A method for identifying the status of electrical cabinet components based on structural reinforcement

Abstract

This invention discloses a method for electrical cabinet component status recognition based on structural enhancement, relating to the field of equipment status recognition technology. The method includes: acquiring an original image of the target electrical cabinet and inputting it into a trained joint model; extracting stable structural information from the original image using a status recognition model within the joint model to generate a structural enhancement map; encoding and fusing the original image and the structural enhancement map to construct a semantic representation of the component status structure; and generating a component status recognition result. The joint model is trained using an overall loss function that includes global visual semantic alignment loss and structural semantic alignment loss. The global visual semantic alignment loss learns the correspondence between electrical cabinet image samples and business status labels, while the structural semantic alignment loss learns the correspondence between structural enhancement map samples and business status labels, endowing the model with inherent resistance to complex operating conditions, thereby improving the accuracy of electrical cabinet component status recognition.

Description

Technical Field

[0001] This invention relates to the field of equipment status identification technology, and more specifically to a method for identifying the status of electrical cabinet components based on structural reinforcement. Background Technology

[0002] In scenarios such as power distribution rooms, substation protection rooms, and user-side applications in industrial and mining enterprises, electrical cabinets are equipped with components such as pointer-type instruments, digital instruments, indicator lights, and protection pressure plates to reflect the equipment's operating status, protection activation / deactivation status, and on-site operating conditions. Currently, on-site maintenance mainly relies on periodic manual inspections, generating maintenance records or alarm notifications upon discovering anomalies.

[0003] With the advancement of unmanned substations, intelligent distribution rooms, and intelligent inspection systems, existing technologies are beginning to utilize inspection robots, fixed cameras, or mobile terminals to collect images of electrical cabinets. These images are then automatically identified through methods such as target detection, OCR recognition, pointer angle regression, and status classification. This solution, to a certain extent, replaces manual verification, reducing the workload of maintenance personnel.

[0004] However, in scenarios such as large industrial parks, auxiliary systems of hydropower plants, underground utility tunnels, and remote substations, images of electrical cabinets are often affected by factors such as glass reflections, oil and dust accumulation, uneven lighting, shooting angle deviations, electromagnetic interference, and cluttered backgrounds. Ordinary image recognition models rely on apparent features such as color, texture, and brightness; once the on-site environment changes, the accuracy of state recognition will decrease significantly. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that existing electrical cabinet component status identification methods are easily affected by reflection, stains, dust accumulation, uneven lighting and shooting angle deviation under complex working conditions, resulting in a decrease in identification accuracy. The purpose is to provide an electrical cabinet component status identification method based on structural enhancement, which solves the above problems.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a method for identifying the status of electrical cabinet components based on structural reinforcement, comprising:

[0008] Acquire the original image of the target electrical cabinet and input the original image into the trained joint model; the joint model includes a state recognition model;

[0009] Using the state recognition model, stable structural information is extracted from the original image to generate a structure enhancement map. The original image and the structure enhancement map are encoded and fused to construct a semantic representation of the component state structure of the target electrical cabinet. Based on the semantic representation of the component state structure, the component state recognition result of the target electrical cabinet is generated.

[0010] The joint model is trained using a training sample set, which includes electrical cabinet image samples and corresponding service status labels. The overall loss function used by the joint model includes global visual semantic alignment loss and structural semantic alignment loss. The global visual semantic alignment loss is used to measure the deviation between the electrical cabinet image samples and the corresponding service status labels. The structural semantic alignment loss is used to measure the deviation between the structural enhancement map samples extracted from the electrical cabinet image samples and the corresponding service status labels.

[0011] Optionally, the joint model further includes an operation and maintenance semantic generation model; after generating the component status identification results of the target electrical cabinet, the method further includes:

[0012] The semantic representation of the component status structure of the target electrical cabinet, the component status identification results, business rules, and prompt templates are input into the operation and maintenance semantic generation model, and the operation and maintenance prompt text of the target electrical cabinet is output.

[0013] Optionally, the step of inputting the semantic representation of the component status structure and component status identification results of the target electrical cabinet, business rules and prompt templates into the operation and maintenance semantic generation model, and outputting the operation and maintenance prompt text of the target electrical cabinet, includes:

[0014] Based on the deviation between the component status identification results of the target electrical cabinet and the SCADA electrical quantity data, the electrical quantity verification results are obtained;

[0015] Based on the deviation between the component status identification results of the target electrical cabinet and historical statistical data, historical trend analysis results are obtained;

[0016] The component status structure semantic representation and component status identification results of the target electrical cabinet, the electrical quantity verification results, the historical trend analysis results, business rules and prompt templates are input into the operation and maintenance semantic generation model, and the operation and maintenance prompt text of the target electrical cabinet is output.

