Power distribution room defect identification method and device based on artificial intelligence

Abstract

The invention relates to the technical field of power equipment detection, in particular to a power distribution room defect identification method and device based on artificial intelligence, and the method comprises the steps: collecting the existing defect sample data of a power distribution room, and marking the corresponding defect features for the existing defect sample data; fusing the defect features on a defect-free picture of the power distribution room by using an artificial intelligence content generation technology to obtain new defect sample data; and screening the new defect sample data based on a three-dimensional data evaluation strategy to obtain a defect sample training set, and training the power distribution room defect identification model of the current version by using the defect sample training set to obtain a power distribution room defect identification model of an updated version so as to identify the power distribution room defect. And in the model application, a sample set training model is constructed by using missing report and false report data, so that missing report is reduced, false report defects are recognized through secondary classification, false report is reduced, a continuous feedback self-evolution mechanism is formed, and the accuracy and the practical level of the model are improved.

Description

Technical Field

[0001] This application relates to the technical field of power equipment detection, and particularly to a method and device for identifying defects in a distribution substation based on artificial intelligence. Background Art

[0002] With the rapid development of information technology and new energy technology, the digital and intelligent management of distribution substations has become a trend. Some existing technical solutions propose to collect and identify information of distribution substations and judge abnormal problems based on an artificial intelligence analysis and processing system. However, this method has the problem of scarce samples. Due to the large differences in the environments and equipment of different distribution substations, the accuracy of model recognition in actual applications is not high, the problems of missed reports and false alarms are serious, and the technical practicability is poor. Summary of the Invention

[0003] To overcome the deficiencies in the prior art, this application provides a method and device for identifying defects in a distribution substation based on artificial intelligence, which can improve the accuracy and practicability of intelligent patrol recognition by expanding samples.

[0004] In a first aspect, this application provides a method for identifying defects in a distribution substation based on artificial intelligence, and the method includes the following steps:

[0005] Collect the existing defect sample data of the distribution substation, label the corresponding defect features for the existing defect sample data, and use artificial intelligence content generation technology to fuse the defect features on the defect-free pictures of the distribution substation to obtain new defect sample data;

[0006] Based on a three-dimensional data evaluation strategy, screen the new defect sample data, merge the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and use the defect sample training set to train the current version of the distribution substation defect recognition model to obtain an updated version of the distribution substation defect recognition model for identifying distribution substation defects.

[0007] In a possible implementation manner, the method further includes the following steps:

[0008] Regularly obtain the missed report / false alarm data of the updated version of the distribution substation defect recognition model during actual operation, label the defect features for the missed report / false alarm data, and use artificial intelligence content generation technology to fuse the defect features of the missed report / false alarm on the defect-free pictures of the distribution substation to obtain missed report / false alarm defect sample data;

[0009] Based on a three-dimensional data evaluation strategy, screen the new defect sample data, and use the screened missed report / false alarm defect sample data to train the updated version of the distribution substation defect recognition model to obtain a self-evolving version of the distribution substation defect recognition model for identifying distribution substation defects.

[0010] In a possible implementation, the defect features include at least one of flame, accumulated water, breakage, and foreign objects; wherein, the defect features are fused at a specified position on the defect-free picture of the distribution substation by using artificial intelligence content generation technology to obtain new defect sample data; the specified position includes at least one of the bus connection point in the switch cabinet breaker chamber, the cable joint in the cable trench, the positive and negative electrode connection points of the storage battery, and the transformer area.

[0011] In a possible implementation, the screening of the new defect sample data based on the three-dimensional data evaluation strategy includes the following steps:

[0012] Input the new defect sample data into the current version of the distribution substation defect recognition model for recognition, and calculate the credibility rate, difference value, and entropy value of each new defect sample;

[0013] Calculate the confidence level of each new defect sample based on the credibility rate, difference value, and entropy value to obtain a confidence matrix, and perform standard deviation processing on the confidence matrix to obtain a standardized confidence level;

[0014] Screen out the new defect samples corresponding to the standardized confidence level less than the first set threshold for merging with the existing defect sample data to obtain a defect sample training set.

