Method and device for diagnosing abnormal working condition of power distribution area

Abstract

The invention relates to the technical field of power grid safety, and particularly provides a power distribution area abnormal working condition diagnosis method and device, and the method comprises the steps: obtaining key state quantity data of a to-be-diagnosed power distribution area; and taking the key state quantity data as input of an edge side diagnosis model corresponding to the to-be-diagnosed power distribution area, and obtaining an abnormal working condition diagnosis result output by the edge side diagnosis model. According to the technical scheme provided by the invention, the problem of target domain sample scarcity can be relieved, the cross-domain diagnosis precision and real-time performance are improved, and engineering landing is facilitated.

Description

Technical Field

[0001] This invention relates to the field of power grid safety technology, specifically to a method and device for diagnosing abnormal operating conditions in a distribution substation. Background Technology With the continuous advancement of the construction of new power systems, distribution substations, as the end-point power supply units of the power grid, play a crucial role in ensuring power supply reliability and the safe and stable operation of the power grid through abnormal condition diagnosis. However, in practical applications, distribution substations face challenges such as the scarcity of fault samples, the diversity of equipment types, and dynamic changes in operating conditions. This leads to false alarms and false negatives due to the single feature of traditional threshold judgment methods. Meanwhile, deep learning-based diagnostic methods are limited by the assumption that training and test data must satisfy independent and identically distributed conditions, making it difficult to effectively address the generalization requirements across substations and operating conditions.

[0002] Against this backdrop, traditional methods for diagnosing abnormal operating conditions in distribution substations have gradually revealed numerous problems. Whether relying on expert experience or analysis methods based on single-source data models, both exhibit significant limitations when processing high-dimensional, multimodal monitoring data: low sensitivity, high false alarm rates, and difficulty in meeting practical needs. Meanwhile, centralized analysis in cloud centers also suffers from significant latency bottlenecks due to the complexity of data transmission and processing, further restricting diagnostic efficiency and real-time performance. These problems collectively highlight the inadequacies of existing methods in dealing with complex operating conditions and provide an urgent impetus for technological innovation to overcome these challenges. Summary of the Invention

[0003] To overcome the above-mentioned defects, the present invention proposes a method and device for diagnosing abnormal operating conditions in a power distribution area.

[0004] Firstly, a method for diagnosing abnormal operating conditions in a distribution transformer area is provided, the method comprising: Acquire key status data of the distribution radio area to be diagnosed; The key state data is used as input to the edge-side diagnostic model corresponding to the distribution area to be diagnosed, and the abnormal operating condition diagnostic results output by the edge-side diagnostic model are obtained.

[0005] Preferably, the process of determining the key state quantities includes: Step 1. Use the XGBoost algorithm to calculate the importance of different state variables in the distribution area; Step 2. Form an initial set X0 with the state variables of high importance, and form a selection set X1 with the remaining state variables; Step 3. Extract a state variable from X1 and add it to X0. Use X0 to train the source domain abnormal working condition diagnosis model and evaluate the diagnostic accuracy of the source domain abnormal working condition diagnosis model. Determine whether the diagnostic accuracy has improved. If so, retain the state variable; otherwise, remove the state variable. Step 4. Repeat step 3 until X1 is an empty set, and use the state variables in X0 as key state variables.

[0006] Preferably, the training process of the edge-side diagnostic model includes: Identify similar distribution stations to the distribution station to be diagnosed within each distribution station area; The similar distribution station area is divided into a first similar distribution station area and a second similar station area; The key state data of the first similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the first source domain training data, the key state data of the second similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the second source domain training data, and the key state data of the distribution transformer area to be diagnosed and its corresponding abnormal operating condition diagnosis results are used as the target domain training data. The diagnostic model of the cloud center is trained using the first source domain training data and the target domain training data, and the model parameters of the edge-side diagnostic model corresponding to the distribution area to be diagnosed are updated using a transfer learning algorithm. The edge-side diagnostic model corresponding to the distribution area to be diagnosed is trained using the second source domain training data and the target domain training data.

[0007] Furthermore, determining similar distribution stations to the distribution station to be diagnosed within each distribution station area includes: Calculate the Kourbak-Leibler divergence between the critical state data of each distribution station area and the critical state data of the distribution station area to be diagnosed. Distribution stations with a Kurbak-Leibler divergence of less than 0.2, whose commissioning year differs from that of the distribution station to be diagnosed by no more than 2 years, and whose equipment type and model are the same as those of the distribution station to be diagnosed are selected as similar distribution stations to the distribution station to be diagnosed.

