Positive and negative sample data processing method and system for substation patrol algorithm

Abstract

The invention discloses a positive and negative sample data processing method and system for a transformer substation patrol algorithm, and the method comprises the steps: setting a transformer substation fault type, determining a first type of parameters, determining a second type of parameters and a third type of parameters from high to low based on a logistic regression algorithm according to the weight of effect variables, and obtaining a positive and negative sample data according to a historical fault database. And performing positive and negative sample labeling on the visual fault sample, the parameter fault sample and the environment fault sample, constructing a model by adopting a convolutional neural network, inputting positive and negative sample data to perform model training, and inputting a first type of parameters, a second type of parameters and a third type of parameters in the real-time inspection data into the model to perform fault judgment. According to the method, positive and negative sample labeling is carried out on the three fault types, the strict requirement for patrol data sample processing is lowered, fault judgment is carried out on the three types of parameters which cause the highest effect quantity weight of the same fault type, and the reliability and accuracy of fault judgment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automated inspection, and in particular to a method and system for processing positive and negative sample data of a substation inspection algorithm. Background Art

[0002] The main methods for processing positive and negative sample data in substation inspection algorithms in the existing technology are: solving the problem of sample category imbalance through the design of loss function for unbalanced samples; improving prediction accuracy based on link prediction methods based on PU learning and random walk; augmenting positive and negative sample images based on deep contrast learning technology to obtain better positive and negative sample feature expressions; improving small target detection capabilities through anchor frame adjustment and feature fusion based on model architecture optimization.

[0003] Patent document 1 with announcement number CN117294015A provides a one-key sequential control monitoring and early warning method for substations based on image recognition. For each monitoring gate position of the one-key sequential control of the substation, the correct and incorrect gate-closing images are recorded as positive and negative samples respectively; the anchor samples at each monitoring gate position are respectively combined with the positive and negative samples to construct positive sample pairs and negative sample pairs; the positive sample pairs and negative sample pairs are used to train the Siamese network by minimizing the loss function to obtain a gate-closing discriminator; the real-time image of each monitoring gate position is obtained, and the real-time image of each monitoring gate position is paired with the comparison image and input into the gate-closing discriminator to obtain whether each monitoring gate position is correctly closed, so as to solve the problem that the inspection of the gate-closing situation in the prior art will consume a lot of manpower and material resources. However, due to the influence of shooting conditions such as shooting equipment, angle, resolution, distance and light intensity, it is difficult to define whether the monitoring image is completely closed, resulting in inaccurate distribution of positive and negative sample images, which in turn affects the judgment result.

[0004] Patent document 2 with announcement number CN117173385B provides a detection method for a substation, which includes: based on a partially supervised learning method, labeling each sample image of the substation, obtaining a predetermined database containing a number of positive sample images and a number of negative sample images, wherein the positive sample images are sample images with labels of non-fault states, and the negative sample images are sample images with labels of fault states. However, in the actual operation of the substation, abnormal samples are usually rarer and less in number than normal samples, resulting in an imbalance between positive and negative samples, which affects the recognition accuracy of the model.

[0005] The above data processing methods either solve the problem of sample imbalance, or solve the problem of low algorithm convergence accuracy, or perform clustering oversampling and data resampling, or enhance sample features through comparative learning. However, in the face of the diversity of sample data types during power system inspections, a single positive and negative sample data processing method cannot accurately adapt to different types of sample data, resulting in reduced accuracy and reliability of the substation inspection algorithm result judgment. Summary of the invention