[0015] In a possible implementation, the confidence level is calculated by the following formula:

[0016] V i =α1hd i +α2mt i +(1 - α1 - α2)es i

[0017] V i represents the confidence level of the i-th new defect sample, hd i represents the credibility rate, mt i represents the difference value, es i represents the entropy value, and α1, α2 are set parameters used to represent the weights of the credibility rate, difference value, and entropy value.

[0018] In a possible implementation, the identification of distribution substation defects by using the self-evolving version of the distribution substation defect recognition model is carried out in the following way, including the following steps:

[0019] Input the screened false alarm defect sample data into the previous self-evolving version of the distribution substation defect recognition model, obtain the output probability result matrix and calculate its probability mean and standard deviation, and calculate the false alarm index according to the current target recognition probability and the calculated probability mean and standard deviation;

[0020] Input the false alarm defect samples corresponding to the false alarm index less than the second set threshold into the current self-evolving version of the distribution substation defect recognition model for secondary recognition, and determine whether to give an alarm according to the recognition result.

[0021] In a possible implementation manner, the false alarm index is calculated by the following formula:

[0022]

[0023] where p represents the current target recognition probability, μ represents the probability mean, and σ represents the standard deviation.

[0024] In a second aspect, the present application provides a distribution substation defect recognition device based on artificial intelligence. The device includes:

[0025] A sample expansion module, configured to collect existing defect sample data of the distribution substation, label corresponding defect features for the existing defect sample data, and fuse the defect features on the defect-free pictures of the distribution substation by using artificial intelligence content generation technology to obtain new defect sample data;

[0026] A defect recognition module, configured to screen the new defect sample data based on a three-dimensional data evaluation strategy, merge the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and use the defect sample training set to train the current version of the distribution substation defect recognition model to obtain an updated version of the distribution substation defect recognition model for identifying distribution substation defects.

[0027] In a third aspect, the present application provides an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps of the artificial intelligence-based distribution substation defect recognition method described in any one of the first aspects are executed.

[0028] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of the artificial intelligence-based distribution substation defect recognition method described in any one of the first aspects are executed.

[0029] A method and device for identifying defects in a distribution substation based on artificial intelligence provided by this embodiment collect existing defect sample data of the distribution substation, label corresponding defect features for the existing defect sample data, and use artificial intelligence content generation technology to fuse the defect features on defect-free pictures of the distribution substation to obtain new defect sample data; screen the new defect sample data based on a three-dimensional data evaluation strategy, and merge the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and use the defect sample training set to train the current version of the distribution substation defect identification model to obtain an updated version of the distribution substation defect identification model for identifying distribution substation defects.

[0030] Furthermore, regularly obtain the false negative / false positive data of the updated version of the distribution substation defect identification model during actual operation, label the defect features for the false negative / false positive data, and use artificial intelligence content generation technology to fuse the false negative / false positive defect features on defect-free pictures of the distribution substation to obtain false negative / false positive defect sample data; screen the new defect sample data based on a three-dimensional data evaluation strategy, and use the screened false negative / false positive defect sample data to train the updated version of the distribution substation defect identification model to obtain a self-evolving version of the distribution substation defect identification model for identifying distribution substation defects.

[0031] Thus, by expanding the existing samples, the applicability of the model is improved in the initial stage of training, the false negative / false positive rate is reduced in actual applications, a self-evolving mechanism is formed, and the accuracy and practical level of the model are improved. Description of the Drawings

[0032] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0033] Figure 1 Shows the flowchart of the method for identifying defects in a distribution substation based on artificial intelligence according to an embodiment of the present application;

[0034] Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 Shows the schematic diagram of using artificial intelligence content generation technology to fuse the defect features at a specified position on defect-free pictures of the distribution substation to obtain new defect sample data;

[0035] Figure 6Shows a flowchart of screening the new defect sample data based on the three-dimensional data evaluation strategy according to an embodiment of the present application;

[0036] Figure 7 , Figure 8 Shows a schematic diagram of using artificial intelligence content generation technology to fuse the defect features of missed reports / false alarms on the defect-free pictures in the power distribution room to obtain the missed report / false alarm defect sample data according to an embodiment of the present application;