[0008] Furthermore, the diagnostic model at the cloud center and the diagnostic model at the edge have the same structure, both including: a first general feature extractor constructed by a CNN-FC module, a second general feature extractor constructed by a CNN-FC module, a third general feature extractor constructed by a CNN-FC module, a specific feature extractor constructed by a ResNet network, a fourth general feature extractor constructed by a CNN-FC module, and a feature classifier, all connected in sequence.

[0009] Furthermore, during the training of the diagnostic model at the cloud center and the diagnostic model at the edge, the loss function is as follows:

[0010] In the above formula, The value of the loss function. The loss function value for the diagnostic model of the cloud center. The loss function value for the edge-side diagnostic model. The MSE loss function is the output of the feature classifier.

[0011] Furthermore, the loss function values ​​of the diagnostic model at the cloud center and the diagnostic model at the edge are as follows:

[0012] In the above formula, The cross-entropy loss function value of the diagnostic model for the cloud center. This represents the cross-entropy loss function value of the edge-side diagnostic model. The coefficient of change of the penalty term. The MK-MMD loss function value of the diagnostic model for the cloud center. The MK-MMD loss function value of the edge-side diagnostic model. The LMMD loss function value of the diagnostic model for the cloud center. This represents the LMMD loss function value of the edge-side diagnostic model.

[0013] Furthermore, the MSE loss function output by the feature classifier is as follows:

[0014] In the above formula, The number of samples in the target domain. The target domain predicted label is output by the feature classifier of the diagnostic model in the cloud center. The target domain predicted label is output by the feature classifier of the edge-side diagnostic model.

[0015] Furthermore, the coefficient of change of the penalty term is as follows:

[0016] In the above formula, It is a natural constant. These are random coefficients.

[0017] Secondly, a diagnostic device for abnormal operating conditions of a distribution transformer area is provided, the diagnostic device for abnormal operating conditions of a distribution transformer area comprising: The acquisition module is used to acquire key status data of the distribution radio area to be diagnosed; The analysis module is used to take the key state data as input to the edge-side diagnostic model corresponding to the distribution area to be diagnosed, and obtain the abnormal operating condition diagnostic results output by the edge-side diagnostic model.

[0018] Thirdly, a computer device is provided, comprising: one or more processors; The processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the method for diagnosing abnormal operating conditions in the distribution radio area is implemented.

[0019] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed, the method for diagnosing abnormal operating conditions of a distribution substation is implemented.

[0020] The above-described technical solutions of the present invention have at least one or more of the following beneficial effects: This invention provides a method and apparatus for diagnosing abnormal operating conditions in a distribution transformer substation, comprising: acquiring key state quantity data of the substation to be diagnosed; using the key state quantity data as input to an edge-side diagnostic model corresponding to the substation to be diagnosed, and obtaining the abnormal operating condition diagnosis result output by the edge-side diagnostic model. The technical solution provided by this invention can alleviate the problem of scarce samples in the target domain, improve the accuracy and real-time performance of cross-domain diagnosis, and is easy to implement in engineering. Specifically: (1) This invention constructs a cloud-edge collaborative diagnostic architecture to achieve “cloud-edge” collaborative optimization. The cloud aggregates abnormal working condition data from multiple sources and trains a frozen layer network containing a general feature extractor (CNN) and a specific feature extractor (ResNet). The model parameters are then distributed to the edge nodes. The edge nodes reconstruct the network based on local target domain data, dynamically fine-tune the deep parameters, and feed back domain difference information, forming a collaborative update mechanism of “global feature sharing - local feature adaptation”.

[0021] (2) This invention proposes a dynamic screening strategy for key state variables. Based on the XGBoost algorithm, the importance index of state variables to abnormal operating conditions is calculated, and effective features are iteratively screened by combining the Sequence Forward Selection (SFS) method: the initial set consists of highly important state variables, the remaining state variables are gradually introduced and the changes in diagnostic accuracy are evaluated, the gain features are retained to eliminate redundancy, and the input dimension of the model is optimized.

