Method, device, computer equipment and readable storage medium for determining line status

Abstract

The present application discloses a method, apparatus, computer device, and readable storage medium for determining line status, relating to the field of circuit inspection technology. The method addresses the current problems of missed line loss detection or misjudgment of faults, and low accuracy in line loss determination. The method comprises: obtaining a training dataset, the training dataset comprising multiple line status categories and multiple target historical inspection feature data corresponding to each of the multiple line status categories; training an established deep network model based on the training dataset until the model converges to obtain a target deep network model; obtaining a prediction dataset; inputting the prediction dataset into the target deep network model to obtain a target line status for the line to be determined, wherein the prediction dataset comprises multiple current inspection feature data indicating the line status of the line to be determined.

Description

Technical Field

[0001] The present application belongs to the field of circuit inspection technology, and more specifically, relates to a method, apparatus, computer equipment, and readable storage medium for determining a line status. Background Art

[0002] Circuit inspection refers to the regular inspection of the operating status of the circuit system and the working conditions of the equipment and facilities to ensure its normal operation and safety. At present, drone inspection has replaced manual inspection as the main method of circuit inspection. It can replace manual confirmation methods such as climbing the tower, thereby improving inspection efficiency. However, the applicant recognizes that the use of drone inspection usually requires drones to take a large number of inspection line images. Relevant personnel judge the line status of the current inspection line based on a large number of inspection line images. Due to the limitation of personnel energy and the influence of subjective factors, there are cases of missed detection or misjudgment of faults, that is, the inspection line with line faults (line loss) cannot be detected or the type of line loss on the inspection line is misjudged, and the accuracy of line loss judgment is low. Summary of the Invention

[0003] In view of this, the present invention provides a method, apparatus, computer device and readable storage medium for determining line status, the main purpose of which is to solve the current problems of missed detection or misjudgment of line loss faults and low accuracy of line loss judgment.

[0004] According to a first aspect of the present application, a method for determining a line status is provided, comprising:

[0005] Acquire a training data set, the training data set including a plurality of line status categories and a plurality of target historical inspection feature data corresponding to each of the plurality of line status categories;

[0006] Establishing a deep network model, and training the deep network model based on the training data set until the model converges to obtain a target deep network model;

[0007] A prediction data set is obtained, and the prediction data set is input into the target deep network model to obtain a target line state of the line to be determined, wherein the prediction data set includes a plurality of current inspection feature data for indicating the line state of the line to be determined.

[0008] Optionally, before obtaining the training data set, the method further includes:

[0009] Acquire a plurality of historical line inspection images, and acquire a regional identification number, rainfall information, temperature information, and wind speed information of a line area corresponding to each of the plurality of historical line inspection images;

[0010] Combining each of the plurality of historical line inspection images with a regional identification number, rainfall information, temperature information, and wind speed information of the corresponding line area to obtain a plurality of historical inspection feature data;

[0011] Acquire the multiple line status categories, and use the multiple line status categories to label the multiple historical inspection feature data to obtain multiple labeled historical inspection feature data;

[0012] An initial training data set is obtained based on the plurality of labeled historical inspection feature data.

[0013] Optionally, after obtaining the initial training data set based on the plurality of annotated historical inspection feature data, the method further includes:

[0014] Identifying at least one first historical inspection feature data with a missing discrete feature index in the initial training data set, deleting the at least one first historical inspection feature data from the initial training data set to obtain an intermediate training data set;

[0015] Identifying at least one second historical inspection feature data in the intermediate training data set that is missing a continuous feature indicator, obtaining identification information, labeling the at least one second historical inspection feature data using the identification information to obtain the at least one labeled second historical inspection feature data, and obtaining an updated intermediate training data set based on the at least one labeled second historical inspection feature data;

[0016] performing normalization processing of continuous characteristic indicators on the updated intermediate training data set to obtain a processed updated intermediate training data set;

[0017] extracting at least one to-be-filled historical inspection feature data corresponding to the at least one second historical inspection feature data from the processed updated intermediate training data set according to the identification information, and filling in missing values ​​of each to-be-filled historical inspection feature data in the at least one to-be-filled historical inspection feature data using a Lagrange interpolation method to obtain at least one target historical inspection feature data;

[0018] The training data set is obtained based on the at least one target historical inspection feature data.

[0019] Optionally, the training the deep network model based on the training data set until the model converges to obtain a target deep network model includes:

[0020] The training data set is processed using one-hot encoding to obtain a target training data set, wherein the target training data set includes a plurality of learning target labels and target historical inspection feature data corresponding to each of the plurality of learning target labels;

[0021] Training the deep network model using the target training data set until the model converges and determining target model parameters;

[0022] The target deep network model is obtained based on the target model parameters.

