneous data in the power grid, as described in this invention, can identify behavioral scenarios such as protection terminal setting modification and periodic inspections. It also utilizes an auxiliary fault analysis program to intelligently identify faults based on specific behavioral scenarios, effectively reducing the false alarm rate of fault reports. Furthermore, the use of artificial intelligence algorithms can greatly improve computing efficiency and classification accuracy, significantly advancing the industry's current status and possessing significant industrial value.

[0103] In this specification, the present invention has been described with reference to specific embodiments thereof. However, it will be apparent that various modifications and variations may be made without departing from the spirit and scope of the present invention. Accordingly, the specification and drawings are to be regarded as illustrative rather than restrictive.

Claims

1. A dual-model protection behavior identification method for multi-source heterogeneous data of power grid, characterized by: The method comprises the following steps: (1) Data standardization and preprocessing: StandardBert, a semantic standardization model for relay protection based on Bert, is used to convert the multi-source heterogeneous data required for substation behavior identification into the original standard sample set OriData suitable for processing by the XGBoost model; (2) Using the grid protection behavior recognition algorithm based on the XGBoost model to train the original standard sample set OriData, a grid protection behavior recognition model RelayActRecognition that meets the error requirements is obtained; (3) Using the semantic standardization model StandardBert and the grid protection behavior recognition model RelayActRecognition, the dual-model behavior recognition and classification processing for the protection equipment in the grid is completed.

2. The dual-model protection behavior identification method for multi-source heterogeneous data of power grid according to claim 1 is characterized in that: The step (1) comprises: (1.1) Collect existing device point meter information data from different sources, classify the information sent by different devices from different manufacturers, and train the BERT-based relay protection semantic standardization model StandardBert; (1.2) Using the semantic standardization large model StandardBert to process the simulation data, semantically standardize the non-standardized action, alarm, switch position change, and recording notification data to obtain standardized data StandardData; (1.3) Count the displacement or upload frequency of each variable data, manually select the appropriate dimension for dimensionality reduction, and use the PCA-based data dimensionality reduction method to reduce the dimension of the standardized data StandardData to obtain the original standard sample set OriData consisting of standardized variable names, variable displacement upload frequencies, and behavior identification labels.

3. The dual-model protection behavior identification method for multi-source heterogeneous data of power grid according to claim 1 is characterized in that: The step (2) comprises: (2.1) XGBoost model initialization: Set the maximum number of iterations Tmax, learning rate η, maximum depth max_depth, subsample ratio subsample, column sampling ratio colsample_bytree, and regularization parameters of the XGBoost algorithm; (2.2) Initial sample allocation weight: Set the initial weight ω(i) of each data sample in the original standard sample set OriData, i = 1, 2, ..., k, where k is the total number of samples in the original standard sample set OriData, and associate the initial weight ω(i) with each sample in the original standard sample set OriData to form a weighted sample set WeightedOriData; (2.3) Training the XGBoost model: Using the original standard sample set OriData to train the XGBoost model, and in each iterative training, sampling is performed from the weighted sample set WeightedOriData according to the current initial weight ω(i) to form a new training sample set SampledData, thereby training a new decision tree-based weak learning model hj, and calculating the predicted value of the current weak learning model hj; (2.4) Calculating the error εj of the weak learning model hj: In each iteration, the error εj of the weak learning model hj on the current sample set is calculated; (2.5) Update sample weight: adjust the current sample weight ω according to the error εj j (i), and the current weight ω j (i) Perform normalization so that the sum of all sample weights is equal to 1; (2.6) Repeat steps (2.3) to (2.5) until the maximum number of iterations Tmax is reached or the preset stopping condition is met; (2.7) Combined weak learning model: The weak learning models hj obtained from all iterations are weightedly combined to form the final strong learning model H(x), and the grid protection behavior recognition model RelayActRecognition that can receive multi-source heterogeneous data input is obtained.

4. The dual-model protection behavior identification method for multi-source heterogeneous data of power grid according to claim 3 is characterized in that: The step (2.5) is specifically as follows: According to the error εj, the current weight ω of the sample is adjusted using an exponential weighting method. j (i) ω j+1 (i)=ω j (i)·exp(α·error(i)) Among them, ω j (i) represents the weight of the i-th sample in the j-th iteration, ω j+1 (i) is the weight of the sample after the j+1th iteration update, α is the importance coefficient of the weak learner, reflecting its classification performance, and error(i) is the prediction error of the weak learner on the i-th sample; And all sample weights are normalized as follows: Make the sum of all sample weights equal to 1.

5. The dual-model protection behavior identification method for multi-source heterogeneous data of power grid according to claim 3 is characterized in that: The step (2.7) is specifically to construct the final strong learning model H(x) in the following manner: Among them, α j represents the weight of the j-th weak classifier (i.e., the tree model trained for the j-th iteration), h j (x) is the predicted output of the jth weak classifier for sample x.

6. The dual-model protection behavior identification method for multi-source heterogeneous data of power grid according to claim 3 is characterized in that: The step (3) comprises: (3.1) Using the protection information received in any period as the input of the grid protection behavior recognition model RelayActRecognition, and preprocessing the information using the semantic standardization model StandardBert to obtain standardized input data for the grid protection behavior recognition model RelayActRecognition; (3.2) using the grid protection behavior recognition model RelayActRecognition to identify the behavior scenario of the pre-processed uploading information and obtain the recognition result; (3.3) Add the identification results into the original standard sample set OriData to complete the identification and classification process.

7. A dual-model protection behavior identification device for multi-source heterogeneous data of power grid, characterized by: The device comprises: a processor configured to execute computer-executable instructions; A memory storing one or more computer executable instructions, wherein when the computer executable instructions are executed by the processor, the steps of the dual-model protection behavior identification method for multi-source heterogeneous data of a power grid according to any one of claims 1 to 6 are implemented.

8. A dual-model protection behavior identification processor for multi-source heterogeneous data of power grid, characterized by: The processor is configured to execute computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the dual-model protection behavior identification method for multi-source heterogeneous data of a power grid according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that A computer program is stored thereon, and the computer program can be executed by a processor to implement the steps of the dual-model protection behavior identification method for multi-source heterogeneous data of a power grid according to any one of claims 1 to 6.

Images