Collection method and system based on power grid data collector
By deploying data collectors in the power grid, acquiring and analyzing multiple grid parameter data, and using Linear Transformer model for fault prediction, the traditional grid monitoring method has solved the shortcomings in processing speed, accuracy and adaptability, and achieved more efficient and reliable grid fault detection and early warning.
Patent Information
- Application Number
- CN202510193328.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-21
- Publication Date
- 2025-06-27
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional grid monitoring methods have obvious shortcomings in processing speed, accuracy and adaptability to complex environments, making it difficult to efficiently and accurately monitor grid status and fault prediction.
The acquisition method based on the power grid data collector is adopted, and the important data is determined based on the importance weight and priority by acquiring multiple types of power grid parameter data, and inputting it into the pre-trained Linear Transformer fault type prediction model to perform fault type prediction.
It realizes a more comprehensive use of multi-source data in the power grid for comprehensive analysis, improves the accuracy and reliability of fault detection, and improves the accuracy and fault warning capabilities of power grid operating status monitoring.
Smart Images
Figure CN120217072A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of smart grids, and in particular, to a data acquisition method and system based on a power grid data collector. Background Art
[0002] With the development of the global economy and the progress of technology, the power system, as one of the infrastructures of modern society, its reliability and stability are crucial for ensuring the normal operation of the national economy and social life. As an advanced form of power network, the smart grid realizes the intelligent monitoring, analysis, control, and decision-making support of the power grid by integrating modern information and communication technologies and traditional power engineering technologies. However, in the face of the increasingly complex power grid structure and the growing power demand, how to efficiently and accurately monitor the power grid status and predict faults has become an important research topic.
[0003] However, traditional power grid monitoring methods mainly rely on manual inspections and simple automation devices, and these methods have obvious deficiencies in terms of processing speed, accuracy, and adaptability to complex environments. Summary of the Invention
[0004] In view of the above-mentioned disadvantages and deficiencies of the prior art, the present invention provides a data acquisition method and system based on a power grid data collector.
[0005] To achieve the above object, the main technical solutions adopted by the present invention include:
[0006] In a first aspect, an embodiment of the present invention provides a data acquisition method based on a power grid data collector, including:
[0007] S100. Obtain the power grid data collected by the data collectors deployed in the power grid;
[0008] The power grid data includes various types of power grid parameter data;
[0009] S200. Determine important data in the power grid data based on the pre-obtained importance weights and the pre-set priorities corresponding to each type of power grid parameter data;
[0010] S300. Input the important data into a pre-trained fault type prediction model to determine the fault type of the power grid;
[0011] The fault type prediction model is a Linear Transformer model;
[0012] Among them, the pre-set fault type prediction model is trained with a first training data set to obtain a trained fault type prediction model;
[0013] The first training data set includes: important data corresponding to multiple different time points within a historical time period, where each piece of important data in the first training data set is corresponding to a preset fault type identifier.
[0014] Preferably, the power grid data includes: voltage level, current intensity, operating frequency of the power grid, and power grid load rate.
[0015] Preferably, before S100, the method further includes:
[0016] S001. According to the power grid data collected by the data collector deployed in the power grid at a preset frequency within a preset first time period and a pre-trained deep network learning model, obtain the importance weight corresponding to each type of power grid parameter data in the power grid data;
[0017] Among them, the pre-set deep network learning model is trained with a second training data set to obtain a trained deep network learning model;
[0018] The second training data set includes: N samples and preset fault type identifiers respectively corresponding to the samples one by one;
[0019] Among them, the i-th sample includes: the power grid data collected by the data collector deployed in the power grid for the i-th time within the first time period, i ∈ N.
[0020] Preferably, S001 specifically includes:
[0021] S001-1. Use the DeepExplainer method in the SHAP library to create an explainer object, and pass the trained deep network learning model and the second training data set into the created parser object;
[0022] S001-2. Use the created explainer object to calculate the SHAP value for each sample in the second training data set, and obtain the SHAP value corresponding to each type of power grid parameter data in each sample;
[0023] S001-3. Based on the SHAP values corresponding to each type of power grid parameter data in each sample, obtain the importance weight corresponding to each type of power grid parameter data.
[0024] Preferably, S001-3 specifically includes:
[0025] Based on the SHAP values corresponding to each type of power grid parameter data in each sample, use formula (1) to obtain the importance weight corresponding to each type of power grid parameter data;
[0026] The formula (1) is:
[0027]
[0028] W j represents the importance weight corresponding to the grid parameter data of the j-th type;
[0029] SHAP ij represents the SHAP value of the grid parameter data of the j-th type in the i-th sample;
[0030] M is the total number of types of grid parameter data;
[0031] k is used to represent traversing all types of grid parameter data.