[0023] Optionally, the training of the deep network model using the target training data set until the model converges and determining target model parameters includes:

[0024] Dividing the target training data set into multiple training subsets, wherein each of the multiple training subsets includes the same amount of annotated target historical inspection feature data;

[0025] Selecting any one training subset from the multiple training subsets, inputting the selected training subset into the deep network model, and obtaining the line loss fault probability corresponding to each labeled target historical inspection feature data output by the deep network model;

[0026] Obtaining a loss function of the deep network model, and extracting a learning target label corresponding to each labeled target historical inspection feature data from the training subset;

[0027] Based on the learning target label and line loss failure probability corresponding to each labeled target historical inspection feature data, the loss value corresponding to the deep network model is calculated using the loss function;

[0028] When it is detected that the loss value does not exceed the target threshold, the model is determined to have converged, and the current model parameters of the deep network model are used as the target model parameters.

[0029] Optionally, the loss function of the deep network model is:

[0030]

[0031] Wherein, loss is the loss value of the loss function, N is the total number of labeled target historical inspection feature data included in the selected data training subset, i is the i-th labeled target historical inspection feature data in the training subset, c is the current classification, M is the multiple learning target labels, y ic is the label value corresponding to the target historical inspection feature data after the i-th annotation under the current classification, p icis the line loss failure probability corresponding to the target historical inspection feature data after the i-th annotation under the current classification.

[0032] Optionally, when it is detected that the loss value exceeds the target threshold, a weight value gradient of the deep network model is calculated based on the loss value using a stochastic gradient descent method;

[0033] Obtain a preset model learning rate of the deep network model, and update the weight value of the deep network model based on the weight value gradient and the preset model learning rate to obtain an updated deep network model, select a training subset other than the training subset from the multiple training subsets, input the selected other training subset into the updated deep network model, obtain the line loss failure probability corresponding to each labeled target historical inspection feature data in the other training subset output by the updated deep network model, calculate the updated loss value corresponding to the updated deep network model using the loss function based on the line loss failure probability corresponding to each labeled target historical inspection feature data in the other training subset, until the updated loss value does not exceed the target threshold, determine model convergence, and use the current model parameters of the updated deep network model as the target model parameters.

[0034] According to a second aspect of the present application, a device for determining a line status is provided, comprising:

[0035] An acquisition module is used to acquire a training data set, wherein the training data set includes multiple line status categories and multiple target historical inspection feature data corresponding to each line status category in the multiple line status categories;

[0036] A training module is used to establish a deep network model and train the deep network model based on the training data set until the model converges to obtain a target deep network model;

[0037] An input module is used to obtain a prediction data set, input the prediction data set into the target deep network model, and obtain a target line state of the line to be determined, wherein the prediction data set includes multiple current inspection feature data for indicating the line state of the line to be determined.

[0038] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0039] According to a fourth aspect of the present application, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0040] Through the above technical solution, the present application provides a method, device, computer equipment and readable storage medium for determining the line status. The present application first obtains a training data set, then establishes a deep network model, trains the deep network model based on the training data set until the model converges, and obtains a target deep network model. Finally, a prediction data set is obtained, and the prediction data set is input into the target deep network model to obtain the target line status of the line to be determined; by using a training data set including multiple historical inspection feature data to train the deep network model until convergence, a target deep network model that can accurately judge the line status can be trained, and the prediction data set including multiple current inspection feature data is input into the target deep network model, which can accurately predict the target line status of the line to be determined. This method has high reusability and can effectively improve the accuracy of line loss judgment.

[0041] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0043] Figure 1 A flow chart showing a method for determining a line status provided by an embodiment of the present application is shown;

[0044] Figure 2 A schematic diagram showing the structure of a deep network model of a method for determining a line state provided in an embodiment of the present application;

[0045] Figure 3 A flowchart showing another method for determining line status provided by an embodiment of the present application is shown;

[0046] Figure 4A A schematic diagram showing the structure of a device for determining a line state provided in an embodiment of the present application is shown;

[0047] Figure 4B A schematic diagram showing the structure of a device for determining a line state provided in an embodiment of the present application is shown;

[0048] Figure 4C A schematic diagram showing the structure of a device for determining a line state provided in an embodiment of the present application is shown;

[0049] Figure 5 A schematic diagram of the device structure of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0050] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0051] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but merely as an example of an embodiment. Other modifications within the scope and spirit of the present application will occur to those skilled in the art.