[0022] (3) The present invention designs an improved multi-source domain deep transfer network. It integrates a dual-stream architecture that combines global feature extraction from CNN and local feature learning from ResNet. It measures the cross-domain global distribution difference through multi-kernel maximum mean difference (MK-MMD), and achieves fine-grained alignment at the class level by combining local maximum mean difference (LMMD). It introduces classifier consistency constraint loss (MSE) to reduce the deviation between the classification results of the source domain and the target domain, and constructs an adaptive transfer loss function with multiple losses to improve the cross-domain feature alignment effect.

[0023] (4) The present invention builds an experimental environment on an elastic cloud server (cloud) and an intelligent fusion terminal (edge ​​node). Typical abnormal working conditions such as low voltage and heavy overload are selected. By comparing the diagnostic accuracy of the proposed method with traditional single-domain transfer (DAN, CNN) and multi-domain transfer (CORAL, DACNN) methods, combined with ablation experiments (to verify the individual and synergistic effects of MK-MMD, LMMD, and MSE), t-SNE feature visualization and computational resource consumption analysis, the advantages of the proposed method in cross-domain diagnostic accuracy, computational efficiency and generalization are verified. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the main steps of the abnormal operating condition diagnosis method for distribution radio stations according to an embodiment of the present invention; Figure 2 This is a ranking diagram of the importance of state variables under low-voltage operating conditions according to an embodiment of the present invention; Figure 3 This is a diagram showing the low voltage diagnostic results under different state variables in an embodiment of the present invention. Figure 4 These are diagrams illustrating the low-voltage anomaly diagnosis results of different methods according to embodiments of the present invention. Figure 5 This is a visualization result of low-voltage operating condition data transfer learning according to an embodiment of the present invention; Figure 6 This is a confusion matrix diagram under the A / B→C migration according to an embodiment of the present invention; Figure 7 This is a graph showing the accuracy of the CNN test set under the A / B→C transfer in an embodiment of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1 See appendix Figure 1 , Figure 1 This is a schematic flowchart illustrating the main steps of a method for diagnosing abnormal operating conditions in a distribution substation according to an embodiment of the present invention. Figure 1 As shown, the abnormal operating condition diagnosis method for distribution transformer areas in this embodiment of the invention mainly includes the following steps: Step S101: Obtain key status data of the distribution radio area to be diagnosed; Step S102: Use the key state quantity data as input to the edge-side diagnostic model corresponding to the distribution area to be diagnosed, and obtain the abnormal operating condition diagnostic results output by the edge-side diagnostic model.

[0028] In this embodiment, the process of determining the key state quantities includes: Step 1. Use the XGBoost algorithm to calculate the importance of different state variables in the distribution area; Step 2. Form an initial set X0 with the state variables of high importance, and form a selection set X1 with the remaining state variables; Step 3. Extract a state variable from X1 and add it to X0. Use X0 to train the source domain abnormal working condition diagnosis model and evaluate the diagnostic accuracy of the source domain abnormal working condition diagnosis model. Determine whether the diagnostic accuracy has improved. If so, retain the state variable; otherwise, remove the state variable. Step 4. Repeat step 3 until X1 is an empty set, and use the state variables in X0 as key state variables.

[0029] Since various abnormal operating conditions in a distribution transformer substation are affected by multiple state variables, this invention classifies these state variables into environmental, electrical, parametric, and statistical categories based on their characteristics. Environmental state variables reflect the meteorological characteristics of the area where the substation is located and can be obtained through temperature sensors, humidity sensors, and timers. Electrical state variables are the measurable characteristics of the substation during operation and can be collected in real time using electricity meters. Parametric state variables belong to the rated characteristics of the substation and its equipment and can be obtained by querying equipment models, etc. Statistical state variables can be calculated in real time from electrical and parametric state variables. Different abnormal operating conditions are affected by different state variables; Table 1 details the state variables corresponding to each abnormal operating condition in the distribution transformer substation.

[0030] Table 1

[0031] In this embodiment, the training process of the edge-side diagnostic model includes: Identify similar distribution stations to the distribution station to be diagnosed within each distribution station area; The similar distribution station area is divided into a first similar distribution station area and a second similar station area; The key state data of the first similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the first source domain training data, the key state data of the second similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the second source domain training data, and the key state data of the distribution transformer area to be diagnosed and its corresponding abnormal operating condition diagnosis results are used as the target domain training data. The diagnostic model of the cloud center is trained using the first source domain training data and the target domain training data, and the model parameters of the edge-side diagnostic model corresponding to the distribution area to be diagnosed are updated using a transfer learning algorithm. The edge-side diagnostic model corresponding to the distribution area to be diagnosed is trained using the second source domain training data and the target domain training data.

