Power transmission line anomaly detection and analysis method and device, electronic equipment and medium
By obtaining transmission line images and prompt information, using an abnormality detection and analysis model to adjust the initial model, identify abnormal characteristics and confusing characteristics of the transmission line, the problem of high false alarm rate of overhead transmission lines is solved, and accurate abnormality identification and timely risk processing is achieved.
Patent Information
- Application Number
- CN202510299307.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-13
- Publication Date
- 2025-07-01
AI Technical Summary
In the prior art, the abnormal identification and false alarm rate of overhead transmission lines is high, resulting in untimely risk handling.
By obtaining transmission line images and prompt information, the abnormal detection and analysis model is used for identification. The model adjusts the initial model parameters through transmission line sample images, sample prompt information and sample reply information to identify abnormal features and confusing features, and reduces the misjudgment rate.
It improves the accuracy of abnormal identification of transmission lines, reduces the rate of misjudgment, ensures that risks can be handled in a timely manner, and improves the safety of transmission lines.
Smart Images
Figure CN120236128A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of power technologies, and in particular, to a method, device, electronic device, and medium for detecting and analyzing abnormalities in a transmission line. Background Art
[0002] An overhead transmission line is a transmission line that uses insulators to fix transmission conductors on poles standing upright on the ground to transmit electric energy.
[0003] The external environment of the overhead transmission line corridor is relatively complex, making the overhead transmission line vulnerable to meteorological and environmental influences, resulting in faults. It is necessary to identify abnormalities in the overhead transmission line to timely discover risks and facilitate timely risk handling.
[0004] However, the current identification of abnormalities in overhead transmission lines has a relatively high false alarm rate. Summary of the Invention
[0005] The present disclosure provides a method, device, electronic device, and medium for detecting and analyzing abnormalities in a transmission line to solve the problem of a relatively high false alarm rate in the identification of abnormalities in overhead transmission lines.
[0006] In a first aspect, the present disclosure provides a method for detecting and analyzing abnormalities in a transmission line, the method including:
[0007] Obtaining a transmission line image and a prompt message; the transmission line image is an image of the surrounding environment of the transmission line taken; the prompt message is used to indicate the type of reply required from the transmission line image;
[0008] Inputting the transmission line image and the prompt message into an abnormality detection and analysis model to obtain an abnormality analysis result corresponding to the prompt message; the abnormality analysis result is used to reply to the prompt message;
[0009] Wherein, the abnormality detection and analysis model is obtained by adjusting parameters of an initial abnormality detection and analysis model through transmission line sample images, sample prompt messages, and sample reply messages; the initial abnormality detection and analysis model is established based on a multi-modal large model; each transmission line sample image contains abnormal features and / or confusing features; the abnormal features refer to features that pose risks to the transmission line; the confusing features are features whose similarity to the abnormal features is greater than a preset threshold; the sample prompt message is used to indicate the type of reply required, and the sample reply message refers to the reply corresponding to the prompt message.
[0010] In some embodiments, the abnormal features include at least one of the following: foreign object features of a transmission line, fire features, smoke features, first crane features, and dangerous object features; the confusing features include at least one of the following: transmission line structure features, light features, cloud features, fog features, and second crane features; the sample prompt information may include at least one of the following: the prompt for detecting foreign objects on the transmission line; the prompt for detecting light, cloud, and fire; the prompt for detecting fog and smoke; the prompt for detecting and analyzing crane scenarios, and the prompt for dangerous objects.
[0011] In some embodiments, the step of inputting the transmission line image and the prompt information into the abnormal detection and analysis model to obtain the abnormal analysis result corresponding to the prompt information includes:
[0012] Decompose the prompt information to obtain sub-prompt information in multiple stages;
[0013] Input the transmission line image and the multi-stage sub-prompt information into the abnormal detection and analysis model in stages according to the order of the multi-stage sub-prompt information to obtain the abnormal analysis result corresponding to the prompt information.
[0014] In some embodiments, the initial abnormal detection and analysis model is a student model obtained by knowledge distillation using a first abnormal detection and analysis model as the teacher model.
[0015] In some embodiments, the method further includes:
[0016] When the abnormal analysis result of the transmission line indicates that the transmission line is abnormal, output an abnormal alarm information, and the abnormal alarm information includes the information of the transmission line and the abnormal type information.
[0017] In some embodiments, the method further includes:
[0018] Obtain a transmission line sample image, sample prompt information, and sample reply information;
[0019] Adjust the parameters of the initial abnormal detection and analysis model through the transmission line sample image, sample prompt information, and sample reply information to obtain the abnormal detection and analysis model.
[0020] In some embodiments, the method further includes:
[0021] Use, based on knowledge distillation, a student model obtained by distillation using a first abnormal detection and analysis model as the teacher model as the initial abnormal detection and analysis model.
[0022] In a second aspect, the present disclosure provides a method for adjusting an abnormal detection and analysis model of a transmission line, the method including:
[0023] Obtain sample images of transmission lines, sample prompt information, and sample response information; each sample image of the transmission line contains abnormal features and / or confusing features; the abnormal features refer to the features that pose risks to the transmission line; the confusing features are the features whose similarity to the abnormal features is greater than a preset threshold; the sample prompt information is used to indicate the type of response to be obtained, and the sample response information refers to the response corresponding to the prompt information.