[0032] Preferably, the S200 specifically includes:
[0033] S200-1. Based on the pre-acquired importance weight corresponding to each type of grid parameter data and the preset priority, use formula (2) to determine the score corresponding to each type of grid parameter data;
[0034] The formula (2) is:
[0035]
[0036] S j is the score of the grid parameter data of the j-th type;
[0037] P j is the priority of the grid parameter data of the j-th type;
[0038] a is the first adjustment parameter;
[0039] b is the second adjustment parameter;
[0040] e is a natural number;
[0041] S200-2. Based on the scores corresponding to each type of grid parameter data, determine important data in the grid data;
[0042] The important data is the grid parameter data of the H types with the largest scores.
[0043] Preferably, after the S300, it further includes:
[0044] S400. Estimate the uncertainty of the predicted fault type to obtain the uncertainty score of each fault type;
[0045] S500. Based on the uncertainty score of each fault type and the preset threshold, determine whether to send a warning message.
[0046] Preferably, the S400 specifically includes:
[0047] Estimate the uncertainty of the predicted fault type, and calculate the uncertainty score of each fault type using formula (3);
[0048] The formula (3) is as follows:
[0049] U x = -p x log(p x );
[0050] where U x is the uncertainty score of the x-th fault type;
[0051] p x is the probability value of the x-th fault type predicted by the fault type prediction model.
[0052] Preferably, the S500 specifically includes:
[0053] When the uncertainty score of the fault type is greater than a preset threshold, a warning message is determined to be issued.
[0054] On the other hand, this embodiment also provides an acquisition system based on a power grid data collector, including:
[0055] At least one processor; and
[0056] At least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor can execute an acquisition method based on a power grid data collector as described in the first aspect by invoking the program instructions.
[0057] The beneficial effects of the present invention are:
[0058] An acquisition method based on a power grid data collector of the present invention, by acquiring various types of power grid parameter data collected by data collectors deployed in the power grid, and determining important data based on the pre-acquired importance weights and preset priorities corresponding to each type of power grid parameter data, and then inputting this important data into a pre-trained fault type prediction model (LinearTransformer model) to determine the fault type of the power grid. Compared with the prior art, it can more comprehensively utilize multi-source data in the power grid for comprehensive analysis, achieving the purpose of improving the accuracy and reliability of fault detection.
[0059] A data acquisition method based on a power grid data collector according to the present invention. Since it is clear that the power grid data includes various key parameters such as voltage level, current intensity, operating frequency of the power grid, and power grid load rate, compared with the prior art, it can accurately identify potential problems in the power grid by monitoring changes in these core indicators, achieving the effect of improving the monitoring accuracy of the power grid operation status and the fault warning ability.
[0060] A data acquisition method based on a power grid data collector according to the present invention. Since it also includes, before step S100, acquiring the power grid data collected by the data collector deployed in the power grid at a preset first moment, and obtaining the importance weights corresponding to each type of power grid parameter data in the power grid data according to the power grid data collected by the data collector deployed in the power grid at a preset frequency within a preset first time period and a pre-trained deep network learning model, where the deep network learning model is trained using a second training data set, compared with the prior art, it can dynamically adjust the importance weights of each type of power grid parameter data according to historical data, achieving the effect of adaptively optimizing the data processing strategy and enhancing the robustness of the fault prediction model. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 It is a flowchart of a data acquisition method based on a power grid data collector according to the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0062] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the drawings and through specific embodiments.
[0063] In order to better understand the above technical solutions, the exemplary embodiments of the present invention will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more clearly and thoroughly understood, and the scope of the present invention can be completely conveyed to those skilled in the art.
[0064] Embodiment 1
[0065] See Figure 1 , this embodiment provides a data acquisition method based on a power grid data collector, including:
[0066] S100. Acquire the power grid data collected by the data collector deployed in the power grid;
[0067] The grid data includes various types of grid parameter data; specifically, the grid data includes: voltage level, current intensity, operating frequency of the grid, and grid load rate. These parameters are important indicators for evaluating the operating state of the grid. By monitoring the changes in these parameters in real time, potential fault hazards can be detected in a timely manner.
[0068] S200. Determine important data in the grid data based on the pre-obtained importance weights corresponding to each type of grid parameter data and the preset priorities.