[0032] In one implementation, determining similar distribution stations to the distribution station to be diagnosed in each distribution station area includes: Calculate the Kourbak-Leibler divergence between the critical state data of each distribution station area and the critical state data of the distribution station area to be diagnosed. Distribution stations with a Kurbak-Leibler divergence of less than 0.2, whose commissioning year differs from that of the distribution station to be diagnosed by no more than 2 years, and whose equipment type and model are the same as those of the distribution station to be diagnosed are selected as similar distribution stations to the distribution station to be diagnosed.

[0033] In one implementation, the diagnostic model at the cloud center and the diagnostic model at the edge have the same structure, both including: a first general feature extractor constructed by a CNN-FC module, a second general feature extractor constructed by a CNN-FC module, a third general feature extractor constructed by a CNN-FC module, a specific feature extractor constructed by a ResNet network, a fourth general feature extractor constructed by a CNN-FC module, and a feature classifier, which are connected in sequence.

[0034] In one implementation, the loss function during the training of the diagnostic model at the cloud center and the diagnostic model at the edge is as follows:

[0035] In the above formula, The value of the loss function. The loss function value for the diagnostic model of the cloud center. The loss function value for the edge-side diagnostic model. The MSE loss function is the output of the feature classifier.

[0036] In one implementation, the loss function values ​​of the diagnostic model at the cloud center and the diagnostic model at the edge are as follows:

[0037] In the above formula, The cross-entropy loss function value of the diagnostic model for the cloud center. This represents the cross-entropy loss function value of the edge-side diagnostic model. The coefficient of change of the penalty term. The MK-MMD loss function value of the diagnostic model for the cloud center. The MK-MMD loss function value of the edge-side diagnostic model. The LMMD loss function value of the diagnostic model for the cloud center. This represents the LMMD loss function value of the edge-side diagnostic model.

[0038] In one implementation, the MSE loss function output by the feature classifier is as follows:

[0039] In the above formula, The number of samples in the target domain. The target domain predicted label is output by the feature classifier of the diagnostic model in the cloud center. The target domain predicted label is output by the feature classifier of the edge-side diagnostic model.

[0040] In one implementation, the penalty term variation coefficient is as follows:

[0041] In the above formula, It is a natural constant. These are random coefficients.

[0042] To verify the feasibility and correctness of the above calculation method, this experiment used a 2-core CPU, 4GB RAM, 60GB hard drive, and 5Mbps bandwidth elastic cloud server as the cloud center for training the multi-source domain diagnostic model; and selected an 8-core processor, built-in GPU enhancement chip, equipped with a 5G communication module and running Linux as the edge device for training the target domain diagnostic model. The cloud center and the intelligent fusion terminal exchanged information such as model parameters, feature distribution, and classification distribution via the MQTT protocol. The hyperparameters of the CNN network model were set as follows: initial learning rate 0.001, Dropout value 0.3, 200 iterations, and the Adam method was used for optimization. During the experiment, the source domain dataset was used entirely for network model training, and the target domain dataset was divided into training and test sets in a 7:3 ratio.

[0043] To verify the performance of the proposed method, some distribution substations in a prefecture-level city were selected as the research objects. Eleven similar substations were chosen, with a commissioning year difference of no more than three years, identical equipment types and models (such as distribution transformers, low-voltage cables, and branch boxes), and a KL divergence of less than 0.25 for six months of data on total daily power load, power supply radius, temperature, and humidity. A total of 56,293 data points were collected for an abnormal operating condition migration diagnosis experiment under cloud-edge collaboration. One substation was selected as the target domain C, and the remaining 10 sets of monitoring data from similar substations were divided into two groups of five as source domains A and B, respectively. To verify the proposed method and achieve refined diagnosis of abnormal operating conditions, a migration experiment was conducted on a low-voltage operating condition scenario.