[0024] Adjust the parameters of the initial anomaly detection analysis model through the sample images of the transmission line, sample prompt information, and sample response information to obtain an anomaly detection analysis model; the initial anomaly detection analysis model is established based on a multimodal large model.
[0025] In some embodiments, the method further includes:
[0026] Based on knowledge distillation, use the first anomaly detection analysis model as the teacher model and the student model obtained by distillation as the initial anomaly detection analysis model.
[0027] In a third aspect, the present disclosure provides a transmission line anomaly detection and analysis device, including:
[0028] An acquisition module, configured to acquire a transmission line image and prompt information; the transmission line image is an image of the surrounding environment of the transmission line; the prompt information is used to indicate the type of response to be obtained through the transmission line image.
[0029] A processing module, configured to input the transmission line image and the prompt information into the anomaly detection analysis model to obtain an anomaly analysis result corresponding to the prompt information; the anomaly analysis result is used to reply to the prompt information; wherein, the anomaly detection analysis model is obtained by adjusting the parameters of the initial anomaly detection analysis model through the sample images of the transmission line, sample prompt information, and sample response information; the initial anomaly detection analysis model is established based on a multimodal large model; each sample image of the transmission line contains abnormal features and / or confusing features; the abnormal features refer to the features that pose risks to the transmission line; the confusing features are the features whose similarity to the abnormal features is greater than a preset threshold; the sample prompt information is used to indicate the type of response to be obtained, and the sample response information refers to the response corresponding to the prompt information.
[0030] In a fourth aspect, the present disclosure provides an electronic device, including a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the program, it implements the method described in the first aspect above.
[0031] Fifth aspect, the present disclosure provides an electronic device, including a memory and a processor, where the memory stores a computer program that can run on the processor, and when the processor executes the program, the method described in the second aspect above is implemented.
[0032] Sixth aspect, the present disclosure provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in the first aspect above is implemented.
[0033] Seventh aspect, the present disclosure provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in the second aspect above is implemented.
[0034] For the transmission line anomaly detection and analysis method, device, electronic device and medium provided by the embodiments of the present disclosure, the server inputs the obtained transmission line image and prompt information into the anomaly detection and analysis model to obtain the transmission line anomaly analysis result. Among them, the anomaly detection and analysis model is pre-stored in the server 102, and the anomaly detection and analysis model is obtained by adjusting the parameters of the pre-trained initial anomaly detection and analysis model through transmission line sample images, sample prompt information and sample reply information. Among them, the transmission line sample image contains anomaly features and / or confounding features. Anomaly features refer to features that pose risks to the transmission line, and confounding features are features with a high similarity to the anomaly features; the sample prompt information is used to indicate the type of reply to be obtained, and the sample reply information refers to the reply corresponding to the prompt information. By jointly inputting the transmission line sample image and the sample prompt information into the initial anomaly detection and analysis model established based on the multi-modal large model, the initial anomaly detection and analysis model can quickly learn the difference between the anomaly features in the transmission line and the confounding features similar to the anomaly features, and quickly learn through the sample prompt information, better stimulating the potential of the model itself. The obtained anomaly detection and analysis model can accurately identify the anomaly situation of the transmission line, reduce the anomaly misjudgment rate, ensure that the risks of the transmission line can be processed in time, and improve the safety of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 It is a schematic structural diagram of a transmission line anomaly detection and analysis system provided by an embodiment of the present disclosure;
[0036] Figure 2 It is a schematic flowchart of a transmission line anomaly detection and analysis method provided by an embodiment of the present disclosure;
[0037] Figure 3 It is a schematic flowchart of a method for adjusting a transmission line anomaly detection and analysis model provided by an embodiment of the present disclosure;
[0038] Figure 4Schematic structural diagram of a transmission line anomaly detection and analysis device provided by an embodiment of the present disclosure. Detailed implementation manners
[0039] In the present disclosure, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone, where A and B may be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (item)" or its similar expression refers to any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) of a alone, b alone, or c alone may represent: a alone, b alone, c alone, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b, and c, where a, b, and c may be single or multiple. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0040] The orientation or positional relationship indicated by terms such as "center", "longitudinal", "transverse", "upper", "lower", "left", "right", "front", "rear", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present disclosure and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present disclosure.
[0041] The terms "connected" and "coupled" should be understood in a broad sense. For example, the "connection" or "coupling" of a circuit structure may refer not only to a physical connection, but also to an electrical connection or a signal connection. For example, it may be a direct connection, that is, a physical connection, or may be indirectly connected through at least one intermediate element, as long as the circuit is connected; the signal connection may be a signal connection through a circuit or may also refer to a signal connection through a media medium, such as radio waves. For those of ordinary skill in the art, the specific meanings of the above terms in the present disclosure can be understood according to specific circumstances.
[0042] The external environment of the overhead transmission line corridor is complex, and there are risk situations such as fire, external force damage, and hanging objects, which may cause faults in the overhead transmission line. In the present disclosure, the overhead transmission line is simply referred to as the transmission line.
[0043] The transmission line anomaly detection and analysis method provided by the present disclosure can be applied in a transmission line anomaly detection and analysis system. A transmission line anomaly detection and analysis system is introduced below.