[0006] Purpose of the invention: In view of the above problems, the present invention proposes a method and system for processing positive and negative sample data of a substation patrol algorithm. Technical solution: In a first aspect, the present invention provides a method for processing positive and negative sample data of a substation patrol algorithm, the method comprising: S1. Set the substation fault type, determine the first type of parameters, and determine the second and third type of parameters based on the effect weight from high to low based on the logistic regression algorithm; S2, positive and negative sample labeling, based on the historical fault database, visual fault samples, parameter fault samples and environmental fault samples are labeled respectively; S3, using convolutional neural network (CNN) to build a model, input positive and negative sample data for model training; S4, inputting the model trained in step S3 according to the real-time patrol data, extracting the first type parameter, the second type parameter and the third type parameter associated with the fault type in the patrol data according to step S1; Input the first type of parameters into the model trained in step S3, and if there is a fault, output the judgment result R1=1; if there is no fault, output the judgment result R1=0; Input the second type of parameters into the model trained in step S3. If there is a fault, output the judgment result R2=1. If there is no fault, output the judgment result R2=0. Compare the judgment result with the judgment result of the first type of parameters. If R1=R2=1, the output result is a fault. If R1≠R2, input the third type of parameters into the model trained in step S3. If a fault exists, output the judgment result R3=1. If no fault exists, output the judgment result R3=0. Further obtain R1+R2+R3. If R1+R2+R3=2, the output result is a fault. If R1+R2+R3=1, the output result is that there is no fault.

[0007] Optionally, the step S1 includes: When S101 sets the substation fault type, based on the fault acquisition method during the inspection process, the fault type is set to visual fault, parameter fault and environmental fault; S102 For visual faults, inspection videos or images are preferred as the first-class parameters; for parameter faults, electrical monitoring data is preferred as the first-class parameters; for environmental faults, sensor collection data is preferred as the first-class parameters; After the first type of parameters are selected in S103, the second type of parameters and the third type of parameters are determined from high to low according to the effect size weights based on the logistic regression algorithm.

[0008] Optionally, when detecting cable damage, a corresponding logical relationship is generated based on the data change relationship of voltage, current, circuit heating and three-phase power supply imbalance caused by the cable damage, and the effect quantity weights corresponding to the voltage, current and temperature are calculated.

[0009] Optionally, step S2 includes: positive and negative sample labeling, and labeling visual fault samples, parameter fault samples, and environment fault samples respectively according to a historical fault database.

[0010] Optionally, inspection images are used as sample data for visual faults, and positive samples and negative samples are selected at a ratio of 10:90.

[0011] Optionally, the substation inspection data with completed result determination is updated to the positive and negative sample database, and the model is iterated and optimized. In a second aspect, the present invention also proposes a positive and negative sample data processing system of a substation patrol algorithm, which is used to execute the positive and negative sample data processing method of the substation patrol algorithm, including a data acquisition module, a digital model construction module, a fault detection module and a strategy output module; Optionally, the substation inspection data collected by the parameter collection module 100 is preprocessed based on the type of data collected, such as image, numerical value or text; Optionally, the pattern recognition module 200 recognizes and classifies the data pre-processed by the acquisition module 100; Optionally, the sample labeling module 300 obtains the substation historical fault data and the historical status data corresponding to the historical fault data from the historical fault database module 700, and performs positive and negative sample labeling; Optionally, the model training module 400 uses a convolutional neural network (CNN) to build a model and inputs the positive and negative samples for training; Optionally, the detection module 500 inputs the real-time patrol data into the model trained by the model training module 400 to determine the patrol result; Optionally, the data acquisition module includes a first data acquisition module and a second data acquisition module. Preprocessing operations include deleting or filling missing data, identifying and processing data points that do not meet expectations, reducing and eliminating random fluctuations and errors in the data, adjusting image size, image enhancement and normalizing pixel values, and word segmentation, stem extraction and vectorization of text data.

[0012] In a third aspect, the present invention further provides a computer device, comprising: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0013] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of any of the above methods when executed by a processor.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention provides a distribution network fault monitoring system and method based on digital twins. By selecting the first type of parameters for the specified fault type, and based on the logistic regression algorithm, determining the second type of parameters and the third type of parameters according to the effect weight from high to low, the substation patrol data is respectively used for fault judgment, thereby improving the accuracy of fault judgment.

[0015] 2. The parameter fault samples and environment fault samples provided by the present invention are labeled with positive and negative samples based on the numerical range, which has the advantages of high accuracy and accurate result judgment, thereby reducing the strict processing standards for visual fault inspection images and solving the problems of blurred image annotation boundaries and inaccurate annotation results.