[0044] For low-voltage abnormal operating conditions, through data aggregation and analysis of actual operation and maintenance results, low-voltage operating conditions are subdivided into three categories: normal, abnormal, and severe. Each operating condition is sequentially used as the source domain, and the remaining operating conditions are used as the target domains. The source domain represents labeled data under known operating conditions, and the target domain represents unlabeled data under the operating conditions to be diagnosed. Samples under the three operating conditions are used as the datasets for the transfer task analysis, denoted as A, B, and C, respectively. Three types of transfer tasks are set: A / B→C, A / C→B, and B / C→A. Taking A / B→C as an example, it represents that the source domain is the datasets of operating conditions A and B, and the target domain is the dataset of operating condition C. The other two transfer tasks are similar. The sliding time window is set to 24 hours, that is, each state variable is composed of the current sampling point combined with the sampling points of the past 23 hours. Based on the feature importance calculated by the XGBoost algorithm, the iteration count is 200 times, and the top 20 state variables in importance are selected, such as... Figure 2 As shown.

[0045] Experimental results using the forward sequence selection method are as follows Figure 3As shown, the top 6 state variables by importance were used as the initial set, and the remaining 14 were used as the residual set. The results showed that the average accuracy of low voltage anomaly diagnosis was highest when 10 residual state variables were added (totaling 16 state variables). Therefore, this low voltage condition migration diagnosis experiment selected 16 state variables to construct abnormal condition samples in the source and target domains.

[0046] To further verify the superiority of the proposed method, five existing transfer learning methods—non-transfer CNN, Domain Adaptive Network (DAN) in transfer learning, CORAL algorithm, Domain Adversarial Convolutional Neural Network (DACNN), and Multi-Feature Space Adaptive Network (MFSAN)—were compared and analyzed with the proposed method.

[0047] Table 2 shows that the accuracy of multi-source domain transfer learning is generally higher than that of single-source domain transfer learning. Furthermore, the proposed method achieves the highest accuracy and superior transfer results across the three different transfer tasks. This confirms the superiority of the proposed method, demonstrating that it can effectively reduce the distributional differences between the source and target domains, thereby achieving better domain adaptation and classification performance.

[0048] Table 2

[0049] And from Figure 4 As can be seen, the proposed method achieves higher transfer diagnostic accuracy than deep convolutional neural networks without a transfer process. This indicates that the measurement and matching of the differences in features and classification distributions between source and target domain samples transfers sample features, thereby effectively improving diagnostic results across different domains. To further analyze the impact of the three loss functions (MK-MMK, LMMD, and MSE) in the proposed transfer learning strategy on the transfer learning results, ablation experiments were conducted for each of the three transfer tasks under low-voltage operating conditions. CELouss, used to eliminate the distribution difference between the predicted and true labels in the source domain, was included in each ablation experiment and therefore not compared in the ablation experiments. The ablation experiment results are shown in Table 3. Table 3

[0050] The impact of a single LMMD on the transfer performance of the proposed model is greater than that of MKMMD and MSE alone. When MK-MMD and LMMD are combined, the transfer performance of the proposed model is improved even more significantly. The adaptive transfer strategy constructed using MK-MMD, LMMD, and MSE can further enhance the alignment of domain features between multiple source domains and between each source domain and the target domain, resulting in the best transfer learning effect.

[0051] t-distribution stochastic neighbor embedding (t-SNE) is a dimensionality reduction technique used to represent high-dimensional datasets in a low-dimensional space (two- or three-dimensional), thereby enabling visualization. To further analyze the performance of the proposed method in unlabeled data feature clustering after transfer learning, the t-SNE algorithm was used to perform feature extraction and clustering visualization analysis on the target domain data extracted from three transfer learning tasks under low-voltage abnormal operating conditions. The results are as follows: Figure 5 , Figure 6 , Figure 7 The figures shown are the MK-MMD heatmap for each category, the confusion matrix under A / B→C transfer, and the CNN test set accuracy graph. These three graphs provide comprehensive support for the performance analysis of the model in multi-source domain transfer diagnosis from three aspects: feature distribution difference quantification, classification accuracy verification, and prediction stability visualization, effectively verifying the effectiveness of the proposed method.

[0052] Example 2 Based on the same inventive concept, the present invention also provides a diagnostic device for abnormal operating conditions of a distribution transformer area, the diagnostic device for abnormal operating conditions of a distribution transformer area comprising: The acquisition module is used to acquire key status data of the distribution radio area to be diagnosed; The analysis module is used to take the key state data as input to the edge-side diagnostic model corresponding to the distribution area to be diagnosed, and obtain the abnormal operating condition diagnostic results output by the edge-side diagnostic model.