[0044] Please refer toFigure 1 , Figure 1 FIG. Figure 1 is a schematic structural diagram of a transmission line anomaly detection and analysis system provided by an embodiment of the present disclosure. The transmission line anomaly detection and analysis system provided by this embodiment includes: a camera device 101 and a server 102. Among them, the camera device is installed around the transmission line for photographing the transmission line and its surrounding environment. The camera device 101 can be one or more. Figure 1 FIG. Figure 1 exemplarily shows one camera device. It can be understood that this is only an example and does not constitute a limitation to the present disclosure. The camera device 101 can be a camera, which can be a camera fixedly installed on the ground around the transmission line, or a camera carried on a drone. The server 102 is communicatively connected to each camera device 101 respectively. It can be understood that the server 102 can be communicatively connected to each camera device 101 in a wired or wireless manner. The server 102 can be a single server or a server cluster composed of two or more servers.
[0045] The camera device 101 can capture transmission line images in real time or periodically and send the transmission line images to the server 102. The server 102 inputs the obtained transmission line images and prompt information into an anomaly detection and analysis model to obtain a transmission line anomaly analysis result. Among them, the anomaly detection and analysis model is pre-stored in the server 102. The anomaly detection and analysis model is obtained by adjusting the parameters of a pre-trained initial anomaly detection and analysis model through transmission line sample images, sample prompt information, and sample reply information. Among them, the transmission line sample images contain anomaly features and / or confounding features. Anomaly features refer to features that pose risks to the transmission line, and confounding features are features with a high similarity to the anomaly features; the sample prompt information is used to indicate the type of reply to be obtained, and the sample reply information refers to the reply corresponding to the prompt information. By jointly inputting the transmission line sample images and the sample prompt information into the initial anomaly detection and analysis model established based on a multi-modal large model, the initial anomaly detection and analysis model can quickly learn the difference between the anomaly features in the transmission line and the confounding features similar to the anomaly features, and quickly learn through the sample prompt information, better stimulating the potential of the model itself. The obtained anomaly detection and analysis model can accurately identify the anomaly situation of the transmission line, reduce the anomaly misjudgment rate, ensure that the risks of the transmission line can be processed in a timely manner, and improve the safety of the transmission line.
[0046] The following uses specific embodiments to elaborate in detail on the technical solutions provided by the present disclosure.
[0047] Please refer to Figure 2 , Figure 2 FIG. Figure 2 is a schematic flow diagram of a transmission line anomaly detection and analysis method provided by an embodiment of the present disclosure. As shown in Figure 2As shown, the method provided in this embodiment is executed by a server, which may be the server 102 in the above Figure 1 shown embodiment. The method provided in this embodiment may include the following steps 201 and 202.
[0048] Step 201: Obtain the transmission line image and the prompt information.
[0049] Among them, the transmission line image is an image of the surrounding environment of the transmission line. The transmission line image can be taken by a camera device. For example, it can be the Figure 1 shown camera device 101.
[0050] Among them, the prompt information is used to indicate the type of reply required to be obtained through the transmission line image. The prompt information is equivalent to a question asked of the model, waiting for the model to give a corresponding reply result. The reply type can be: yes / no. For example, the prompt information can be: Is there a fire in the current transmission line image, etc.
[0051] Step 202: Input the transmission line image and the prompt information into the anomaly detection and analysis model to obtain the anomaly analysis result of the transmission line corresponding to the prompt information.
[0052] Among them, the anomaly analysis result of the transmission line is used to indicate whether there is an anomaly in the transmission line.
[0053] Furthermore, in the case where the anomaly analysis result of the transmission line indicates that there is an anomaly in the transmission line, the anomaly analysis result of the transmission line may further include the anomaly type information of the transmission line, etc.
[0054] Among them, the anomaly detection and analysis model is pre-stored in the server. It can be obtained by the server through model training, or can be obtained by other electronic devices through model training and sent to the server. The method for obtaining the anomaly detection and analysis model is introduced below.
[0055] An initial anomaly detection and analysis model is established based on a multi-modal large model. Among them, the large model refers to a "large parameter" model trained using large-scale data and powerful computing capabilities. The initial anomaly detection and analysis model can be a pre-trained model. Usually, pre-training is an unsupervised learning using a large-scale unlabeled dataset, and this initial anomaly detection and analysis model belongs to the base model.
[0056] The parameters of the initial anomaly detection and analysis model are adjusted through the transmission line sample image and the sample prompt information to obtain the anomaly detection and analysis model.
[0057] Each of the above transmission line sample images contains abnormal features and / or confounding features. An abnormal feature refers to a feature that poses a risk to the transmission line, and a confounding feature is a feature whose similarity to an abnormal feature is greater than a preset threshold. The preset threshold is a value set in advance. A confounding feature is usually a feature that is relatively similar to a certain abnormal feature. When the model identifies an abnormal feature, it is likely to misidentify the confounding feature as an abnormal feature. For example, light and fire are relatively similar in an image. Light does not affect the transmission line, while fire may pose a risk to the safety of the transmission line. Therefore, in anomaly detection, the fire in the transmission line sample image is defined as an abnormal feature, and the light in the transmission line sample image is defined as a confounding feature. Exemplarily, the abnormal features and confounding features can be manually marked on the transmission line sample image.