[0052] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0053] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0054] It should also be understood that although the present application has been described with reference to certain specific examples, those skilled in the art will readily be able to implement many other equivalent forms of the present application.

[0055] The above and other aspects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.

[0056] Specific embodiments of the present application will be described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments described are merely examples of the present application and may be implemented in a variety of ways. Familiar and / or repetitive functions and structures are not described in detail to avoid obscuring the present application with unnecessary or redundant details. Therefore, the specific structural and functional details described herein are not intended to be limiting, but rather serve merely as a basis and representative basis for the claims to teach those skilled in the art to variously utilize the present application with substantially any suitable detailed structure.

[0057] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," which may all refer to one or more of the same or different embodiments according to the present application.

[0058] The present application embodiment provides a method for determining line status, such as Figure 1 Shown, including:

[0059] 101. Obtain a training dataset.

[0060] In an embodiment of the present application, it is necessary to first obtain a training data set including multiple inspection feature data, that is, the training data set includes multiple line status categories and multiple target historical inspection feature data corresponding to each line status category in the multiple line status categories.

[0061] 102. Establish a deep network model and train the deep network model based on the training data set until the model converges to obtain the target deep network model.

[0062] In an embodiment of the present application, after obtaining a training data set, a deep network model is established and initialized, and then the training value of the deep network model is converged based on the training data set to obtain a target deep network model.

[0063] It should be noted that if Figure 2As shown, the deep network model consists of three parts: an input layer, hidden layers, and an output layer. The input layer is divided into image input and non-image feature input. The image input layer's vector dimension is [3, 32, 32], where 32 represents the number of pixels in the image's height and width, and the number of channels for the 3-bit RGB primary colors. The non-image input layer contains four neurons with dimension 1, corresponding to the following four features: the route's region ID, rainfall over the past week, average temperature over the past week, and average wind speed over the past week. The hidden layer contains 8211 neurons, some of which come from the image input after convolution with the convolution kernel and flattening, with a dimension of 8192, and some from the non-image input for embedding, with a dimension of 19. The convolution kernel parameters are: 8, size 3*3*3, stride 1, max-pooling, and padding. Therefore, the image output after convolution is 32*32*8, and the vector dimension after flattening is 8192. The output layer contains four neurons, corresponding to the four labels: normal, broken conductor, damaged insulator, and contaminated insulator. Let x1 be the image input, x2 be the non-image input, Conv be the convolution function, W12 be the weight corresponding to the non-image features from the input layer to the hidden layer, and Concat be the vector concatenation function. Then, the hidden layer h1 = Concat(Flattening(Conv(x1)), W12*x2), the weights from the hidden layer to the output layer are W2 and b2, and the activation function of the output layer is Softmax. The output of the output layer p = Softmax(W2*h1+b2), where p is a 4-dimensional vector, each dimension representing a probability value between 0 and 1, corresponding to the four labels (i.e., line status categories) of normal, broken conductor, damaged insulator, and contaminated insulator. The label category corresponding to the highest probability value is the line loss category corresponding to that sample (which can be understood as the prediction dataset prediction sample set), i.e., the line status.

[0064] 103. Obtain a prediction data set, input the prediction data set into a target deep network model, and obtain a target line state of the line to be determined.

[0065] In an embodiment of the present application, after obtaining the target deep network model, a prediction data set including multiple current inspection feature data for indicating the line status of the line to be determined is obtained, and the prediction data set is input into the target deep network model, so as to obtain the target line status of the line to be determined output by the target deep network model.

[0066] It should be noted that an initial prediction data set can also be obtained, which includes multiple initial inspection feature data for indicating the line status of the line to be determined. The initial prediction data set can then be processed, for example, the format of the initial prediction data set can be sorted out and unified. For example, if the characteristic value indicated by the target characteristic indicator in the initial inspection feature data is missing, for discrete target characteristic indicators, the corresponding initial inspection feature data can be deleted from the initial prediction data set, and for continuous target characteristic indicators, the interpolation method can be used to supplement the characteristic value corresponding to the target characteristic indicator.

[0067] The method provided in the embodiment of the present application first obtains a training data set, then establishes a deep network model, trains the deep network model based on the training data set until the model converges, and obtains a target deep network model, and finally obtains a prediction data set, inputs the prediction data set into the target deep network model, and obtains the target line state of the line to be determined; by using a training data set including multiple historical inspection feature data to train the deep network model until convergence, a target deep network model that can accurately judge the line state can be trained, and the prediction data set including multiple current inspection feature data is input into the target deep network model, and the target line state of the line to be determined can be accurately predicted. This method has high reusability and can effectively improve the accuracy of line loss judgment.