[0016] 3. The positive and negative sample data processing method provided by the present invention does not require huge data processing calculations, and the inspection data result judgment process does not need to rely on excessive deep learning algorithms, thereby reducing the hardware requirements of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of a method for processing positive and negative sample data of a substation inspection algorithm provided by an embodiment of the present invention; Figure 2 A schematic diagram of a positive and negative sample data processing system for a substation patrol algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Obviously, many modifications and changes made by those skilled in the art based on the purpose of the present invention belong to the protection scope of the present invention.

[0019] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when an element or component is said to be "connected" to another element or component, it may be directly connected to the other element or component, or there may be an intermediate element or component. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1

[0021] The embodiment of the present invention provides a method for processing positive and negative sample data of a substation patrol algorithm. For details, please refer to Figure 1 , Figure 1 A flow chart of a method for processing positive and negative sample data of a substation patrol algorithm provided by an embodiment of the present invention, the method comprising the steps of: S1. Set the substation fault type, determine the first type of parameters, and determine the second and third type of parameters based on the effect weight from high to low based on the logistic regression algorithm; Optionally, the step S1 further includes: When S101 sets the substation fault type, based on the fault acquisition method during the inspection process, the fault type is set to visual fault, parameter fault and environmental fault.

[0022] Preferably, visual faults are usually acquired based on inspection image recognition, including electrical equipment housing faults, electrical equipment connection faults, insulation faults, line damage faults, and external intrusion behavior detection. Parameter faults are usually acquired based on the operating parameters of electrical equipment, including overvoltage / current faults, grounding faults, short circuit faults, and leakage faults. Environmental faults are usually acquired based on sensor data, including temperature and humidity anomalies, water leakage detection anomalies, and fire detection anomalies.

[0023] S102 For visual faults, inspection videos or images are preferred as the first type of parameters. During the specific implementation, a camera device can be used to collect videos or images of the substation. The camera device can be an inspection robot, a drone, a fixed camera, etc. In combination with the camera environment, a fill light device can be added to the camera device or an infrared camera device can be used. For parameter faults, electrical monitoring data is preferred as the first type of parameters. During the specific implementation, current / voltage parameters can be obtained through current / voltage transformers, and leakage current, short-circuit current, etc. can be obtained through electrical fire monitoring devices. For environmental faults, sensor acquisition data is preferred as the first type of parameters. During the specific implementation, temperature and humidity sensors, fire detectors, water leakage monitoring sensors, etc. can be used to obtain substation environment-related parameters.

[0024] After the first type of parameters are selected in S103, the second type of parameters and the third type of parameters are determined from high to low according to the effect size weights based on the logistic regression algorithm.

[0025] Specifically, when detecting cable damage, a corresponding logical relationship is generated according to the change relationship of voltage, current, circuit heating and three-phase power supply imbalance data caused by cable damage, which is: According to the following parameters, X 1: Voltage, X 2: Current, X 3: Wire temperature, Y :When the cable is damaged Y= 1. When the cable is not damaged, when the cable is damaged Y=0， Build a logistic regression model: p ( Y =1| X 1, X 2, X 3)=(1+ e −(β0+β1X1+β2X2+β3X3) ) -1 in p ( Y =1| X 1, X 2, X 3) is the probability of wire breakage under given voltage, current and wire temperature, β 0, β 1, β 2, β 3 is the effect size weight, which needs to be estimated through training data, as follows: 1. Initialization parameters: Randomly initialize model parameters β 0, β 1, β 2, β 3.

[0026] 2. Construct feature matrix: Take voltage, current and wire temperature as features to construct feature matrix X。

[0027] 3. Calculate the predicted probability: Using the current model parameters, calculate the predicted probability of each sample: p ( Y =1| X )=(1+ e −(β0+β1X1+β2X2+β3X3) ) -1 .

[0028] 4. Calculate gradients: Calculate the gradient of the loss function with respect to each parameter:

[0029] in, L is the loss function, n is the sample size, yi It is i The actual labels of samples, Xi It is i The sample j feature values.

[0030] 5. Update parameters: Use gradient descent method to update model parameters:

[0031] in, η is the learning rate.

[0032] 6. Iterative optimization: Repeat steps 3-5 until the model parameters converge or the maximum number of iterations is reached. After the training is completed, the effect size weight can be obtained. β 0, β 1, β 2, β 3.