[0058] Furthermore, the abnormal features may include, but are not limited to, at least one of the following: foreign object features on the transmission line, fire features, smoke features, and the first crane feature. Among them, the foreign object feature on the transmission line refers to the feature that there are foreign objects on the transmission line and its surrounding area. The fire feature refers to the feature that there is fire on the transmission line and its surrounding area. The smoke feature refers to the feature of the smoke generated when a fire breaks out on the transmission line and its surrounding area. The first crane feature refers to the feature of a crane that may come into contact with the transmission line. If the crane comes into contact with the transmission line, it may cause damage to the transmission line.
[0059] Correspondingly, the confounding features include at least one of the following: transmission line structure features, light features, cloud features, fog features, and the second crane feature. Among them, the transmission line structure feature refers to the features of each structure of the transmission line. The light feature refers to the feature that there is light on the transmission line and its surrounding area. The cloud feature refers to the feature that there are clouds above the transmission line. The fog feature refers to the feature that there is fog around the transmission line. The second crane feature refers to the feature of a crane that will not come into contact with the transmission line.
[0060] It can be understood that the confounding feature corresponding to the foreign object feature on the transmission line is the transmission line structure feature; the confounding feature corresponding to the fire feature is the light feature or cloud feature, etc.; the confounding feature corresponding to the smoke feature is the fog feature, etc.; the confounding feature corresponding to the first crane feature is the second crane feature.
[0061] The above sample prompt information is used to indicate the type of response that needs to be obtained. The sample response information refers to the response corresponding to the prompt information. The sample prompt information can also be used to indicate the difference information between the abnormal feature and the confounding feature, and / or the risk brought by the abnormal feature to the transmission line. The sample prompt information can be in the form of text or voice, etc. It should be noted that there may be multiple types of abnormal features. Correspondingly, the sample prompt information refers to the difference information between any abnormal feature and the confounding feature corresponding to this abnormal feature.
[0062] Furthermore, the sample prompt information may include at least one of the following: transmission line foreign object detection prompt; light, cloud, fog, and fire detection prompt; crane scene detection and analysis prompt; and dangerous object prompt.
[0063] Exemplarily, there is sometimes fog in the environment. Fog refers to tiny water droplets formed by the condensation of water vapor in the air and floating in the air near the ground. Fog is a natural phenomenon and usually does not pose a risk to transmission lines. However, sometimes when a fire occurs, there may be smoke characteristics in the transmission line image. The smoke in this application can also be referred to as fume, which means the smoke generated when a fire breaks out. Smoke and fog are usually easily confused. Considering that fog usually floats in mid-air while smoke usually rises from the ground, the sample prompt information can be to indicate the difference between smoke and fog, including: the location where the feature is located, so as to determine whether it is fog or smoke.
[0064] Exemplarily, fire and smoke usually exist together. Considering that clouds or lights usually do not carry smoke, the sample prompt information can be to indicate the difference between fire and clouds / lights, including: whether there is smoke around the feature. If it is detected that there is smoke around a light feature, it can be determined that the currently detected light feature is a fire feature.
[0065] Exemplarily, if there is a crane in the transmission line image, the sample prompt information can also be the information indicating the topographic features of the reference transmission line and the crane, so that the initial anomaly detection and analysis model can learn how to distinguish whether it is a first crane feature or a second crane feature if a crane appears in the image.
[0066] Furthermore, there can be various ways to adjust the initial anomaly detection and analysis model. The model parameter fine-tuning method can be adopted. For example, it can be the Low-Rank Adaptation (LoRa) method.
[0067] Since the transmission line sample image contains anomaly features that may cause transmission line anomalies, confusing features similar to the anomaly features, and sample prompt information indicating the difference between the anomaly features and the confusing features, inputting them into the initial anomaly detection and analysis model for fine-tuning can obtain the anomaly detection and analysis model.
[0068] In this embodiment, the server inputs the obtained transmission line image and prompt information into the anomaly detection and analysis model to obtain the transmission line anomaly analysis result. Among them, the anomaly detection and analysis model is pre-stored in the server 102. The anomaly detection and analysis model is obtained by adjusting the parameters of the pre-trained initial anomaly detection and analysis model through transmission line sample images, sample prompt information, and sample response information. Among them, the transmission line sample image contains anomaly features and / or confounding features. Anomaly features refer to features that pose risks to the transmission line, and confounding features are features with a high similarity to the anomaly features; the sample prompt information is used to indicate the type of response to be obtained, and the sample response information refers to the response corresponding to the prompt information. By jointly inputting the transmission line sample image and the sample prompt information into the initial anomaly detection and analysis model established based on the multi-modal large model, the initial anomaly detection and analysis model can quickly learn the difference between the anomaly features in the transmission line and the confounding features similar to the anomaly features, and quickly learn through the sample prompt information, better stimulating the potential of the model itself. The obtained anomaly detection and analysis model can accurately identify the abnormal conditions of the transmission line, reduce the anomaly misjudgment rate, ensure that the risks of the transmission line can be processed in a timely manner, and improve the safety of the transmission line.
[0069] In some embodiments, step 202 can be implemented in the following manner:
[0070] Step 2021: Decompose the prompt information to obtain sub-prompt information in multiple stages.