[0068] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation of this embodiment, the embodiment of the present application provides another method for determining the line status, such as Figure 3 Shown, including:

[0069] 201. Obtain an initial training data set, and process the initial training data set to obtain a training data set.

[0070] In an embodiment of the present application, a plurality of historical line inspection images are first obtained, and the regional identity identification number, rainfall information, temperature information, and wind information of the area to which each historical line inspection image corresponds in the plurality of historical line inspection images are obtained. Subsequently, each historical line inspection image in the plurality of historical line inspection images is combined with the regional identity identification number, rainfall information, temperature information, and wind information of the area to which the corresponding line belongs to obtain a plurality of historical inspection feature data. Subsequently, a plurality of line status categories are obtained, and the plurality of historical inspection feature data are labeled using the plurality of line status categories to obtain a plurality of labeled historical inspection feature data. Finally, an initial training data set is obtained based on the plurality of labeled historical inspection feature data. In other words, the massive amount of data that no one in the power grid platform obtains is effectively utilized to form an initial training data set with a large amount of data. As we all know, the amount of data often restricts the effect of model training.

[0071] It should be noted that the multiple line status categories include line fault conditions such as normal, broken conductor, damaged insulator, and contaminated insulator.

[0072] Furthermore, after obtaining the initial training data set, the initial training data set needs to be processed, specifically: first, identify at least one first historical inspection feature data in the initial training data set that is missing a discrete feature indicator, which can be understood as traversing each feature of each data to confirm whether it is missing, so that all missing data can be identified, delete at least one first historical inspection feature data from the initial training data set to obtain an intermediate training data set, and identify at least one second historical inspection feature data in the intermediate training data set that is missing a continuous feature indicator, obtain identification information, and use the identification information to mark the at least one second historical inspection feature data to obtain at least one marked first historical inspection feature data. Second historical inspection feature data, based on the at least one second historical inspection feature data after labeling, an updated intermediate training data set is obtained, and then the updated intermediate training data set is normalized with continuous feature indicators to obtain a processed updated intermediate training data set, and then at least one to-be-filled historical inspection feature data corresponding to the at least one second historical inspection feature data is extracted from the processed updated intermediate training data set according to the identification information, and the Lagrange interpolation method is used to fill the missing values ​​of each to-be-filled historical inspection feature data in the at least one to-be-filled historical inspection feature data to obtain at least one target historical inspection feature data, and finally a training data set is obtained based on the at least one target historical inspection feature data.

[0073] It should be noted that each piece of historical inspection feature data should include the historical line inspection image and the regional identification number, rainfall information, temperature information, and wind information of the area to which the corresponding line belongs. Then, in the actual training data set, some sample data will be incomplete. For example, the historical inspection feature data lacks discrete feature indicators. This discrete feature indicator can be understood as the regional identification number, that is, the data is discrete. At this time, the historical inspection feature data that lacks discrete feature indicators needs to be deleted from the initial training data set. Another case is that the historical inspection feature data lacks continuous feature indicators. This continuous feature indicator can be understood as rainfall information, temperature information, and wind information. These data have a common feature, that is, they are continuous. For historical inspection feature data that lack continuous feature indicators, interpolation can be used to fill them. The historical inspection feature data adds environmental features, so that the target deep network model trained based on this environmental feature will have a higher recognition accuracy in subsequent image recognition. Among them, the normalization formula used can be expressed as:

[0074]

[0075] in, is the eigenvalue indicated by the normalized characteristic index, x j is the data value indicated by a feature indicator in the j-th historical inspection feature data in the training data set, x max is the maximum value indicated by the feature index in the training data set, x min The minimum value of the feature indicator in the training dataset. For example, taking temperature as an example, the temperature value in the 11th historical training feature data in the training dataset is 28°, while the maximum temperature value of all data in the training dataset is 36° and the minimum temperature value is 0°. The normalized temperature value in the 11th historical training feature data is:

[0076]

[0077] The above method can be used for all data in the training data set, and the characteristic values ​​indicated by the relevant characteristic indicators are mapped to the interval of 0-1. Other characteristic indicators, such as pressure, precipitation, etc., can be normalized using the above method to obtain the normalized training data set.

[0078] 202. Establish a deep network model and train the deep network model based on the training data set until the model converges to obtain the target deep network model.

[0079] In the implementation of this application, after establishing the deep network model, the training data set is processed using one-hot encoding to obtain a target training data set, which includes multiple learning target labels and target historical inspection feature data corresponding to each of the multiple learning target labels; the deep network model is trained using the target training data set until the model converges and the target model parameters are determined; and the target deep network model is obtained based on the target model parameters.