[0053] Preferably, the process of determining the key state quantities includes: Step 1. Use the XGBoost algorithm to calculate the importance of different state variables in the distribution area; Step 2. Form an initial set X0 with the state variables of high importance, and form a selection set X1 with the remaining state variables; Step 3. Extract a state variable from X1 and add it to X0. Use X0 to train the source domain abnormal working condition diagnosis model and evaluate the diagnostic accuracy of the source domain abnormal working condition diagnosis model. Determine whether the diagnostic accuracy has improved. If so, retain the state variable; otherwise, remove the state variable. Step 4. Repeat step 3 until X1 is an empty set, and use the state variables in X0 as key state variables.

[0054] Preferably, the training process of the edge-side diagnostic model includes: Identify similar distribution stations to the distribution station to be diagnosed within each distribution station area; The similar distribution station area is divided into a first similar distribution station area and a second similar station area; The key state data of the first similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the first source domain training data, the key state data of the second similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the second source domain training data, and the key state data of the distribution transformer area to be diagnosed and its corresponding abnormal operating condition diagnosis results are used as the target domain training data. The diagnostic model of the cloud center is trained using the first source domain training data and the target domain training data, and the model parameters of the edge-side diagnostic model corresponding to the distribution area to be diagnosed are updated using a transfer learning algorithm. The edge-side diagnostic model corresponding to the distribution area to be diagnosed is trained using the second source domain training data and the target domain training data.

[0055] Furthermore, determining similar distribution stations to the distribution station to be diagnosed within each distribution station area includes: Calculate the Kourbak-Leibler divergence between the critical state data of each distribution station area and the critical state data of the distribution station area to be diagnosed. Distribution stations with a Kurbak-Leibler divergence of less than 0.2, whose commissioning year differs from that of the distribution station to be diagnosed by no more than 2 years, and whose equipment type and model are the same as those of the distribution station to be diagnosed are selected as similar distribution stations to the distribution station to be diagnosed.

[0056] Furthermore, the diagnostic model at the cloud center and the diagnostic model at the edge have the same structure, both including: a first general feature extractor constructed by a CNN-FC module, a second general feature extractor constructed by a CNN-FC module, a third general feature extractor constructed by a CNN-FC module, a specific feature extractor constructed by a ResNet network, a fourth general feature extractor constructed by a CNN-FC module, and a feature classifier, all connected in sequence.

[0057] Furthermore, during the training of the diagnostic model at the cloud center and the diagnostic model at the edge, the loss function is as follows:

[0058] In the above formula, The value of the loss function. The loss function value for the diagnostic model of the cloud center. The loss function value for the edge-side diagnostic model. The MSE loss function is the output of the feature classifier.

[0059] Furthermore, the loss function values ​​of the diagnostic model at the cloud center and the diagnostic model at the edge are as follows:

[0060] In the above formula, The cross-entropy loss function value of the diagnostic model for the cloud center. This represents the cross-entropy loss function value of the edge-side diagnostic model. The coefficient of change of the penalty term. The MK-MMD loss function value of the diagnostic model for the cloud center. The MK-MMD loss function value of the edge-side diagnostic model. The LMMD loss function value of the diagnostic model for the cloud center. This represents the LMMD loss function value of the edge-side diagnostic model.

[0061] Furthermore, the MSE loss function output by the feature classifier is as follows:

[0062] In the above formula, The number of samples in the target domain. The target domain predicted label is output by the feature classifier of the diagnostic model in the cloud center. The target domain predicted label is output by the feature classifier of the edge-side diagnostic model.

[0063] Furthermore, the coefficient of change of the penalty term is as follows:

[0064] In the above formula, It is a natural constant. These are random coefficients.

[0065] Example 3 Based on the same inventive concept, this invention also provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of the distribution area abnormal operating condition diagnosis method in the above embodiments.

[0066] Example 4 Based on the same inventive concept, this invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of the distribution area abnormal operating condition diagnosis method in the above embodiments.

[0067] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing abnormal operating conditions in a distribution radio area, characterized in that, The method includes: Acquire key status data of the distribution radio area to be diagnosed; The key state data is used as input to the edge-side diagnostic model corresponding to the distribution area to be diagnosed, and the abnormal operating condition diagnostic results output by the edge-side diagnostic model are obtained.