[0071] Furthermore, the prompt information can be decomposed according to the type of prompt information. The type of prompt information is obtained by classifying the prompt information according to its complexity.
[0072] Step 2022: According to the order of the sub-prompt information in multiple stages, input the transmission line image and the sub-prompt information in multiple stages into the anomaly detection and analysis model in stages to obtain the anomaly analysis result corresponding to the prompt information.
[0073] Among them, multiple stages mean processing step by step according to the type of prompt information. Each stage processes one type of prompt information.
[0074] In this embodiment, the anomaly detection and analysis model is usually a multi-modal large model in the billions. The complexity of the prompt information it can handle is limited. The original prompt information can be decomposed and modified into a prompt combination for the multi-step multi-modal large model in the billions. Prompt the multi-modal large model in the billions according to the combined multi-steps for reasoning. That is, according to the type of prompt information, classify the prompt information according to its complexity, and prompt the anomaly detection and analysis model step by step to obtain the output result of each stage.
[0075] Exemplarily, the first-step prompt is a preliminary recognition prompt, and the second-step and above prompts are for further complex recognition and analysis based on the results of the previous-step judgment and analysis. For example, the first-step prompt is to recognize whether there is light of a special color, and the second-step prompt is to analyze whether it is light / fireworks; or the first-step prompt is to recognize whether there is a crane, and the second-step prompt is an analysis prompt for the distance of the crane.
[0076] In this embodiment, since the transmission line scenario is complex and the prompt information can also be relatively complex compared to other fields, in order to obtain better output results through the anomaly detection analysis model, the prompt information can be disassembled to obtain multi-stage sub-prompt information. According to the sequence of the multi-stage sub-prompt information, the transmission line image and the multi-stage sub-prompt information are input into the anomaly detection analysis model in stages respectively, and the anomaly analysis results corresponding to the prompt information are obtained, enabling the anomaly detection analysis model to fully understand the prompt information, thereby accurately identifying the anomalies of the transmission line, reducing the anomaly misjudgment rate, predicting the potential risks existing in the transmission line in a timely manner, ensuring that the risks of the transmission line can be processed in a timely manner, and improving the safety of the transmission line.
[0077] In some embodiments, after step 202, the following steps 203 and 204 may further be included.
[0078] Step 203: Determine whether the anomaly analysis result of the first transmission line indicates that the first transmission line has an anomaly.
[0079] If so, continue to execute step 204; if not, return to execute step 201.
[0080] Step 204: Output an anomaly alarm message.
[0081] Among them, the anomaly alarm message includes the information of the transmission line and the anomaly type information.
[0082] The server may need to process the anomalies of multiple transmission lines simultaneously. Therefore, the anomaly alarm message needs to include the information of the transmission line where the current anomaly occurs. This information of the transmission line can be the identification information of the transmission line or the identification information of the camera device. Usually, different lines of the transmission line have different names or numbers and other identification information. Through the identification information of the transmission line, the location of the transmission line where the anomaly occurs can be determined. Usually, the camera device is fixed around the transmission line, and the server can determine the sender of the transmission line image with an anomaly, that is, the identification information of the camera device that sends the transmission line image. Through the identification information of the camera device, the transmission line where the current anomaly occurs can also be determined.
[0083] The abnormal alarm information may further include abnormal type information, which refers to the type of the currently detected abnormality. For example, the current abnormal type information may be the presence of a fire.
[0084] In this embodiment, when the server determines that the abnormal analysis result of the transmission line indicates that the transmission line is abnormal, it outputs an abnormal alarm information including the information of the transmission line and the abnormal type information, timely reminding the user of the risks existing in the current transmission line, and accurately positioning the location and type of the risk, so that the user can quickly locate the transmission line with risks for risk investigation, ensuring that the risks of the transmission line can be processed in time and improving the safety of the transmission line.
[0085] In some embodiments, the initial abnormal detection and analysis model is a student model obtained by distilling with the first abnormal detection and analysis model as the teacher model based on knowledge distillation.
[0086] Among them, knowledge distillation is a method of model compression. Knowledge distillation usually trains a lightweight small model by using the supervision information of a large model with better performance, in order to achieve better performance and accuracy. The large model can be called the teacher model, and the small model can be called the student model. The supervision information output from the teacher model is called knowledge, and the process by which the student model learns and transfers the supervision information from the teacher model is called distillation.
[0087] In this embodiment, usually the first abnormal detection and analysis model is a multi-modal large model with tens of billions, hundreds of billions or trillions of parameters. In the actual application process, there are situations such as large consumption of processing resources. Therefore, based on knowledge distillation, a student model with tens of billions of parameters obtained by distilling with the first abnormal detection and analysis model as the teacher model can be used as the initial abnormal detection and analysis model. This makes the initial abnormal detection and analysis model a lighter-weight model, and further makes the obtained abnormal detection and analysis model a lighter-weight model, improving the processing efficiency of the abnormal detection and analysis model. At the same time, it improves the efficiency of using the abnormal detection and analysis model for abnormal detection and analysis of transmission lines.
[0088] In some scenarios, the abnormal detection and analysis model is obtained through model training. The following introduces an adjustment method for the abnormal detection and analysis model of transmission lines.