[0080] In an optional implementation of an embodiment of the present application, a target training data set is used to train a deep network model until the model converges, and a specific method for determining the target model parameters is as follows: dividing the target training data set into multiple training subsets, wherein each training subset in the multiple training subsets includes the same number of labeled target historical inspection feature data; selecting any training subset from the multiple training subsets, inputting the selected training subset into the deep network model, and obtaining the line loss fault probability corresponding to each labeled target historical inspection feature data output by the deep network model; obtaining the loss function of the deep network model, and extracting the learning target label corresponding to each labeled target historical inspection feature data from the training subset; calculating the loss value corresponding to the deep network model using the loss function based on the learning target label and line loss fault probability corresponding to each labeled target historical inspection feature data; when it is detected that the loss value does not exceed the target threshold, determining that the model has converged, and using the current model parameters of the deep network model as the target model parameters.

[0081] Furthermore, the loss function of the deep network model is:

[0082]

[0083] Among them, loss is the loss value of the loss function, N is the total number of labeled target historical inspection feature data included in the selected data training subset, i is the i-th labeled target historical inspection feature data in the training subset, c is the current classification, M is multiple learning target labels, y ic is the label value corresponding to the target historical inspection feature data after the i-th annotation under the current classification, p ic is the line loss fault probability corresponding to the target historical inspection feature data after the i-th annotation under the current classification. For example, in this application, there are multiple line status categories: normal, broken conductor, damaged insulator, and contaminated insulator. That is, one normal and three abnormal. The corresponding multiple learning target labels can be [0, 1, 0, 0]. If M = 1, when predicting, the second one in [0, 1, 0, 0] is predicted as 1, and the others are also predicted as 0. All 1 to M are summed, and this is learned simultaneously.

[0084] In another optional implementation of the embodiment of the present application, when it is detected that the loss value exceeds the target threshold, the weight value gradient of the deep network model is calculated based on the loss value using the stochastic gradient descent method; the preset model learning rate of the deep network model is obtained, and the weight value of the deep network model is updated based on the weight value gradient and the preset model learning rate to obtain an updated deep network model, and a training subset other than the training subset is selected from multiple training subsets, and the selected other training subsets are input into the updated deep network model to obtain the line loss failure probability corresponding to each annotated target historical inspection feature data in the other training subsets output by the updated deep network model, and the updated loss value corresponding to the updated deep network model is calculated using the loss function based on the line loss failure probability corresponding to each annotated target historical inspection feature data in the other training subsets, until the updated loss value does not exceed the target threshold, the model convergence is determined, and the current model parameters of the updated deep network model are used as the target model parameters. That is to say, the target threshold can be understood as the convergence condition, and the loss value exceeding the target threshold can be understood as the deep network model not meeting the convergence condition. At this time, the weight value gradient of the deep network model is calculated based on the loss value, and the weight value in the deep network model is updated based on the weight value gradient and the preset model learning rate. The deep network model with updated weight value is used to train the next batch of learning sample data until the deep network model converges.

[0085] It should be noted that after the network parameters are randomly initialized, the stochastic gradient descent method is used to optimize the loss. The specific method is to calculate the gradient of the loss for each weight value of the model after the forward propagation is completed. Add the gradient to the required gradient multiplied by the learning rate, such as the weight w i The update method is:

[0086] w i =w i +lr*grad wi ,

[0087] Among them, lr is the learning rate, which can be set to 0.001.

[0088] Furthermore, in the actual application process, the number of training times can also be determined. A training subset is randomly selected from multiple training subsets, and the deep network model is repeatedly trained using the training subset according to the number of training times. If the number of training times is 10 times, then 10 loss rates will be obtained. The average loss rate is calculated based on these loss rates. When it is detected that the average loss rate is less than the target threshold, the model is determined to have converged. The target threshold can be selected as 0.001. The advantage of this is that the judgment volatility is small and more stable.

[0089] 203. Obtain a prediction data set, input the prediction data set into a target deep network model, and obtain a target line state of the line to be determined.

[0090] In actual application, by integrating the trained target deep network model into the drone inspection data processing process, an accurate prediction of fault results for the inspection line can be made continuously 24 hours a day. If a corresponding fault is detected, the fault type can be quickly displayed on the staff's device for easy viewing. The data preprocessing process for the prediction sample data can be referred to the data preprocessing process for the training dataset above. In addition, based on the existing converged model, the above model training process can be executed with newly collected data, and only the existing model construction can be used, further improving accuracy.