[0017] Optionally, the training sample set also includes maintenance prompt samples corresponding to the electrical cabinet image samples and local business status labels corresponding to local structural regions in the electrical cabinet image samples; the overall loss function is obtained by weighted summation of joint alignment loss, local structural association loss, state recognition loss, and maintenance semantic generation loss;

[0018] The joint alignment loss is obtained by weighted summation of the global visual semantic alignment loss and the structural semantic alignment loss;

[0019] The local structure association loss is used to measure the deviation between the local structure region and the corresponding local business status label.

[0020] The state recognition loss is used to measure the deviation between the component state recognition result output by the state recognition model and the corresponding business state label;

[0021] The operation and maintenance semantic generation loss is used to measure the deviation between the operation and maintenance prompt text output by the operation and maintenance semantic generation model and the corresponding operation and maintenance prompt sample.

[0022] Optionally, the functional expression for the global visual semantic alignment loss is as follows:

[0023]

[0024] The functional expression for the structural semantic alignment loss is as follows:

[0025]

[0026] in, The global visual semantic alignment loss; The structural semantic alignment loss; Represents the similarity function; Indicates the temperature coefficient; Indicates the first The original visual features of each electrical cabinet image sample Indicates from the first Structural features of structural enhancement maps extracted from image samples of electrical cabinets; Indicates the first Business semantic features corresponding to each electrical cabinet image sample.

[0027] Optionally, after generating the component status identification results of the target electrical cabinet, the method further includes:

[0028] Based on the component status identification results of the target electrical cabinet, a visual reliability score, an electrical quantity consistency score, and a historical trend anomaly score are calculated and weighted to obtain a comprehensive status assessment value for the target electrical cabinet. The visual reliability score measures the reliability of the component status identification results themselves; the electrical quantity consistency score measures the deviation between the component status identification results and SCADA electrical quantity data; and the historical trend anomaly score measures the deviation between the component status identification results and historical statistical data.

[0029] If the comprehensive status evaluation value is greater than or equal to the preset alarm threshold, an abnormal alarm is triggered.

[0030] Optionally, the formula for calculating the visual credibility score is as follows:

[0031]

[0032] in, This represents the confidence level of the component state recognition result output by the state recognition model; This represents the structural consistency score between the structural enhancement diagram of the target electrical cabinet and the component status identification result; The image quality score is the original image of the target electrical cabinet. , , The weights of each item.

[0033] Optionally, the formula for calculating the electrical quantity consistency score is as follows:

[0034]

[0035] in, The instrument readings in the component status identification results; ϵ represents the corresponding measurement point data in the SCADA electrical quantity data, where ϵ represents a minimum constant to prevent the denominator from being zero; max() represents taking the maximum value.

[0036] Optionally, the formula for calculating the historical trend anomaly score is as follows:

[0037]

[0038] in, This indicates the instrument readings in the component status identification results; This represents the mean of historical statistical data. The standard deviation of historical statistical data; This represents a minimal constant to prevent the denominator from being zero.

[0039] Optionally, the stable structural information includes the dial center, pointer axis, scale arc, and pointer tip position of the pointer-type instrument; the character area boundary, digital tube strokes, and decimal point position of the digital instrument; and the pressure plate edge, rotation center, end connection point, and engagement / disengagement angle of the protective pressure plate.

[0040] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0041] This application provides a method for electrical cabinet component status recognition based on structure enhancement. The method includes: acquiring an original image of the target electrical cabinet and inputting it into a trained joint model; extracting stable structural information from the original image using a status recognition model within the joint model to generate a structure enhancement map; encoding and fusing the original image and the structure enhancement map to construct a semantic representation of the component status structure; and generating a component status recognition result. The joint model is trained using an overall loss function that includes global visual semantic alignment loss and structural semantic alignment loss. The global visual semantic alignment loss learns the correspondence between electrical cabinet image samples and business status labels, while the structural semantic alignment loss learns the correspondence between structure enhancement map samples and business status labels. Stable structural information remains unchanged regardless of reflections, oil stains, dust accumulation, uneven lighting, or shooting angle shifts, providing a stable and reliable recognition basis for the model, thereby achieving stable and reliable component status recognition in real industrial environments. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0043] Figure 1 A schematic flowchart of an electrical cabinet component status identification method based on structural reinforcement provided in an embodiment of this application;

[0044] Figure 2 This is a schematic diagram of the operation and maintenance semantic generation process provided in the embodiments of this application;

[0045] Figure 3 This is a schematic diagram of the structural reinforcement multimodal alignment process provided in an embodiment of this application;