[0089] Please refer to Figure 3 , Figure 3 which is a schematic flowchart of an adjustment method for an abnormal detection and analysis model of transmission lines provided by an embodiment of the present disclosure. As Figure 3 shown, the method provided in this embodiment is executed by an electronic device, and the electronic device may be the aboveFigure 2 The server in the illustrated embodiment may also be other electronic devices different from the above Figure 2 illustrated embodiment. The method provided in this embodiment may include the following steps 301 and 302. If Figure 3 the illustrated embodiment is combined with Figure 2 the illustrated embodiment, then steps 301 and 302 are executed before step 202.
[0090] Step 301: Obtain a transmission line sample image, sample prompt information, and sample response information.
[0091] Among them, the transmission line sample image contains labels of abnormal features and labels of confounding features; the confounding feature is a feature similar to the abnormal feature.
[0092] Among them, the sample prompt information is used to indicate the difference information between the abnormal feature and the confounding feature.
[0093] Step 302: Adjust the parameters of the initial anomaly detection analysis model through the transmission line sample image, sample prompt information, and sample response information to obtain an anomaly detection analysis model.
[0094] Among them, the initial anomaly detection analysis model is established based on a multimodal large model. The initial anomaly detection analysis model is established based on a multimodal large model. The initial anomaly detection analysis model may be a pre-trained model. Generally, pre-training is unsupervised learning using a large-scale unlabeled dataset, and this initial anomaly detection analysis model belongs to a base model.
[0095] Each of the above transmission line sample images contains abnormal features and / or confounding features. An abnormal feature refers to a feature that poses a risk to the transmission line, and a confounding feature is a feature whose similarity to the abnormal feature is greater than a preset threshold. The preset threshold is a value set in advance. The confounding feature is usually a feature that is relatively similar to a certain abnormal feature. When the model identifies the abnormal feature, it is easy to misidentify the confounding feature as the abnormal feature. For example, the light and the fire are relatively similar in the image. The light will not affect the transmission line, while the fire may pose a risk to the safety of the transmission line. Therefore, in anomaly detection, the fire in the transmission line sample image is defined as an abnormal feature, and the light in the transmission line sample image is defined as a confounding feature. Exemplarily, the abnormal feature and the confounding feature can be manually labeled on the transmission line sample image.
[0096] Further, the abnormal features may include, but are not limited to, at least one of the following: foreign object features of the transmission line, fire features, smoke features, and first crane features. Among them, the foreign object features of the transmission line refer to the features of the presence of foreign objects on and around the transmission line. The fire features refer to the features of the presence of fire on and around the transmission line. The smoke features refer to the features of the smoke generated when a fire breaks out on and around the transmission line. The first crane feature refers to the features of a crane that may come into contact with the transmission line. If the crane comes into contact with the transmission line, it may cause damage to the transmission line.
[0097] Correspondingly, the confusing features include at least one of the following: transmission line structure features, light features, cloud features, fog features, and second crane features. Among them, the transmission line structure features refer to the features of the various structures of the transmission line. The light features refer to the features of the presence of light on and around the transmission line. The cloud features refer to the features of the presence of clouds above the transmission line. The fog features refer to the features of the presence of fog around the transmission line. The second crane feature refers to the features of a crane that will not come into contact with the transmission line.
[0098] It can be understood that the confusing feature corresponding to the foreign object feature of the transmission line is the transmission line structure feature; the confusing features corresponding to the fire feature are the light feature or the cloud feature, etc.; the confusing feature corresponding to the smoke feature is the fog feature, etc.; the confusing feature corresponding to the first crane feature is the second crane feature.
[0099] The above sample prompt information is used to indicate the type of response to be obtained. The sample response information refers to the response corresponding to the prompt information. The sample prompt information can also be used to indicate the difference information between the abnormal features and the confusing features, and / or the risks brought by the abnormal features to the transmission line. The sample prompt information can be in the form of text or voice, etc. It should be noted that there may be multiple types of abnormal features. Correspondingly, the sample prompt information refers to the difference information between any abnormal feature and the confusing feature corresponding to the abnormal feature.
[0100] Further, the sample prompt information may include at least one of the following: foreign object detection prompt for the transmission line; detection prompts for light, cloud, fog, and fire; crane scene detection and analysis prompts; and dangerous object prompts.
[0101] Exemplarily, there is sometimes fog in the environment. Fog refers to tiny water droplets formed by the condensation of water vapor in the air and floating in the air close to the ground. Fog is a natural phenomenon and usually does not pose a risk to the transmission line. However, sometimes when a fire occurs, there may be smoke features in the transmission line image. The smoke in this application can also be referred to as fume, which refers to the smoke generated when a fire breaks out. Smoke and fog are usually easily confused. Considering that fog usually floats in mid-air while smoke usually rises from the ground, the sample prompt information can be to indicate the differences between smoke and fog, including: the location where the feature is located, so as to determine whether it is fog or smoke.
[0102] Exemplarily, firelight and smoke usually exist together. Considering that clouds or lights usually do not carry smoke, the sample prompt information can indicate the difference between firelight and clouds / lights, including whether there is smoke around the feature. If smoke is detected around a light feature, it can be determined that the currently detected light feature is a firelight feature.