[0037] Figure 9 Shows a flowchart of using a self-evolving version of the power distribution room defect recognition model to recognize power distribution room defects according to an embodiment of the present application;

[0038] Figure 10 Shows a schematic structural diagram of a power distribution room defect recognition device based on artificial intelligence according to an embodiment of the present application;

[0039] Figure 11 Shows a structural block diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. It should be understood that the accompanying drawings in the present application are only for the purposes of illustration and description, and are not used to limit the protection scope of the present application. In addition, it should be understood that the schematic drawings are not drawn to actual scale. The flowcharts used in the present application show the operations implemented according to some embodiments of the present application. It should be understood that the operations in the flowchart may not be implemented in sequence, and steps without logical context relationships may be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of the present application.

[0041] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application to be protected, but only represents the selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative efforts belong to the protection scope of the present application.

[0042] It should be noted that the term "including" will be used in the embodiments of the present application to indicate the existence of the features stated thereafter, but does not exclude the addition of other features.

[0043] In view of the technical problems proposed in the background art, the present application provides an artificial intelligence-based distribution substation defect identification method and device, which can improve the accuracy and practicality of intelligent patrol identification by expanding samples.

[0044] In one embodiment, referring to the attached drawings of the specification Figure 1 The present application provides an artificial intelligence-based distribution substation defect identification method, and the method includes the following steps:

[0045] S1. Collect the existing defect sample data of the distribution substation, label the corresponding defect features for the existing defect sample data, and use artificial intelligence content generation technology to fuse the defect features on the defect-free pictures of the distribution substation to obtain new defect sample data;

[0046] S2. Screen the new defect sample data based on the three-dimensional data evaluation strategy, merge the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and use the defect sample training set to train the current version of the distribution substation defect identification model to obtain an updated version of the distribution substation defect identification model for identifying distribution substation defects;

[0047] S3. Regularly obtain the false negative / false positive data of the updated version of the distribution substation defect identification model during actual operation, label the defect features for the false negative / false positive data, and use artificial intelligence content generation technology to fuse the defect features of the false negative / false positive on the defect-free pictures of the distribution substation to obtain false negative / false positive defect sample data;

[0048] S4. Screen the new defect sample data based on the three-dimensional data evaluation strategy, and use the screened false negative / false positive defect sample data to train the updated version of the distribution substation defect identification model to obtain a self-evolving version of the distribution substation defect identification model for identifying distribution substation defects.

[0049] In step S1, mainly by expanding the existing defect sample data to generate new defect sample data to overcome the problem of scarce abnormal samples in the current artificial intelligence patrol of distribution substations. Specifically, label the defect features of the existing defect sample data, and use artificial intelligence content generation technology to fuse the defect features at the specified positions on the defect-free pictures to obtain new defect sample data.

[0050] For example, first collect a certain number of pictures of substations without fire. These pictures should cover common equipment areas such as switchgear cabinets, battery cabinets, cable trenches of different models, as well as different substation layouts and environmental conditions to ensure the diversity of the samples. At the same time, organize the existing fire defect sample data and carefully label the characteristic information such as the color, shape, size, and position of the flame. Select an artificial intelligence content generation model such as Midjourney or Doubao to fuse and generate the labeled flame characteristics on the pictures of substations without fire. For example, for the battery cabinet area, according to the possible positions and situations of actual fires, generate flame combustion characteristics at the positive and negative connections of the battery, the battery case, etc.; for the circuit breaker room of the switchgear cabinet, generate flames at the positions of components such as contacts and busbars that are prone to heat generation and ignition. During the generation process, manually assist in specifying the generation positions of the flames according to the actual structure and fire risk points of the substation to ensure the authenticity and effectiveness of the generated samples. In one embodiment, the newly generated defect sample data can be referred to in the attached Figures 2 - 5 , where the attached Figure 2 and the attached Figure 3 are sample data with flame defects at the switchgear cabinet in the distribution room; the attached Figure 4 and the attached Figure 5 are sample data with water accumulation defects on the ground in the distribution room.