[0033] 7. Determine the second and third type parameters according to the effect size weights from high to low.

[0034] The above implementation is only one way to determine the parameters of a cable damage fault. Of course, other types of faults can also be achieved through the above method. The protection scope of the present invention is not limited to this. Any changes or replacements that can be easily thought of by anyone familiar with the technology within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

[0035] S2, positive and negative sample labeling, based on the historical fault database, visual fault samples, parameter fault samples and environmental fault samples are labeled separately; Optionally, step S2 includes: If the parameter corresponds to a visual fault parameter, an inspection image is obtained, and the inspection image is divided into a fault image and a non-fault image. The fault image is marked as a positive sample, and the non-fault image is marked as a negative sample. The positive sample and the negative sample are selected at a ratio of 10:90 to establish a positive and negative sample training set; If the parameter corresponds to a parameter fault, positive and negative samples are selected according to the substation historical fault database. Specifically, for parameters such as voltage, current, overvoltage, overcurrent and leakage current, the value range A under normal operating conditions and the value range B under faulty operating conditions are determined according to the periodic operating parameters. The electrical parameter values ​​within the value range B are marked as positive samples, and the electrical parameter values ​​within the value range A are marked as negative samples. If the parameter corresponds to an environmental fault, the sensor data is obtained. For the temperature, humidity, water immersion and smoke concentration parameters collected by the sensor, the corresponding normal working range C is set according to the operation requirements of the substation. The control room, transformer room, distribution room, battery room and high-voltage room in the substation facilities require different working range standards, which are not limited here. Environmental fault parameters that exceed the working range are marked as positive samples, and parameters within the standard working range are marked as negative samples.

[0036] In the prior art, the number of substation equipment has increased, and the existing equipment defect sample images cannot cover all equipment defect types. In addition, the image sample preprocessing process is cumbersome and the sample annotation results are inaccurate. Therefore, the substation operation and maintenance personnel only use the equipment defect image recognition results as a reference for inspection. For parameter faults and environmental fault samples, positive and negative samples are annotated within a numerical range. The data processing and sample annotation process are simple and accurate, so the image sample annotation results can be verified to make up for the inaccurate judgment results of the image sample results.

[0037] In other implementations of the present invention, positive and negative sample annotation of electric power text data is also included, specifically including cleaning, splicing, random deletion and replacement of text data, and then annotating the enhanced text data.

[0038] Regarding the above-mentioned different types of sample data labeling implementation methods, the difficulty of labeling a single type of sample data and the accuracy of the results are not easy to evaluate. When labeling, when determining positive and negative samples for image samples and text samples, the above-mentioned numerical sample labeling results can be used for verification and positive and negative sample labeling determination, thereby improving the overall accuracy of the sample labeling results.

[0039] S3, using convolutional neural network (CNN) to build a model, input positive and negative sample data for model training; Optionally, step S3 includes: S301 sets hyperparameters such as input size, number of categories, learning rate, batch size, and training cycle according to the labeled data samples; S302 defines the CNN model architecture, including convolutional layers, pooling layers, fully connected layers, etc., and moves it to the GPU or CPU; S303 Define loss and optimizer: Select the cross entropy loss function and Adam optimizer to compile the model; S304 trains the network for multiple epochs. In each epoch, data is fed into the model in batches, the loss is calculated, and the model weights are updated through backpropagation. After each epoch, the performance of the model on the validation set is evaluated to monitor overfitting.

[0040] Other implementation methods of the embodiments of the present invention also include, when performing model training, performing model training separately according to different data types under the same fault type in historical fault data, cross-checking and parameter optimization of respective training results, so as to achieve accuracy and consistency in the judgment of the training model results.