[0046] Figure 4 This is a schematic diagram of local structural region association learning provided in an embodiment of this application;

[0047] Figure 5 This is a schematic diagram of electrical quantity linkage verification and abnormal early warning provided in the embodiments of this application;

[0048] Figure 6 Another flowchart illustrating the method for identifying the state of electrical cabinet components based on structural reinforcement provided in this application embodiment;

[0049] Figure 7 A schematic diagram of the structure of the electrical cabinet component status identification system based on structural reinforcement provided in this application embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0051] Please refer to Figure 1 This is a schematic flowchart of a method for identifying the status of electrical cabinet components based on structural reinforcement, provided in an embodiment of this application. The following is a description of... Figure 1 The method for identifying the status of electrical cabinet components based on structural reinforcement is introduced.

[0052] S1. Obtain the original image of the target electrical cabinet and input the original image into the trained joint model.

[0053] In the specific implementation process, raw images of the target electrical cabinet are acquired through inspection robots, fixed cameras, mobile terminals, or edge acquisition devices. The target electrical cabinet refers to any electrical cabinet to be identified, including pointer-type instruments, digital instruments, indicator lights, protective pressure plates, and other internal components. After preprocessing the raw images, such as glare removal, stain removal, adaptive contrast enhancement, distortion correction, noise suppression, and target region cropping, the images are input into a trained joint model. The joint model includes a state recognition model and an operation and maintenance semantic generation model. The state recognition model can generate component state recognition results based on the electrical cabinet images, and the operation and maintenance semantic generation model can generate operation and maintenance prompt text based on the component state recognition results.

[0054] S2. Using a state recognition model, extract stable structural information from the original image to generate a structure enhancement map. Encode and fuse the original image and the structure enhancement map to construct a semantic representation of the component state structure of the target electrical cabinet. Based on the semantic representation of the component state structure, generate the component state recognition result of the target electrical cabinet.

[0055] In the specific implementation process, firstly, the structural enhancement module of the state recognition model extracts stable structural information from the original image to generate a structural enhancement map. Secondly, the visual encoder of the state recognition model encodes the original image to obtain original visual features, and the structural encoder of the state recognition model encodes the structural enhancement map to obtain structural features. Then, the multimodal fusion module of the state recognition model fuses the original visual features and structural features to construct a semantic representation of the component state structure of the target electrical cabinet. Finally, the state recognizer of the state recognition model processes the semantic representation of the component state structure and outputs the component state recognition results of the target electrical cabinet. These component state recognition results include the readings and over-limit states of pointer instruments, the displayed values ​​and over-limit states of digital instruments, the illumination status and color of indicator lights, and the engagement / disengagement status of protective pressure plates, etc.

[0056] In one possible embodiment, stable structural information refers to the structural information of each component of the electrical cabinet that does not change with variations in appearance characteristics such as color, texture, brightness, and reflection. Stable structural information includes the dial center, pointer axis, scale arc, and pointer tip position of pointer-type instruments; the character area boundary, digital tube strokes, and decimal point position of digital instruments; and the edge, rotation center, end connection point, and engagement / disengagement angle of the protective pressure plate.

[0057] In the specific implementation process, the outer contour of the instrument, the edge of the pressure plate, and the boundary of the digital characters are extracted by the edge detection algorithm; the center of the dial and the circular scale area of ​​the pointer instrument are determined by the Hough circle detection; the pointer axis, the scale line direction, and the pressure plate main axis direction are determined by the line detection and skeleton extraction algorithm; the pressure plate end, connection point, and rotation center are extracted by the key point positioning algorithm; and the pointer deflection angle and the pressure plate advance and retreat angle are obtained based on the geometric angle calculation method.

[0058] For pointer-type instruments, the mapping relationship between the dial center, pointer axis, scale area, and angle is used as the structural input; for digital instruments, the character area boundary, digital tube strokes, and decimal point position are used as the structural input; for protective pressure plates, the pressure plate edge, rotation angle, connection position, and deployment / retraction angle are used as the structural input, thereby generating a structural enhancement diagram.

[0059] In one possible embodiment, the joint model further includes an operation and maintenance semantic generation model; after executing S2, the method further includes: inputting the semantic representation of the component state structure of the target electrical cabinet and the component state identification results, business rules and prompt templates into the operation and maintenance semantic generation model, and outputting the operation and maintenance prompt text of the target electrical cabinet.