[0051] In step S2, it is mainly to screen the newly generated defect sample data for model training. Specifically, first, in accordance with the principle that the sample background cannot be modified and only fusion can be superimposed on the defect-free sample, eliminate unreasonable and invalid samples; such as whether the flame has a reasonable interaction with the surrounding equipment and whether there are obvious logical errors. Then, refer to the attached Figure 6 , and screen the newly generated defect sample data based on the three-dimensional data evaluation strategy, including the following steps:

[0052] S201. Input the newly generated defect sample data into the current version of the distribution room defect recognition model for recognition, and calculate the credibility rate, difference value, and entropy value of each new defect sample;

[0053] S202. Calculate the confidence level of each new defect sample based on the credibility rate, difference value, and entropy value to obtain a confidence matrix, and perform standard deviation processing on the confidence matrix to obtain a standardized confidence level;

[0054] S203. Screen out the new defect samples corresponding to the standardized confidence level less than the first set threshold for merging with the existing defect sample data to obtain a defect sample training set.

[0055] That is, the three-dimensional data evaluation strategy described in this application is set based on the credibility rate, difference value, and entropy value. This is because the credibility rate can reflect the model's classification ability and sample scarcity, the difference value can measure the sample uncertainty and information content, and the entropy value can describe the sample information chaos degree and value. Therefore, by evaluating the value of samples for model training from the three dimensions of the credibility rate, difference value, and entropy value, samples that are more helpful for improving the model performance can be selected.

[0056] Specifically, in step S201, the credibility rate is calculated through the following formula:

[0057] hd i = max k p(y i = k|x i ; w)

[0058] hd i represents the credibility rate of the i-th new defect sample, max k is the maximum operator, p(y i = k|x i ; w) represents the probability of being predicted as class k given the new defect sample x i and the model parameter w, and y i is the true classification label of the i-th new defect sample. Among them, selecting a sample with a low credibility rate indicates that the model has a poor classification ability for this sample, and the model encounters fewer such samples during training, so this sample is more valuable.

[0059] The difference value is calculated through the following formula:

[0060] mt i = p(y i = k1|x i ; w) - p(y i = k2|x i ; w)

[0061] mt i represents the difference value of the i-th new defect sample, p(y i = k1|x i ; w) represents the probability of being predicted as class k1 given the new defect sample x i and the model parameter w, and y i is the true classification label of the i-th new defect sample, k1 is the class corresponding to the maximum value in the predicted probability of the model for the i-th sample, and k1, k2 are the classes corresponding to the maximum value and the second maximum value in the predicted probability of the model for the i-th sample. Among them, the difference value sampling refers to the difference between the maximum value and the second maximum value in the classification prediction probability. The smaller this difference value is, the higher the sample uncertainty and the higher the sample information content.

[0062] The entropy value is calculated by the following formula:

[0063]

[0064] es i represents the difference of the i-th new defect sample. m represents the total number of all possible categories when the model classifies the sample. n is the category index to which the model predicts the sample may belong, taking values from 1 to m. p(y i =n|x i ; w) represents the probability that the sample is predicted to be category n given the new defect sample x i and the model parameter w. y i is the true classification label of the i-th new defect sample. Among them, the higher the information entropy value, the more chaotic the sample information and the higher the information content.

[0065] In step S202, the confidence level is further calculated mainly based on the calculated credibility rate, difference and entropy value. The following formula is as follows:

[0066] V i =α1hd i +α2mt i +(1 - α1 - α2)es i

[0067] V i represents the confidence level of the i-th new defect sample. hd i represents the credibility rate. mt i represents the difference. es i represents the entropy value. α1 and α2 are set parameters used to represent the weights of the credibility rate, difference and entropy value. Among them, after evaluating the confidence level of the new defect sample, the confidence level matrix [V i (0 ≤ i ≤ n) of n samples is obtained. The matrix V i is processed by Z-score standard deviation to obtain the standardized confidence level. The formula is as follows:

[0068]

[0069] In step S203, the first set threshold is 1, that is, the new defect samples with Z i <1 are screened out. These samples have poor confidence levels under the existing model, that is, the model has a low certainty in classifying them. However, it also means that they contain more information that the model has not fully learned. Therefore, retaining these samples for subsequent training helps to improve the defect recognition accuracy of the model in different distribution substation room environments and equipment conditions and reduce the problem of inaccurate recognition caused by environmental and equipment differences.