[0069] S300. Input the important data into a pre-trained fault type prediction model to determine the fault type of the grid.
[0070] The fault type prediction model is a Linear Transformer model.
[0071] Among them, the pre-set fault type prediction model is trained using the first training data set to obtain the trained fault type prediction model.
[0072] The first training data set includes: important data corresponding to multiple different time points within a historical time period. Among them, each important data in the first training data set corresponds to a pre-set fault type identifier, which is convenient for the model to learn and predict new fault types.
[0073] In this embodiment, the Linear Transformer model is used to process the important data determined from the grid data and predict the fault type of the grid. Due to the adoption of a linearized self-attention mechanism, the Linear Transformer model can significantly reduce the calculation time while maintaining a high accuracy. This is crucial for real-time monitoring and rapid response to grid faults. Compared with traditional methods, this method can process a large amount of grid parameter data faster and detect potential faults in a timely manner.
[0074] The Linear Transformer model has strong adaptability and can easily integrate different types of data (such as voltage level, current intensity, operating frequency of the grid, grid load rate, etc.). This multi-source data fusion ability helps to comprehensively evaluate the grid state and improve the reliability of fault detection.
[0075] In this embodiment, before S100, the method further includes:
[0076] S001. Obtain the importance weights corresponding to each type of grid parameter data in the grid data according to the grid data collected by the data collector deployed in the grid at a preset frequency within a preset first time period and the pre-trained deep network learning model.
[0077] Among them, the pre-set deep network learning model is trained using the second training data set to obtain the trained deep network learning model;
[0078] In this embodiment, the deep network learning model can be any one of a Multilayer Perceptron (MLP), a Convolutional Neural Networks (CNN), a Transformer model, and an Autoencoder. Regardless of which model is selected, the SHAP (SHapley Additive exPlanations) value can be combined to explain the importance of the model output. The SHAP value is based on the Shapley value concept in game theory and provides a quantitative evaluation of the contribution degree of each feature.
[0079] The second training data set includes: N samples and pre-set fault type identifiers respectively corresponding to the samples one by one;
[0080] Among them, the i-th sample includes: the power grid data collected by the data collector deployed in the power grid for the i-th time within the first time period, where i ∈ N.
[0081] In the actual application of this embodiment, S001 specifically includes:
[0082] S001-1. Create an explainer object using the DeepExplainer method in the SHAP library, and pass the trained deep network learning model and the second training data set into the created parser object;
[0083] S001-2. Use the created explainer object to calculate the SHAP value for each sample in the second training data set, and obtain the SHAP values respectively corresponding to each type of power grid parameter data in each sample. These SHAP values quantify the influence degree of each feature on the prediction result of the sample.
[0084] S001-3. Based on the SHAP values respectively corresponding to each type of power grid parameter data in each sample, obtain the importance weight corresponding to each type of power grid parameter data.
[0085] Specifically, S001-3 specifically includes:
[0086] Based on the SHAP values respectively corresponding to each type of power grid parameter data in each sample, use formula (1) to obtain the importance weight corresponding to each type of power grid parameter data;
[0087] The formula (1) is:
[0088]
[0089] W j represents the importance weight corresponding to the grid parameter data of the j-th type;
[0090] SHAP ij represents the SHAP value of the grid parameter data of the j-th type in the i-th sample;
[0091] M is the total number of types of grid parameter data;
[0092] k is used to represent traversing all types of grid parameter data.
[0093] In this embodiment, formula (1) quantifies the influence degree of each type of grid parameter data on the prediction result by calculating the SHAP value of the grid parameter data of each type in each sample. This helps to identify which parameter data is the most critical for fault prediction.
[0094] The denominator part in formula (1) normalizes the SHAP values of all types of grid parameter data, ensuring a fair weight comparison between different parameter types. This can avoid certain parameter types being overemphasized due to large numerical values. By calculating the average value of the SHAP values over the entire training data set, formula (1) provides the importance weight from a global perspective. This helps to discover generally important parameter types rather than only focusing on individual sample cases.
[0095] In summary, formula (1) quantifies the importance weight of each type of grid parameter data by calculating the SHAP value, thereby helping to identify key parameters. This method not only improves the interpretability of the model but also ensures a fair weight comparison between different parameter types, ultimately enhancing the accuracy and reliability of fault prediction.
[0096] This embodiment can effectively extract key information from a large amount of grid data and use advanced machine learning models for fault prediction, improving the safety and stability of grid operation.