[0091] Through the above technical solution, the massive data in the power grid platform is effectively utilized; on the one hand, the algorithm judges the line status based on the convolutional network with obvious advantages in image data and image information extraction; on the other hand, it incorporates different environmental characteristics in different regions, which will further improve the accuracy of image recognition compared to general pure convolutional networks; this method is low-cost and highly reusable, greatly saving manpower and material resources in judging drone inspection results.

[0092] The method provided in the embodiment of the present application first obtains an initial training data set, processes the initial training data set to obtain a training data set, then establishes a deep network model, trains the deep network model based on the training data set until the model converges to obtain a target deep network model, then obtains a prediction data set, inputs the prediction data set into the target deep network model, and obtains a target line state of the line to be determined. By using a training data set including multiple historical inspection feature data to train the deep network model until convergence, a target deep network model that can accurately judge the line state can be trained. By inputting the prediction data set including multiple current inspection feature data into the target deep network model, the target line state of the line to be determined can be accurately predicted. This method has high reusability and can effectively improve the accuracy of line loss judgment.

[0093] Further, as Figure 1 The specific implementation of the method is as follows: Figure 4A As shown, an embodiment of the present invention provides a device for determining a line status, including: an acquisition module 401 , a training module 402 and an input module 403 .

[0094] The acquisition module 401 is used to acquire a training data set, wherein the training data set includes multiple line status categories and multiple target historical inspection feature data corresponding to each line status category in the multiple line status categories;

[0095] The training module 402 is used to establish a deep network model and train the deep network model based on the training data set until the model converges to obtain a target deep network model;

[0096] The input module 403 is used to obtain a prediction data set and input the prediction data set into the target deep network model to obtain the target line state of the line to be determined, wherein the prediction data set includes multiple current inspection feature data used to indicate the line state of the line to be determined.

[0097] In specific application scenarios, such as Figure 4B As shown, the device further includes: a marking module 404.

[0098] The labeling module 404 is used to obtain multiple historical line inspection images, and obtain the regional identity identification number, rainfall information, temperature information, and wind force information of the area to which the line belongs corresponding to each historical line inspection image in the multiple historical line inspection images; combine each historical line inspection image in the multiple historical line inspection images with the regional identity identification number, rainfall information, temperature information, and wind force information of the area to which the line belongs to obtain multiple historical inspection feature data; obtain the multiple line status categories, and use the multiple line status categories to label the multiple historical inspection feature data to obtain multiple labeled historical inspection feature data; and obtain an initial training data set based on the multiple labeled historical inspection feature data.

[0099] In specific application scenarios, such as Figure 4C As shown, the device further includes: a processing module 405.

[0100] The processing module 405 is used to identify at least one first historical inspection feature data with missing discrete feature indicators in the initial training data set, delete the at least one first historical inspection feature data from the initial training data set, and obtain an intermediate training data set; identify at least one second historical inspection feature data with missing continuous feature indicators in the intermediate training data set, obtain identification information, use the identification information to label the at least one second historical inspection feature data to obtain at least one labeled second historical inspection feature data, and obtain an updated intermediate training data set based on the at least one labeled second historical inspection feature data; normalize the continuous feature indicators of the updated intermediate training data set to obtain a processed updated intermediate training data set; extract at least one to-be-filled historical inspection feature data corresponding to the at least one second historical inspection feature data from the processed updated intermediate training data set according to the identification information, use the Lagrange interpolation method to fill the missing values ​​of each to-be-filled historical inspection feature data in the at least one to-be-filled historical inspection feature data to obtain at least one target historical inspection feature data; and obtain the training data set based on the at least one target historical inspection feature data.

[0101] In a specific application scenario, the training module 402 is also used to: process the training data set using one-hot encoding to obtain a target training data set, wherein the target training data set includes multiple learning target labels and target historical inspection feature data corresponding to each of the multiple learning target labels; train the deep network model using the target training data set until the model converges and determines the target model parameters; and obtain the target deep network model based on the target model parameters.

[0102] In a specific application scenario, the training module 402 is also used to: divide the target training data set into multiple training subsets, wherein each training subset in the multiple training subsets includes the same number of labeled target historical inspection feature data; select any training subset from the multiple training subsets, input the selected training subset into the deep network model, and obtain the line loss fault probability corresponding to each labeled target historical inspection feature data output by the deep network model; obtain the loss function of the deep network model, and extract the learning target label corresponding to each labeled target historical inspection feature data from the training subset; calculate the loss value corresponding to the deep network model using the loss function based on the learning target label and line loss fault probability corresponding to each labeled target historical inspection feature data; when it is detected that the loss value does not exceed the target threshold, determine that the model has converged, and use the current model parameters of the deep network model as the target model parameters.