[0070] Further, in step S3, for the missed alarm and false alarm data that occur during the actual operation of the model, by inputting the time period when the missed alarm and false alarm occur, the corresponding original data is accurately screened from the database with the help of a program. Then, the missed alarm and false alarm data are carefully labeled and classified manually. For example, the red light on the secondary instrument panel of the switch cabinet in the false alarm data is labeled as an indicator light, and the glass reflection of the battery cabinet glass is labeled as glass; for the missed alarm data, the target defect features that are not recognized due to environmental and background influences are labeled.

[0071] Also using artificial intelligence content generation technology, the labeled missed alarm and false alarm defects are fused and generated on defect-free pictures to construct a large number of sample data sets with missed alarm and false alarm defects. For example, red indicator light samples are generated on the secondary instrument panels of different models of switch cabinets, and glass reflection samples are generated on different models of battery cabinets / remote control cabinets, etc. In one embodiment, the missed alarm defect samples generated by fusing the defect of the switch cabinet on fire in other defect-free substation building pictures can be seen in the attached Figure 7 ; the false alarm defect samples generated by fusing the false alarm defect feature of the panel red light misreported as on fire on the defect-free substation building pictures can be seen in the attached Figure 8 .

[0072] In step S4, for the generated missed alarm / false alarm defect sample data, the above-mentioned three-dimensional data evaluation strategy is used for filtering and screening to ensure the sample quality. And the current version model M q is trained with the filtered missed alarm / false alarm defect sample data to obtain the new version model M q+1 , and at the same time, version M q is retained for subsequent comparative analysis and backtracking.

[0073] Among them, to further solve the false alarm problem, the false alarm data is reclassified. Specifically, see the attached Figure 9 , and use the self-evolving version of the distribution room defect recognition model to identify distribution room defects, including the following steps:

[0074] S401. Input the filtered false alarm defect sample data into the previous self-evolving version of the distribution room defect recognition model, obtain the output probability result matrix and calculate its probability mean and standard deviation, and calculate the false alarm index according to the current target recognition probability and the calculated probability mean and standard deviation;

[0075] S402. Input the false alarm defect samples corresponding to the false alarm index less than the second set threshold into the current self-evolving version of the distribution room defect recognition model for secondary recognition, and determine whether to give an alarm according to the recognition result.

[0076] In step S401, input the false alarm defect sample data into model Mq Perform recognition and output the probability result matrix [p i (0 ≤ i ≤ n), where n is the number of samples, and calculate its probability mean μ and standard deviation σ; and in the actual on-site application scenario, calculate and analyze the target recognition probability p in real time, and calculate the false alarm index H through the formula Calculate the false alarm index H.

[0077] In step S402, the second set threshold is 2. When H < 2, it is determined that the false alarm probability is relatively high. At this time, send the false alarm defect sample into the model M q+1 Perform secondary recognition. If the model M q+1 The recognition result is a known false alarm result (such as a red light, glass, etc.), record the recognition result but do not send an alarm signal to avoid unnecessary alarm interference; if the recognition result is not a known false alarm result, it is determined that it may be a real abnormal situation, and the system continues to trigger an alarm to remind the staff to handle potential problems in time to ensure the safe and stable operation of the substation building.

[0078] For the missing alarm defect sample data, directly use the model M q+1 Perform recognition to complete the defect recognition types of the previous version model and reduce the missing alarm rate.

[0079] It can be seen that a method for identifying defects in a distribution substation based on artificial intelligence provided by this application, in the initial stage of model training, based on a small amount of defect sample data, uses artificial intelligence content generation technology to fuse and generate defect features on defect-free distribution substation pictures in different environments and equipment, realizing simple, fast, and low-cost defect sample expansion, and improving the applicability of the model in different distribution substation scenarios. In the model application, using the missing alarm and false alarm data that appear, use artificial intelligence technology to fuse and generate missing alarm and false alarm defect features on defect-free distribution substation pictures in different environments and equipment, construct a missing alarm and false alarm defect sample set, train the model to improve the recognition accuracy and reduce the missing alarm, perform secondary classification on the model training to identify false alarm defects, reduce the false alarm, and form a continuous feedback self-evolution mechanism based on the missing alarm and false alarm data, and gradually improve the model accuracy and practical level with the application.