[0097] In this embodiment, the S200 specifically includes:
[0098] S200-1. Based on the pre-obtained importance weight corresponding to each type of grid parameter data and the preset priority, use formula (2) to determine the score corresponding to each type of grid parameter data;
[0099] The formula (2) is:
[0100]
[0101] S j is the score of the grid parameter data of the j-th type;
[0102] P j is the priority of the grid parameter data of the jth type;
[0103] a is the first adjustment parameter;
[0104] b is the second adjustment parameter;
[0105] e is a natural number;
[0106] In this embodiment, the priority and importance weight are both considered in formula (2), ensuring that the scoring depends not only on a single factor but on a comprehensive evaluation. This helps to more comprehensively evaluate different types of grid parameter data. The exponential part in formula (2) introduces a non-linear adjustment, making the change of the scoring under different priorities and importance weights smoother. This non-linear adjustment can better reflect the complex relationships in the actual application scenario. By adjusting parameters a and b, the calculation method of the scoring can be flexibly adjusted according to specific requirements. And the exponential part in the formula ensures that the scoring will not fluctuate too much due to extreme values, improving the stability of the scoring.
[0107] S200-2. For the scoring corresponding to each type of grid parameter data, determine the important data in the grid data;
[0108] The important data are the H types of grid parameter data with the largest scores.
[0109] By screening out the top H types of grid parameter data with the highest scores through the scoring mechanism, the data types that are most critical for fault prediction can be accurately identified. This method helps to focus on the truly important information, avoiding the interference of irrelevant or secondary information, thus improving the accuracy and efficiency of fault prediction. When performing fault prediction, only the data of the top H types with the highest scores need to be processed, which greatly reduces the amount of data to be processed. Reducing the amount of data not only reduces the demand for computing resources but also speeds up the model training and prediction, making real-time monitoring and rapid response possible. Focusing on the key analysis and monitoring of the data of the top H types can optimize the resource allocation, investing more human and material resources in the most critical part, and improving the overall operation and maintenance efficiency. This is particularly important for large-scale smart grids because it can significantly improve the reliability and security of the system without incurring excessive costs.
[0110] The scoring mechanism in this embodiment comprehensively considers the importance weights of each type of power grid parameter data and the preset priorities, ensuring that the scoring does not solely rely on a single factor but is the result of a comprehensive evaluation. This comprehensive evaluation method can more accurately reflect which parameter data is most critical for fault prediction, providing a solid foundation for subsequent selection. By limiting the selection to the H types of data with the highest scores instead of all data, this strategy effectively balances the relationship between accuracy and efficiency. On the one hand, it ensures the quality of the selected data, that is, only the most relevant part is included; on the other hand, it controls the data volume, making the calculation more efficient.
[0111] In this embodiment, after S300, it further includes:
[0112] S400. Estimate the uncertainty of the predicted fault types to obtain the uncertainty scores of each fault type;
[0113] Specifically, S400 includes:
[0114] Estimate the uncertainty of the predicted fault types and calculate the uncertainty scores of each fault type using formula (3);
[0115] The formula (3) is:
[0116] U x =-p x log(p x );
[0117] where U x is the uncertainty score of the x-th fault type;
[0118] p x is the probability value of the x-th fault type predicted by the fault type prediction model.
[0119] By calculating the uncertainty score U x , the uncertainty of the prediction result can be quantified. This helps to identify those fault types with unreliable prediction results, thereby improving the overall reliability of the prediction.
[0120] If the uncertainty score of a certain fault type is relatively high, it indicates that the model is not very confident in predicting this fault type and further attention or additional measures are required.
[0121] In this embodiment, formula (3) uses the concept of information entropy. Information entropy is a standard method for measuring information uncertainty and can accurately reflect the uncertainty of the prediction result. When p x is close to 0 or 1, the uncertainty score is relatively low; when p x is close to 0.5, the uncertainty score is relatively high. This measurement method can intuitively reflect the credibility of the prediction result.
[0122] S500 determines whether to issue a warning message based on the uncertainty score of each fault type and a preset threshold.
[0123] The S500 specifically includes:
[0124] When the uncertainty score of the fault type is greater than the preset threshold, it is determined to issue a warning message. In this embodiment, by adjusting the threshold, the risks of false alarms and missed alarms can be balanced to ensure the warning accuracy in different scenarios. In summary, through uncertainty estimation and threshold setting, the system can effectively improve the reliability of fault prediction, reduce false alarms and missed alarms, optimize resource allocation, and provide better decision support.