[0103] Exemplarily, if there is a crane in the image of the transmission line, the sample prompt information can also be the information indicating the topographic features of the reference transmission line and the crane, so that the initial anomaly detection and analysis model can learn how to distinguish whether it is the first crane feature or the second crane feature if a crane appears in the image.
[0104] Furthermore, there can be multiple ways to adjust the initial anomaly detection and analysis model. The model parameter fine-tuning method can be adopted. For example, it can be LoRa.
[0105] In this embodiment, the transmission line sample image contains abnormal features and / or confusing features. Abnormal features refer to the features that pose risks to the transmission line, and confusing features are the features with a high similarity to the abnormal features. The sample prompt information is used to indicate the type of response to be obtained, and the sample response information refers to the response corresponding to the prompt information. By jointly inputting the transmission line sample image and the sample prompt information into the initial anomaly detection and analysis model established based on the multimodal large model, the initial anomaly detection and analysis model can quickly learn the difference between the abnormal features in the transmission line and the confusing features similar to the abnormal features, and quickly learn through the sample prompt information, better stimulating the potential of the model itself. The obtained anomaly detection and analysis model can accurately identify the abnormal conditions of the transmission line, reduce the abnormal misjudgment rate, ensure that the risks of the transmission line can be processed in time, and improve the safety of the transmission line.
[0106] In some embodiments, before step 302 of the method provided in this embodiment, step 300 may further be included.
[0107] Step 300: Based on knowledge distillation, use the first anomaly detection and analysis model as the teacher model, and the student model obtained by distillation as the initial anomaly detection and analysis model.
[0108] In this embodiment, generally, the first anomaly detection and analysis model is a multi-modal large model with tens of billions, hundreds of billions, or trillions of parameters. During actual application, there are situations such as high consumption of processing resources. Therefore, based on knowledge distillation, the first anomaly detection and analysis model can be used as the teacher model, and the distilled student model with tens of billions of parameters can be used as the initial anomaly detection and analysis model. This makes the initial anomaly detection and analysis model a lighter-weight model, and further makes the obtained anomaly detection and analysis model a lighter-weight model, improving the processing efficiency of the anomaly detection and analysis model. At the same time, the efficiency of using the anomaly detection and analysis model for power transmission line anomaly detection and analysis is improved.
[0109] Please refer to Figure 4 , Figure 4 , which is a schematic structural diagram of a power transmission line anomaly detection and analysis device provided by an embodiment of the present disclosure. The device provided in this embodiment includes:
[0110] An acquisition module 401, configured to acquire power transmission line images; the power transmission line images are images of the surrounding environment of the power transmission line taken.
[0111] A processing module 402, configured to input the power transmission line images into an anomaly detection and analysis model to obtain an anomaly analysis result of the power transmission line.
[0112] Among them, the anomaly analysis result of the power transmission line is used to indicate whether there is an anomaly in the power transmission line; the anomaly detection and analysis model is obtained by adjusting the parameters of the initial anomaly detection and analysis model through power transmission line sample images and sample prompt information; the initial anomaly detection and analysis model is established based on a multi-modal large model; the power transmission line sample images contain labels of anomaly features and labels of confounding features; the confounding features are features similar to the anomaly features; the sample prompt information is used to indicate the difference information between the anomaly features and the confounding features, and / or the risks brought by the anomaly features to the power transmission line.
[0113] In some embodiments, the anomaly features include at least one of the following: foreign object features of the power transmission line, fire light features, first crane features, and dangerous object features; the confounding features include at least one of the following: power transmission line structure features, light features, cloud features, fog features, and second crane features.
[0114] In some embodiments, the anomaly detection and analysis model is specifically obtained in the following manner: by using power transmission line sample images and sample prompt information, the parameters of the initial anomaly detection and analysis model are adjusted to obtain a fine-tuned initial anomaly detection and analysis model, and based on the type of sample prompt information, the fine-tuned initial anomaly detection and analysis model is adjusted in multiple stages.
[0115] In some embodiments, the initial anomaly detection analysis model is a student model obtained by knowledge distillation, using the first anomaly detection analysis model as the teacher model for distillation.
[0116] In some embodiments, the apparatus further includes:
[0117] An anomaly alarm module, configured to output an anomaly alarm message when the anomaly analysis result of the transmission line indicates that the transmission line has an anomaly. The anomaly alarm message includes information about the transmission line and information about the anomaly type.
[0118] In some embodiments, the apparatus further includes:
[0119] A training module, configured to obtain transmission line sample images and sample prompt information; and adjust the parameters of the initial anomaly detection analysis model through the transmission line sample images and the sample prompt information to obtain an anomaly detection analysis model.
[0120] The implementation principle and beneficial effects of the apparatus in this embodiment are similar to those of the above embodiments, and will not be elaborated here.
[0121] The present disclosure provides an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, it implements the steps of the method in any of the above embodiments.
[0122] Based on the transmission line anomaly detection and analysis method described in any of the above embodiments, the embodiments of the present disclosure further provide a computer-readable storage medium. For example, a non-transitory computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc. Computer instructions are stored on the storage medium for executing the transmission line anomaly detection and analysis method described in any of the above embodiments, which will not be elaborated here.