[0080] Based on the same inventive concept, an apparatus for identifying defects in a distribution substation based on artificial intelligence is also provided in an embodiment of this application. Since the principle of solving problems by the apparatus in the embodiment of this application is similar to that of the method for identifying defects in a distribution substation based on artificial intelligence in the above embodiment of this application, the implementation of the apparatus can refer to the implementation of the method, and the repeated parts will not be described again.

[0081] As shown in the specification appendix Figure 10 As shown, an apparatus for identifying defects in a distribution substation based on artificial intelligence provided in an embodiment of this application, the apparatus includes:

[0082] The sample augmentation module 1001 is used to collect the existing defect sample data of the distribution substation, label the corresponding defect features for the existing defect sample data, and fuse the defect features on the defect-free pictures of the distribution substation by using artificial intelligence content generation technology to obtain new defect sample data;

[0083] The defect recognition module 1002 is used to screen the new defect sample data based on the three-dimensional data evaluation strategy, merge the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and use the defect sample training set to train the current version of the distribution substation defect recognition model to obtain an updated version of the distribution substation defect recognition model for identifying distribution substation defects.

[0084] In one embodiment, the device further includes:

[0085] The acquisition module 1003 is used to regularly obtain the false negative / false positive data of the updated version of the distribution substation defect recognition model during actual operation, label the defect features for the false negative / false positive data, and fuse the defect features of false negative / false positive on the defect-free pictures of the distribution substation by using artificial intelligence content generation technology to obtain false negative / false positive defect sample data;

[0086] The update module 1004 is used to screen the new defect sample data based on the three-dimensional data evaluation strategy, and use the screened false negative / false positive defect sample data to train the updated version of the distribution substation defect recognition model to obtain a self-evolving version of the distribution substation defect recognition model for identifying distribution substation defects.

[0087] In one embodiment, the defect features include at least one of flame, water accumulation, damage, and foreign objects; wherein, the defect features are fused at the specified positions on the defect-free pictures of the distribution substation by using artificial intelligence content generation technology to obtain new defect sample data; the specified positions include at least one of the bus connection points in the switch cabinet breaker room, the cable joints in the cable trench, the positive and negative battery connection points, and the transformer area.

[0088] In one embodiment, the defect recognition module 1002 screens the new defect sample data based on the three-dimensional data evaluation strategy, including: inputting the new defect sample data into the current version of the distribution substation defect recognition model for recognition, and calculating the credibility rate, difference value, and entropy value of each new defect sample; calculating the confidence level of each new defect sample based on the credibility rate, difference value, and entropy value to obtain a confidence matrix, and performing standard deviation processing on the confidence matrix to obtain a standardized confidence level; screening out the new defect samples corresponding to the standardized confidence level less than the first set threshold for merging with the existing defect sample data to obtain a defect sample training set. Among them, the confidence level is calculated by the following formula:

[0089] V i = α1hd i + α2mt i +(1 - α1 - α2)es i

[0090] V i represents the confidence of the i-th new defect sample, hd i represents the credibility rate, mt i represents the difference, es i represents the entropy value, and α1, α2 are set parameters used to represent the weights of the credibility rate, difference, and entropy value.

[0091] In one embodiment, the update module 1004 uses the self-evolving version of the distribution substation defect recognition model to recognize distribution substation defects, including: inputting the filtered false alarm defect sample data into the previous self-evolving version of the distribution substation defect recognition model, obtaining the output probability result matrix and calculating its probability mean and standard deviation, and calculating the false alarm index according to the current target recognition probability and the calculated probability mean and standard deviation; inputting the false alarm defect samples corresponding to the false alarm index less than the second set threshold into the current self-evolving version of the distribution substation defect recognition model for secondary recognition, and determining whether to give an alarm according to the recognition result. Among them, the false alarm index is calculated by the following formula:

[0092]

[0093] where p represents the current target recognition probability, μ represents the probability mean, and σ represents the standard deviation.