[0125] This embodiment also provides a collection system based on a power grid data collector, including: at least one processor; and at least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor can execute a collection method based on a power grid data collector as described in the embodiment by invoking the program instructions.
[0126] Embodiment Two
[0127] With the development of smart grids, real-time monitoring and fault prediction of grid states have become increasingly important. Traditional monitoring methods are difficult to cope with complex grid environments, while Embodiment Two provides a more efficient and accurate solution by combining advanced machine learning models (such as Linear Transformer) and deep network learning models.
[0128] Step 100: First, it is necessary to obtain the importance weights corresponding to each type of grid parameter data in the grid data. This step includes:
[0129] Use the DeepExplainer method in the SHAP library to create an explainer object and pass in a pre-trained deep network learning model and a second training dataset.
[0130] Use the created explainer object to calculate the SHAP values of each sample in the second training dataset to obtain the SHAP values corresponding to each type of grid parameter data in each sample.
[0131] Based on the SHAP values corresponding to each type of grid parameter data in each sample, use formula (1) to calculate the importance weight corresponding to each type of grid parameter data.
[0132] The formula (1) is:
[0133]
[0134] W j represents the importance weight corresponding to the grid parameter data of the j-th type; SHAP ij represents the SHAP value of the grid parameter data of the j-th type in the i-th sample; M is the total number of grid parameter data types; k is used to represent traversing all grid parameter data types.
[0135] In this embodiment, by using the SHAP value to dynamically adjust the importance weight of the grid parameter data, it is ensured that the model can adaptively optimize its performance according to the actual grid operation conditions. This mechanism significantly improves the sensitivity and accuracy of fault detection.
[0136] Step 200: Obtain various types of grid parameter data from the data collectors deployed in the grid, including but not limited to key parameters such as voltage level, current intensity, grid operating frequency, grid load rate, etc.
[0137] Step 300: Based on the pre-obtained importance weight and the pre-set priority corresponding to each type of grid parameter data, use formula (2) to determine the score corresponding to each type of grid parameter data, and select the H types of grid parameter data with the highest scores as important data.
[0138] The formula (2) is:
[0139]
[0140] S j is the score of the grid parameter data of the j-th type; P j is the priority of the grid parameter data of the j-th type; a is the first adjustment parameter; b is the second adjustment parameter; e is the natural number;
[0141] Formula (2) is introduced to calculate the score of each type of grid parameter data, which not only considers the importance weight of the data but also combines the pre-set priority. This comprehensive evaluation method enables the system to more comprehensively identify potential problems and improves the overall decision-making quality.
[0142] Step 400: Input the important data determined in step 300 into the pre-trained fault type prediction model (LinearTransformer model) to determine the fault type of the grid. This model is trained through the first training dataset, and each important data corresponds to a pre-set fault type identifier.
[0143] Step 500: Estimate the uncertainty of the predicted fault type, and use formula (3) to calculate the uncertainty score of each fault type. The formula (3) is:
[0144] U x =-px log(p x );
[0145] where U x is the uncertainty score of the x-th type of fault;
[0146] p x is the probability value of the x-th type of fault predicted by the fault type prediction model.
[0147] In this embodiment, the uncertainty score of the fault type is calculated by formula (3), and the corresponding warning threshold is set. This method helps the operation and maintenance personnel to timely discover those prediction results with low confidence and take additional verification measures. This greatly enhances the reliability and security of the system.
[0148] Step 600: Determine whether to send a warning message based on the uncertainty score of each fault type and the preset threshold. If the uncertainty score of a certain fault type is greater than the preset threshold, send a warning message.
[0149] This second embodiment demonstrates how to construct an efficient and accurate power grid fault prediction system by a series of carefully designed steps using advanced machine learning techniques. By introducing a dynamic adjustment mechanism, a comprehensive scoring system, and an uncertainty estimation and warning mechanism, the present invention not only improves the accuracy of fault prediction, but also enhances the robustness and reliability of the system, providing strong support for the safe and stable operation of modern smart power grids.
[0150] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.
[0151] In the present invention, unless otherwise clearly defined and limited, the terms "installed", "connected", "connected", "fixed", etc. should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium; it may be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0152] In the present invention, unless otherwise clearly specified or limited, a first feature being "on" or "under" a second feature may mean that the first and second features are in direct contact, or that the first and second features are indirectly in contact via an intermediate medium. Further, a first feature being "above", "over" and "on top of" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the horizontal height of the first feature is higher than that of the second feature. A first feature being "under", "below" and "beneath" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the horizontal height of the first feature is lower than that of the second feature.