[0123] Based on the adjustment method of the transmission line anomaly detection analysis model described in any of the above embodiments, the embodiments of the present disclosure further provide a computer-readable storage medium. For example, a non-transitory computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc. Computer instructions are stored on the storage medium for executing the adjustment method of the transmission line anomaly detection analysis model described in any of the above embodiments, which will not be elaborated here.
[0124] Those of ordinary skill in the art can understand that all or part of the steps to implement the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a disk, an optical disc, etc.
[0125] After considering the specification and practicing the disclosure herein, those skilled in the art will readily conceive of other embodiments of the present disclosure. This application is intended to cover any variations, uses, or adaptations of the present disclosure, which follow the general principles of the present disclosure and include known common general knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.
Claims
1. A method for detecting and analyzing abnormalities in a power transmission line, characterized in that: The method comprises: Acquire a power transmission line image and prompt information; the power transmission line image is an image of the surrounding environment of the power transmission line; the prompt information is used to indicate the type of reply required by the power transmission line image; Inputting the transmission line image and the prompt information into an abnormality detection analysis model to obtain an abnormality analysis result corresponding to the prompt information; the abnormality analysis result is used to reply to the prompt information; Among them, the anomaly detection analysis model is obtained by adjusting the parameters of the initial anomaly detection analysis model through transmission line sample images, sample prompt information and sample reply information; the initial anomaly detection analysis model is established based on a multimodal large model; each transmission line sample image contains abnormal features and / or confusion features; the abnormal features refer to features that bring risks to the transmission line; the confusion features are features whose similarity with the abnormal features is greater than a preset threshold; the sample prompt information is used to indicate the type of reply required, and the sample reply information refers to the reply corresponding to the prompt information.
2. The method according to claim 1, characterized in that: The abnormal feature includes at least one of the following: a foreign object feature of a transmission line, a fire feature, a smoke feature, a first crane feature and a dangerous object feature; the confusing feature includes at least one of the following: a transmission line structure feature, a light feature, a cloud feature, a fog feature and a second crane feature; the sample prompt information may include at least one of the following: a foreign object detection prompt of the transmission line; a light, cloud and fire detection prompt; a fog and smoke detection prompt; a crane scene detection and analysis prompt and a dangerous object prompt.
3. The method according to claim 1, characterized in that: The step of inputting the transmission line image and the prompt information into an abnormality detection analysis model to obtain an abnormality analysis result corresponding to the prompt information includes: Decomposing the prompt information to obtain multi-stage sub-prompt information; According to the sequence of the sub-prompt information of the multiple stages, the power transmission line image and the sub-prompt information of the multiple stages are input into the abnormality detection analysis model in stages to obtain the abnormality analysis results corresponding to the prompt information.
4. The method according to claim 1, characterized in that: The initial anomaly detection analysis model is based on knowledge distillation, using the first anomaly detection analysis model as a teacher model to obtain a student model through distillation.
5. The method according to claim 1, characterized in that The method further comprises: When the abnormality analysis result of the power transmission line indicates that the power transmission line is abnormal, abnormality alarm information is output, wherein the abnormality alarm information includes information of the power transmission line and abnormality type information.
6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Obtaining a sample image of a power transmission line, sample prompt information, and sample reply information; The parameters of the initial anomaly detection analysis model are adjusted through the transmission line sample images, sample prompt information and sample reply information to obtain the anomaly detection analysis model.
7. A method for adjusting a transmission line anomaly detection and analysis model, characterized in that: The method comprises: Acquire a transmission line sample image, sample prompt information, and sample reply information; each transmission line sample image contains abnormal features and / or confusing features; the abnormal features refer to features that bring risks to the transmission line; the confusing features refer to features whose similarity to the abnormal features is greater than a preset threshold; the sample prompt information is used to indicate the type of reply required, and the sample reply information refers to the reply corresponding to the prompt information; The parameters of the initial anomaly detection analysis model are adjusted through the transmission line sample images, sample prompt information and sample reply information to obtain the anomaly detection analysis model; the initial anomaly detection analysis model is established based on a multimodal large model.
8. The method according to claim 7, characterized in that The method further comprises: Based on knowledge distillation, the first anomaly detection analysis model is used as a teacher model, and the student model obtained by distillation is used as the initial anomaly detection analysis model.
9. A transmission line anomaly detection and analysis device, characterized in that: include: An acquisition module, used to acquire transmission line images and prompt information; The transmission line image is a photographed image of the surrounding environment of the transmission line; The prompt information is used to indicate the type of response required through the power transmission line image; A processing module, used for inputting the transmission line image and the prompt information into an abnormality detection analysis model to obtain an abnormality analysis result corresponding to the prompt information; The abnormality analysis result is used to reply to the prompt information; wherein, the abnormality detection analysis model is obtained by adjusting the parameters of the initial abnormality detection analysis model through the transmission line sample image, sample prompt information and sample reply information; the initial abnormality detection analysis model is established based on a multimodal large model; each transmission line sample image contains abnormal features and / or confusion features; the abnormal feature refers to a feature that brings risks to the transmission line; the confusion feature is a feature whose similarity with the abnormal feature is greater than a preset threshold; the sample prompt information is used to indicate the type of reply required, and the sample reply information refers to the reply corresponding to the prompt information.
10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.
11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.