[0094] A distribution substation defect recognition device based on artificial intelligence provided by the present application collects the existing defect sample data of the distribution substation through a sample expansion module, annotates the corresponding defect features for the existing defect sample data, and uses artificial intelligence content generation technology to fuse the defect features on the defect-free pictures of the distribution substation to obtain new defect sample data; the defect recognition module screens the new defect sample data based on a three-dimensional data evaluation strategy, combines the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and uses the defect sample training set to train the current version of the distribution substation defect recognition model to obtain an updated version of the distribution substation defect recognition model to recognize distribution substation defects.

[0095] Further, the obtaining module periodically obtains the false negative / false positive data of the updated version of the distribution substation defect identification model during actual operation, performs defect feature annotation on the false negative / false positive data, and uses artificial intelligence content generation technology to fuse the false negative / false positive defect features on the defect-free pictures of the distribution substation to obtain false negative / false positive defect sample data; the updating module screens the new defect sample data based on the three-dimensional data evaluation strategy, and uses the screened false negative / false positive defect sample data to train the updated version of the distribution substation defect identification model to obtain a self-evolving version of the distribution substation defect identification model for identifying distribution substation defects.

[0096] Thus, by expanding the existing samples, the applicability of the model is improved in the initial stage of training, the false negative / false positive rate is reduced in actual applications, a self-evolving mechanism is formed, and the accuracy and practical level of the model are improved.

[0097] Based on the same concept of the present invention, the specification appendix Figure 11 As shown, the structure of an electronic device 1100 provided by an embodiment of the present application includes: at least one processor 1101, at least one network interface 1104 or other user interfaces 1103, a memory 1105, and at least one communication bus 1102. The communication bus 1102 is used to implement connection communication between these components. The electronic device 1100 optionally includes a user interface 1103, including a display (for example, a touch screen, LCD, CRT, holographic imaging, or projector, etc.), a keyboard, or a pointing device (for example, a mouse, trackball, touchpad, or touch screen, etc.).

[0098] The memory 1105 may include a read-only memory and a random access memory, and provide instructions and data to the processor 1101. A part of the memory 1105 may also include a non-volatile random access memory (NVRAM).

[0099] In some embodiments, the memory 1105 stores the following elements, executable modules, or data structures, or subsets thereof, or extended sets thereof:

[0100] An operating system 11051, including various system programs, for implementing various basic services and processing hardware-based tasks;

[0101] An application program module 11052, including various application programs, such as a launcher, a media player, a browser, etc., for implementing various application services.

[0102] In the embodiments of the present application, by invoking the programs or instructions stored in the memory 1105, the processor 1101 is configured to execute the steps in a method for identifying defects in a power distribution room based on artificial intelligence, and can improve the accuracy and practicality of intelligent patrol identification by augmenting samples.

[0103] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps in the method for identifying defects in a power distribution room based on artificial intelligence.

[0104] Specifically, the storage medium can be a general storage medium, such as a removable disk, a hard disk, etc. When the computer program on the storage medium is run, it can execute the above-mentioned method for identifying defects in a power distribution room based on artificial intelligence.

[0105] In the embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some communication interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical or other forms.

[0106] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0107] In addition, the functional units in the embodiments provided by the present application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0108] If a function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of this application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0109] Finally, it should be noted that the above embodiments are only specific implementation manners of this application, used to illustrate the technical solutions of this application, rather than limiting it. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any technician familiar with this technical field can still modify the technical solutions recorded in the foregoing embodiments, or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A method for identifying defects in a power distribution room based on artificial intelligence, characterized in that: The method comprises the following steps: Collect existing defect sample data of the power distribution room, annotate the existing defect sample data with corresponding defect features, and use artificial intelligence content generation technology to integrate the defect features with the defect-free picture of the power distribution room to obtain new defect sample data; The new defect sample data is screened based on the three-dimensional data evaluation strategy, and the screened new defect sample data is merged with the existing defect sample data to obtain a defect sample training set, and the defect sample training set is used to train the current version of the distribution room defect recognition model to obtain an updated version of the distribution room defect recognition model to identify distribution room defects.