2. The method as described in claim 1, characterized in that, The process of determining the key state quantities includes: Step 1. Use the XGBoost algorithm to calculate the importance of different state variables in the distribution area; Step 2. Form an initial set X0 with the state variables of high importance, and form a selection set X1 with the remaining state variables; Step 3. Extract a state variable from X1 and add it to X0. Use X0 to train the source domain abnormal working condition diagnosis model and evaluate the diagnostic accuracy of the source domain abnormal working condition diagnosis model. Determine whether the diagnostic accuracy has improved. If so, retain the state variable; otherwise, remove the state variable. Step 4. Repeat step 3 until X1 is an empty set, and use the state variables in X0 as key state variables.

3. The method as described in claim 1, characterized in that, The training process of the edge-side diagnostic model includes: Identify similar distribution stations to the distribution station to be diagnosed within each distribution station area; The similar distribution station area is divided into a first similar distribution station area and a second similar station area; The key state data of the first similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the first source domain training data, the key state data of the second similar distribution transformer area and its corresponding abnormal operating condition diagnosis results are used as the second source domain training data, and the key state data of the distribution transformer area to be diagnosed and its corresponding abnormal operating condition diagnosis results are used as the target domain training data. The diagnostic model of the cloud center is trained using the first source domain training data and the target domain training data, and the model parameters of the edge-side diagnostic model corresponding to the distribution area to be diagnosed are updated using a transfer learning algorithm. The edge-side diagnostic model corresponding to the distribution area to be diagnosed is trained using the second source domain training data and the target domain training data.

4. The method as described in claim 3, characterized in that, The step of identifying similar distribution stations to the distribution station to be diagnosed in each distribution station area includes: Calculate the Kourbak-Leibler divergence between the critical state data of each distribution station area and the critical state data of the distribution station area to be diagnosed. Distribution stations with a Kurbak-Leibler divergence of less than 0.2, whose commissioning year differs from that of the distribution station to be diagnosed by no more than 2 years, and whose equipment type and model are the same as those of the distribution station to be diagnosed are selected as similar distribution stations to the distribution station to be diagnosed.

5. The method as described in claim 3, characterized in that, The diagnostic model at the cloud center and the diagnostic model at the edge have the same structure, both including: a first general feature extractor constructed by a CNN-FC module, a second general feature extractor constructed by a CNN-FC module, a third general feature extractor constructed by a CNN-FC module, a specific feature extractor constructed by a ResNet network, a fourth general feature extractor constructed by a CNN-FC module, and a feature classifier, all connected in sequence.

6. The method as described in claim 3, characterized in that, During the training of the diagnostic model at the cloud center and the diagnostic model at the edge, the loss function is as follows: In the above formula, The value of the loss function. The loss function value for the diagnostic model of the cloud center. The loss function value for the edge-side diagnostic model. The MSE loss function is the output of the feature classifier.

7. The method as described in claim 6, characterized in that, The loss function values ​​of the diagnostic model for the cloud center and the diagnostic model for the edge are as follows: In the above formula, The cross-entropy loss function value of the diagnostic model for the cloud center. This represents the cross-entropy loss function value of the edge-side diagnostic model. The coefficient of change of the penalty term. The MK-MMD loss function value of the diagnostic model for the cloud center. The MK-MMD loss function value of the edge-side diagnostic model. The LMMD loss function value of the diagnostic model for the cloud center. This represents the LMMD loss function value of the edge-side diagnostic model.

8. The method as described in claim 6, characterized in that, The MSE loss function output by the feature classifier is as follows: In the above formula, The number of samples in the target domain. The target domain predicted label is output by the feature classifier of the diagnostic model in the cloud center. The target domain predicted label is output by the feature classifier of the edge-side diagnostic model.

9. The method as described in claim 7, characterized in that, The coefficient of change for the penalty term is as follows: In the above formula, It is a natural constant. These are random coefficients.

10. An apparatus for diagnosing abnormal operating conditions in a distribution substation based on any one of claims 1-9, characterized in that, The device includes: The acquisition module is used to acquire key status data of the distribution radio area to be diagnosed; The analysis module is used to take the key state data as input to the edge-side diagnostic model corresponding to the distribution area to be diagnosed, and obtain the abnormal operating condition diagnostic results output by the edge-side diagnostic model.

11. A computer device, characterized in that, include: One or more processors; The processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the abnormal operating condition diagnosis method for distribution radio areas as described in any one of claims 1 to 9 is implemented.

12. A computer-readable storage medium, characterized in that, It contains a computer program, which, when executed, implements the method for diagnosing abnormal operating conditions of a distribution substation as described in any one of claims 1 to 9.

Images