[0041] S4, judging the inspection result according to the real-time inspection data input into the model trained in step S3; Optionally, step S4 includes: S401 extracts first-category parameters, second-category parameters, and third-category parameters associated with the fault type in the inspection data according to step S1; S402: Input the first type of parameters into the model trained in step S3 to judge the inspection result. Optionally, if there is a fault, output the judgment result R1=1; if there is no fault, output the judgment result R1=0; S403 inputs the second type of parameters into the model trained in step S3, and judges the inspection result. Optionally, if there is a fault, the judgment result R2=1 is output, and if there is no fault, the judgment result R2=0 is output. The judgment result is further compared with the result of step S402. If R1=R2=1, the output result is a fault. If R1=R2=0, the output result is that there is no fault. The result is stored in the positive and negative sample database, and the positive and negative data samples are updated; if R1≠R2, go to step S404; S404 Input the third type of parameters into the model trained in step S3, and judge the inspection result. Optionally, if there is a fault, output the judgment result R3=1, if there is no fault, output the judgment result R3=0, and further obtain R1+R2+R3. If R1+R2+R3=2, the output result is a fault. If R1+R2+R3=1, the output result is that there is no fault. The result is stored in the positive and negative sample database, and the positive and negative sample data are updated.

[0042] Using the updated sample data, step 3 is continuously iterated and optimized for fault determination of subsequent substation inspection data samples.

[0043] On the basis of the implementation method of the above-mentioned embodiment of the present invention, based on the accuracy and reliability analysis and requirements of the result judgment of the substation patrol data, the fourth type of parameters and the fifth type of parameters can be further obtained according to the effect quantity weight, and the model trained in step S3 can be input for result judgment to improve the accuracy requirements of the result judgment. Embodiment 2

[0044] The embodiment of the present invention also provides a positive and negative sample data processing system for a substation patrol algorithm. For details, please refer to Figure 2 , Figure 2 A schematic diagram of a positive and negative sample data processing system for a substation patrol algorithm provided by an embodiment of the present invention, the system includes a parameter acquisition module 100, a pattern recognition module 200, a sample annotation module 300, a model training module 400, a detection module 500, a sample database module 600 and a historical fault database module 700; Optionally, the parameter acquisition module 100 is used to collect substation inspection data, and adopts corresponding preprocessing operations based on the collected data type such as image, value or text, including but not limited to the following processing methods: deleting or filling missing data, identifying and processing data points that do not meet expectations, reducing and eliminating random fluctuations and errors in the data, adjusting image size, image enhancement and normalizing pixel values, word segmentation, stem extraction and vectorization of text data to improve data quality.

[0045] Optionally, the pattern recognition module 200 is used to identify and classify the data pre-processed by the acquisition module 100. Specifically, the substation fault types are set to be visual faults, parameter faults and environmental faults, and the collected data are arranged in groups.

[0046] Preferably, the pattern recognition module 200 also selects the first type of parameters for fault judgment based on specific fault types, such as line damage, ground fault, short circuit fault, etc., and determines the second type of parameters and the third type of parameters from high to low according to the effect size weight based on the logistic regression algorithm, which are used to jointly determine the substation fault result.

[0047] Optionally, the historical fault database module 700 is used to store the historical fault types and corresponding fault data of the substation, and the sample labeling module 300 obtains the historical fault data of the substation and the historical status data corresponding to the historical fault data as sample data, and labels the above three types of fault sample data with positive and negative samples based on the fault type, and stores the labeled data in the sample database module 600.

[0048] Optionally, the model training module 400 uses a convolutional neural network (CNN) to build a model, inputs positive and negative sample data obtained from the sample database module 600 for model training, and the trained model is used for substation patrol data fault determination.

[0049] Optionally, the detection module 500 inputs the real-time patrol data into the model training module 400 to complete the training model, and determines the patrol result. Specifically, the first type of parameters, the second type of parameters and the third type of parameters in the pattern recognition module 200 are obtained.

[0050] Preferably, the first type of parameter input is tested, and if a fault exists, the judgment result R1=1 is output, and if there is no fault, the judgment result R1=0 is output; the second type of parameter input is tested, and if a fault exists, the judgment result R2=1 is output, and if there is no fault, the judgment result R2=0 is output, and the judgment result is further compared with the judgment result of the first type of parameter. If R1=R2=1, the output result is a fault, and the result is stored in the positive and negative sample database, and the positive and negative data samples are updated; if R1≠R2, the third type of parameter input is tested, and if a fault exists, the judgment result R3=1 is output, and if there is no fault, the judgment result R3=0 is output, and R1+R2+R3 is further obtained. If R1+R2+R3=2, the output result is a fault, and if R1+R2+R3=1, the output result is that there is no fault, and the result is stored in the positive and negative sample database, and the positive and negative sample data are updated.