[0153] In the description of this specification, the descriptions of terms such as "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.
[0154] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A collection method based on a power grid data collector, characterized in that: include: S100, acquiring power grid data collected by a data collector deployed in the power grid; The power grid data includes various types of power grid parameter data; S200, determining important data in the power grid data based on the pre-acquired importance weight and the pre-set priority corresponding to each type of power grid parameter data; S300, inputting the important data into a pre-trained fault type prediction model to determine the fault type of the power grid; The fault type prediction model is a Linear Transformer model; Wherein, a pre-set fault type prediction model is trained using a first training data set to obtain a trained fault type prediction model; The first training data set includes: important data corresponding to a plurality of different time points in a historical time period, wherein each important data in the first training data set corresponds to a preset fault type identifier.
2. The data acquisition method based on the power grid data collector according to claim 1 is characterized in that: The grid data include: voltage level, current intensity, grid operating frequency, and grid load rate.
3. The collection method based on the power grid data collector according to claim 2 is characterized in that: The method further includes before S100: S001. Obtain the importance weight corresponding to each type of power grid parameter data in the power grid data according to the power grid data collected by the data collector deployed in the power grid at a preset frequency within a preset first time period and a pre-trained deep network learning model; wherein, a pre-set deep network learning model is trained using a second training data set to obtain a trained deep network learning model; The second training data set includes: N samples and preset fault type identifiers corresponding to the samples one by one; The i-th sample includes: the power grid data collected by the data collector deployed in the power grid for the i-th time within the first time period, i∈N.
4. The collection method based on the power grid data collector according to claim 3 is characterized in that: The S001 specifically includes: S001-1. Use the DeepExplainer method in the SHAP library to create an interpreter object, and pass the trained deep network learning model and the second training data set into the created parser object; S001-2. Using the created interpreter object, calculate the SHAP value for each sample in the second training data set to obtain the SHAP value corresponding to each type of power grid parameter data in each sample; S001-3. Based on the SHAP values corresponding to each type of power grid parameter data in each sample, obtain the importance weight corresponding to each type of power grid parameter data.
5. The collection method based on the power grid data collector according to claim 4 is characterized in that: The S001-3 specifically includes: Based on the SHAP values corresponding to each type of power grid parameter data in each sample, the importance weight corresponding to each type of power grid parameter data is obtained using formula (1); The formula (1) is: W j represents the importance weight corresponding to the j-th type of power grid parameter data; SHAP ij represents the SHAP value of the j-th type of power grid parameter data in the i-th sample; M is the total number of power grid parameter data types; k is used to indicate traversal of all power grid parameter data types.
6. The data acquisition method based on the power grid data collector according to claim 5 is characterized in that: The S200 specifically includes: S200-1. Based on the pre-acquired importance weight and pre-set priority corresponding to each type of power grid parameter data, determine the score corresponding to each type of power grid parameter data using formula (2); The formula (2) is: S j is the score of the j-th type of power grid parameter data; P j is the priority of the j-th type of power grid parameter data; a is the first adjustment parameter; b is the second adjustment parameter; e is a natural number; S200-2. Scoring corresponding to each type of grid parameter data, and determining important data in the grid data; The important data are H types of power grid parameter data with the largest scores.
7. The data acquisition method based on the power grid data collector according to claim 6 is characterized in that: The step S300 further includes: S400, estimating the uncertainty of the predicted fault type to obtain an uncertainty score for each fault type; S500: Determine whether to issue a warning message based on the uncertainty score of each fault type and a preset threshold.
8. The collection method based on the power grid data collector according to claim 7 is characterized in that: The S400 specifically includes: The uncertainty of the predicted fault type is estimated, and the uncertainty score of each fault type is calculated using formula (3); The formula (3) is: U x =-p x log(p x ); Among them, U x is the uncertainty score of the xth fault type; p x is the probability value of the xth fault type predicted by the fault type prediction model.
9. The collection method based on the power grid data collector according to claim 8 is characterized in that: The S500 specifically includes: When the uncertainty score of the fault type is greater than a preset threshold, it is determined that a warning message is issued.
10. A data acquisition system based on a power grid data collector, characterized in that: include: at least one processor; as well as At least one memory is communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute a collection method based on a power grid data collector as described in any one of claims 1-9.