[0103] In a specific application scenario, the training module 402 is further used to: the loss function of the deep network model is:

[0104]

[0105] Among them, loss is the loss value of the loss function, N is the total number of labeled target historical inspection feature data included in the selected data training subset, i is the i-th labeled target historical inspection feature data in the training subset, c is the current classification, M is the multiple learning target labels, yic is the label value corresponding to the i-th labeled target historical inspection feature data under the current classification, and pic is the line loss fault probability corresponding to the i-th labeled target historical inspection feature data under the current classification.

[0106] In a specific application scenario, the training module 402 is also used to: when it is detected that the loss value exceeds the target threshold, calculate the weight value gradient of the deep network model based on the loss value using the stochastic gradient descent method; obtain the preset model learning rate of the deep network model, and update the weight value of the deep network model based on the weight value gradient and the preset model learning rate to obtain an updated deep network model, select another training subset other than the training subset from the multiple training subsets, input the selected other training subset into the updated deep network model, obtain the line loss failure probability corresponding to each annotated target historical inspection feature data in the other training subset output by the updated deep network model, calculate the updated loss value corresponding to the updated deep network model using the loss function based on the line loss failure probability corresponding to each annotated target historical inspection feature data in the other training subsets, until the updated loss value does not exceed the target threshold, determine model convergence, and use the current model parameters of the updated deep network model as the target model parameters.

[0107] The device provided in the embodiment of the present application first obtains a training data set through an acquisition module, then establishes a deep network model through a training module, trains the deep network model based on the training data set until the model converges to obtain a target deep network model, and finally obtains a prediction data set through an input module, inputs the prediction data set into the target deep network model to obtain a target line state of the line to be determined; by using a training data set including multiple historical inspection feature data to train the deep network model until convergence, a target deep network model that can accurately judge the line state can be trained, and by inputting the prediction data set including multiple current inspection feature data into the target deep network model, the target line state of the line to be determined can be accurately predicted. This method has high reusability and can effectively improve the accuracy of line loss judgment.

[0108] It should be noted that for other corresponding descriptions of the functional units involved in the apparatus for determining the line status provided in the embodiment of the present invention, reference can be made to Figure 1 and Figures 4A to 4C The corresponding description in will not be repeated here.

[0109] In an exemplary embodiment, see Figure 5 A computer device is also provided, comprising a bus, a processor, a memory, and a communication interface. The computer device may also include an input / output interface and a display device, wherein the various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor is configured to execute the program stored in the memory and perform the line status determination method described in the above embodiment.

[0110] A computer-readable storage medium stores a computer program, which implements the steps of the line status determination method when executed by a processor.

[0111] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented through hardware or by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0112] Those skilled in the art will understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application.

[0113] Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the implementation scenario description, or can be modified accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module or further split into multiple submodules.

[0114] The above application serial numbers are for description only and do not represent the advantages or disadvantages of the implementation scenarios.

[0115] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.

Claims

1. A method for determining line status, characterized in that: include: Acquire a plurality of historical line inspection images, and acquire a regional identification number, rainfall information, temperature information, and wind speed information of a line area corresponding to each of the plurality of historical line inspection images; Combining each of the plurality of historical line inspection images with a regional identification number, rainfall information, temperature information, and wind speed information of the corresponding line area to obtain a plurality of historical inspection feature data; Acquire the multiple line status categories, and use the multiple line status categories to label the multiple historical inspection feature data to obtain multiple labeled historical inspection feature data; Obtaining an initial training data set based on the plurality of annotated historical inspection feature data; Acquire a training data set, the training data set including a plurality of line status categories and a plurality of target historical inspection feature data corresponding to each of the plurality of line status categories; Establishing a deep network model, training the deep network model based on the training data set until the model converges, and obtaining a target deep network model, wherein the deep network model includes an input layer, a hidden layer, and an output layer, and the input layer includes an image input and a non-image feature input; A prediction data set is obtained, and the prediction data set is input into the target deep network model to obtain a target line state of the line to be determined, wherein the prediction data set includes a plurality of current inspection feature data for indicating the line state of the line to be determined.