2. According to the method for identifying defects in a power distribution room based on artificial intelligence according to claim 1, it is characterized in that: The method further comprises the following steps: Regularly obtain the missed / false alarm data of the updated version of the distribution room defect recognition model in actual operation, and annotate the missed / false alarm data with defect features, and use artificial intelligence content generation technology to fuse the missed / false alarm defect features with the defect-free pictures of the distribution room to obtain missed / false alarm defect sample data; The new defect sample data is screened based on the three-dimensional data evaluation strategy, and the screened missed / false alarm defect sample data is used to train the updated version of the distribution room defect recognition model to obtain a self-evolving version of the distribution room defect recognition model to identify distribution room defects.

3. According to the method for identifying defects in a power distribution room based on artificial intelligence as described in claim 2, it is characterized in that: The defect features include at least one of flame, water accumulation, damage, and foreign matter; wherein, the defect features are integrated at designated locations on the defect-free picture of the distribution room using artificial intelligence content generation technology to obtain new defect sample data; the designated locations include at least one of the busbar connection of the switch cabinet circuit breaker room, the cable joint of the cable trench, the positive and negative pole connection of the battery, and the transformer area.

4. According to the method for identifying defects in a power distribution room based on artificial intelligence as claimed in claim 3, it is characterized in that: The screening of the new defect sample data based on the three-dimensional data evaluation strategy includes the following steps: Input the new defect sample data into the current version of the distribution room defect recognition model for recognition, and calculate the credibility rate, difference and entropy value of each new defect sample; Calculating the confidence of each new defect sample based on the confidence rate, the difference and the entropy value to obtain a confidence matrix, and performing standard deviation processing on the confidence matrix to obtain a standardized confidence; New defect samples corresponding to the standardized confidence level being less than the first set threshold are screened out and used to be merged with the existing defect sample data to obtain a defect sample training set.

5. According to the method for identifying defects in a power distribution room based on artificial intelligence as claimed in claim 4, it is characterized in that: in, The confidence level is calculated using the following formula: V i =α1hd i +α2mt i +(1-α1-α2)es i V i Indicates the confidence of the i-th new defect sample, hd i Represents the credibility rate, mt i Indicates the difference, es i represents the entropy value, α1 and α2 are setting parameters used to represent the weights of the credibility rate, difference and entropy value.

6. According to the method for identifying defects in a power distribution room based on artificial intelligence according to claim 5, it is characterized in that: in, The self-evolving version of the distribution room defect recognition model is used to identify distribution room defects in the following manner, including the following steps: Input the filtered false alarm defect sample data into the previous self-evolving version of the distribution room defect recognition model, obtain the output probability result matrix and calculate its probability mean and standard deviation, and calculate the false alarm index based on the current target recognition probability and the calculated probability mean and standard deviation; The false alarm defect samples corresponding to the false alarm index being less than the second set threshold are input into the current self-evolving version of the distribution room defect recognition model for secondary recognition, and whether to alarm is determined according to the recognition result.

7. According to claim 6, a method for identifying defects in a power distribution room based on artificial intelligence is characterized in that: The false positive index is calculated using the following formula: Among them, p represents the current target recognition probability, μ represents the mean probability, and σ represents the standard deviation.

8. A power distribution room defect identification device based on artificial intelligence, characterized in that: The device comprises: The sample expansion module is used to collect existing defect sample data of the power distribution room, annotate the existing defect sample data with corresponding defect features, and use artificial intelligence content generation technology to integrate the defect features with the defect-free picture of the power distribution room to obtain new defect sample data; A defect identification module is used to screen the new defect sample data based on a three-dimensional data evaluation strategy, and merge the screened new defect sample data with the existing defect sample data to obtain a defect sample training set, and use the defect sample training set to train the current version of the distribution room defect identification model to obtain an updated version of the distribution room defect identification model to identify distribution room defects.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the method for defect identification in a distribution room based on artificial intelligence are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for identifying defects in a power distribution room based on artificial intelligence as described in any one of claims 1 to 7.

Images