[0051] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0052] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0053] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0054] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0055] Finally, it should be noted that, in this article, relationships such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

Claims

1. A method for processing positive and negative sample data of a substation inspection algorithm, characterized in that: The method comprises: S1. Set the substation fault type, determine the first type of parameters, and determine the second and third type of parameters based on the effect weight from high to low based on the logistic regression algorithm; S2, positive and negative sample labeling, based on the historical fault database, visual fault samples, parameter fault samples and environmental fault samples are labeled respectively; S3, using convolutional neural network (CNN) to build a model, input positive and negative sample data for model training; S4, inputting the model trained in step S3 according to the real-time patrol data, extracting the first type parameter, the second type parameter and the third type parameter associated with the fault type in the patrol data according to step S1; Input the first type of parameters into the model trained in step S3, and if there is a fault, output the judgment result R1=1; if there is no fault, output the judgment result R1=0; Input the second type of parameters into the model trained in step S3. If there is a fault, output the judgment result R2=1. If there is no fault, output the judgment result R2=0. Compare the judgment result with the judgment result of the first type of parameters. If R1=R2=1, the output result is a fault. If R1≠R2, input the third type of parameters into the model trained in step S3. If a fault exists, output the judgment result R3=1. If no fault exists, output the judgment result R3=0. Further obtain R1+R2+R3. If R1+R2+R3=2, the output result is a fault. If R1+R2+R3=1, the output result is that there is no fault.

2. The method according to claim 1, characterized in that The step S1 comprises: When S101 sets the substation fault type, based on the fault acquisition method during the inspection process, the fault type is set to visual fault, parameter fault and environmental fault; S102 For visual faults, inspection videos or images are preferred as the first-class parameters; for parameter faults, electrical monitoring data is preferred as the first-class parameters; for environmental faults, sensor collection data is preferred as the first-class parameters; After the first type of parameters are selected in S103, the second type of parameters and the third type of parameters are determined from high to low according to the effect size weights based on the logistic regression algorithm.

3. The method according to claim 2, characterized in that When detecting cable damage, the corresponding logical relationship is generated according to the change relationship of voltage, current, circuit heating and three-phase unbalanced power supply data caused by cable damage, and the effect quantity weights corresponding to voltage, current and temperature are calculated.

4. The method according to claim 1, characterized in that: The step S2 includes: positive and negative sample labeling, and labeling visual fault samples, parameter fault samples and environmental fault samples respectively according to the historical fault database.

5. The method according to claim 2, characterized in that: For visual faults, inspection images are used as sample data, and positive samples and negative samples are selected at a ratio of 10:

90.

6. The method according to claim 5, characterized in that The substation inspection data with completed result determination is updated to the positive and negative sample database, and the model is iterated and optimized.

7. A positive and negative sample data processing system for a substation patrol algorithm, used to implement the positive and negative sample data processing method for a substation patrol algorithm according to any one of claims 1 to 6, characterized in that: Includes the following modules: The substation inspection data collected by the parameter collection module (100) is pre-processed based on the collected data type, such as image, numerical value or text; The pattern recognition module (200) recognizes and classifies the data pre-processed by the acquisition module (100); The sample labeling module (300) obtains the substation historical fault data and the historical status data corresponding to the historical fault data from the historical fault database module (700), and performs positive and negative sample labeling; The model training module (400) uses a convolutional neural network (CNN) to build a model and inputs the positive and negative samples for training; The detection module (500) inputs the real-time patrol data into the model training module (400) to complete the training model and determine the patrol result.

8. The system according to claim 7, characterized in that The preprocessing operations include deleting or filling missing data, identifying and processing data points that do not meet expectations, reducing and eliminating random fluctuations and errors in the data, adjusting image size, image enhancement and normalizing pixel values, and word segmentation, stem extraction and vectorization of text data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images