2. The method for determining line status according to claim 1, wherein: After obtaining the initial training data set based on the plurality of annotated historical inspection feature data, the method further includes: Identifying at least one first historical inspection feature data with a missing discrete feature index in the initial training data set, deleting the at least one first historical inspection feature data from the initial training data set to obtain an intermediate training data set; Identifying at least one second historical inspection feature data in the intermediate training data set that is missing a continuous feature indicator, obtaining identification information, labeling the at least one second historical inspection feature data using the identification information to obtain the at least one labeled second historical inspection feature data, and obtaining an updated intermediate training data set based on the at least one labeled second historical inspection feature data; performing normalization processing of continuous characteristic indicators on the updated intermediate training data set to obtain a processed updated intermediate training data set; extracting at least one to-be-filled historical inspection feature data corresponding to the at least one second historical inspection feature data from the processed updated intermediate training data set according to the identification information, and filling in missing values ​​of each to-be-filled historical inspection feature data in the at least one to-be-filled historical inspection feature data using a Lagrange interpolation method to obtain at least one target historical inspection feature data; The training data set is obtained based on the at least one target historical inspection feature data.

3. The method for determining line status according to claim 1, wherein: The deep network model is trained based on the training data set until the model converges to obtain a target deep network model, including: The training data set is processed using one-hot encoding to obtain a target training data set, wherein the target training data set includes a plurality of learning target labels and target historical inspection feature data corresponding to each of the plurality of learning target labels; Training the deep network model using the target training data set until the model converges and determining target model parameters; The target deep network model is obtained based on the target model parameters.

4. The method for determining line status according to claim 3, wherein: The method of training the deep network model using the target training data set until the model converges and determining target model parameters includes: Dividing the target training data set into multiple training subsets, wherein each of the multiple training subsets includes the same amount of annotated target historical inspection feature data; Selecting any one training subset from the multiple training subsets, inputting the selected training subset into the deep network model, and obtaining the line loss fault probability corresponding to each labeled target historical inspection feature data output by the deep network model; Obtaining a loss function of the deep network model, and extracting a learning target label corresponding to each labeled target historical inspection feature data from the training subset; Based on the learning target label and line loss failure probability corresponding to each labeled target historical inspection feature data, the loss value corresponding to the deep network model is calculated using the loss function; When it is detected that the loss value does not exceed the target threshold, the model is determined to have converged, and the current model parameters of the deep network model are used as the target model parameters.

5. The method for determining line status according to claim 4, characterized in that: The loss function of the deep network model is: Wherein, loss is the loss value of the loss function, N is the total number of labeled target historical inspection feature data included in the selected data training subset, i is the i-th labeled target historical inspection feature data in the training subset, c is the current classification, M is the multiple learning target labels, y ic is the label value corresponding to the target historical inspection feature data after the i-th annotation under the current classification, p ic is the line loss failure probability corresponding to the target historical inspection feature data after the i-th annotation under the current classification.

6. The method for determining line status according to claim 4, characterized in that: When it is detected that the loss value exceeds the target threshold, calculating the weight value gradient of the deep network model based on the loss value using the stochastic gradient descent method; Obtain a preset model learning rate of the deep network model, and update the weight value of the deep network model based on the weight value gradient and the preset model learning rate to obtain an updated deep network model, select a training subset other than the training subset from the multiple training subsets, input the selected other training subset into the updated deep network model, obtain the line loss failure probability corresponding to each labeled target historical inspection feature data in the other training subset output by the updated deep network model, calculate the updated loss value corresponding to the updated deep network model using the loss function based on the line loss failure probability corresponding to each labeled target historical inspection feature data in the other training subset, until the updated loss value does not exceed the target threshold, determine model convergence, and use the current model parameters of the updated deep network model as the target model parameters.

7. A device for determining line status, characterized in that: include: a labeling module for acquiring a plurality of historical line inspection images and acquiring a regional identification number, rainfall information, temperature information, and wind speed information of a line area corresponding to each of the plurality of historical line inspection images; Combining each of the plurality of historical line inspection images with a regional identification number, rainfall information, temperature information, and wind speed information of the corresponding line area to obtain a plurality of historical inspection feature data; Acquire the multiple line status categories, and use the multiple line status categories to label the multiple historical inspection feature data to obtain multiple labeled historical inspection feature data; Obtaining an initial training data set based on the plurality of annotated historical inspection feature data; An acquisition module is used to acquire a training data set, wherein the training data set includes multiple line status categories and multiple target historical inspection feature data corresponding to each line status category in the multiple line status categories; A training module is used to establish a deep network model, train the deep network model based on the training data set until the model converges, and obtain a target deep network model, wherein the deep network model includes an input layer, a hidden layer, and an output layer, wherein the input layer includes image input and non-image feature input; An input module is used to obtain a prediction data set, input the prediction data set into the target deep network model, and obtain a target line state of the line to be determined, wherein the prediction data set includes multiple current inspection feature data for indicating the line state of the line to be determined.

8. 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.

9. A 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