[0060] In the specific implementation process, firstly, based on the deviation between the component status identification results of the target electrical cabinet and the SCADA electrical quantity data, electrical quantity verification results are obtained. The SCADA electrical quantity data refers to the power operation parameters such as current, voltage, power, and switch status collected in real time by the SCADA system. Secondly, based on the deviation between the component status identification results of the target electrical cabinet and historical statistical data, historical trend analysis results are obtained. Finally, the semantic representation of the component status structure of the target electrical cabinet, along with the component status identification results, electrical quantity verification results, historical trend analysis results, business rules, and prompt templates, are input into the operation and maintenance semantic generation model to output the operation and maintenance prompt text for the target electrical cabinet.

[0061] Please refer to Figure 2This is a schematic diagram of the operation and maintenance semantic generation process provided in this application embodiment. The business rules include instrument over-limit threshold rules, pressure plate activation / deactivation rules, indicator light alarm rules, and operation mode verification rules. The prompt templates include status description templates, risk warning templates, and handling suggestion templates. The operation and maintenance prompt text output by the model includes the current operating status, abnormal risk warnings, trend change descriptions, and suggested handling measures. The operation and maintenance prompt text may also include instrument reading descriptions, pressure plate activation / deactivation status, abnormal status descriptions, and suggested inspection actions.

[0062] In this embodiment, through the operation and maintenance semantic generation model, the system no longer simply outputs numerical values ​​or categories, but can automatically generate textual operation and maintenance prompts that include status descriptions, abnormal risks, trend changes, and handling suggestions, significantly reducing the manual judgment burden on operation and maintenance personnel. By incorporating electrical quantity verification results and historical trend analysis results into the input of the operation and maintenance semantic generation model, the generated operation and maintenance prompts are not only based on visual recognition results but also integrate SCADA real-time data and historical operating patterns, making the prompts more accurate and comprehensive. By introducing business rules and prompt templates, the operation and maintenance semantic generation model can generate standardized and regulated prompts according to power industry standards and operation and maintenance experience, improving the system's professionalism and interpretability.

[0063] The above describes the actual prediction process of the joint model. During the training process, the state recognition model and the operation and maintenance semantic generation model are jointly trained based on the overall loss function to obtain the trained joint model. The training process of this joint model is described below.

[0064] First, a training sample set is obtained, which includes electrical cabinet image samples, corresponding business status labels and maintenance prompt samples, and local business status labels corresponding to local structural regions in the electrical cabinet image samples.

[0065]

[0066] in, Represents a set of business status labels; For the first There are 1 business status label; M is the total number of business status labels.

[0067]

[0068] in, This represents a collection of operation and maintenance prompt texts; For the first The number of maintenance prompt texts; L represents the total number of maintenance prompt texts.

[0069] Each electrical cabinet image sample undergoes preprocessing including glare removal, stain removal, adaptive contrast enhancement, distortion correction, noise suppression, and target region cropping to form the original visual input set.

[0070]

[0071] in, This represents a set of electrical cabinet image samples; For the first There are 10 electrical cabinet image samples; N is the total number of electrical cabinet image samples.

[0072] The structural enhancement module extracts stable structural information from each electrical cabinet image sample to generate a structural enhancement map.

[0073]

[0074] in, This indicates a structural enhancement module; Indicates the parameters of the structural reinforcement module; This represents the set of structure enhancement graphs; the nth structure enhancement graph sample can be represented as:

[0075]

[0076] Each electrical cabinet image sample is encoded using a visual encoder to obtain the original visual features:

[0077]

[0078] in, Represents a visual encoder. The parameters represent the visual encoder. This represents the set of original visual features; the nth original visual feature can be represented as:

[0079]

[0080] The structural enhancement map is encoded using a structural encoder to obtain structural features:

[0081]

[0082] in, Indicates a structure encoder. The parameters represent the structure encoder. Represents the set of structural features; the nth structural feature can be represented as:

[0083]

[0084] The business state labels are encoded by a business semantic encoder to obtain business semantic features;

[0085]

[0086] in, This represents the business semantic encoder. The parameters represent the business semantic encoder. This represents the set of business semantic features; the m-th business semantic feature can be represented as:

[0087]

[0088] The original visual features, structural features, and business semantic features are spatially aligned and fused using a multimodal fusion module to construct a semantic representation of the component state structure.

[0089]

[0090] in, Represents a multimodal fusion function. These represent the parameters of the multimodal fusion module. Indicates the first Semantic representation of component state structure of an electrical cabinet image sample.

[0091] The semantic representation of the component's state structure is processed by a state recognizer, and the component state recognition result is output:

[0092]

[0093] in, Indicates a state recognizer. The parameters represent the state recognizer. This indicates the component state recognition result predicted by the state recognizer.

[0094] semantic representation of component state structure and component status identification results Input the O&M semantic generation model, output O&M prompt text:

[0095]

[0096] in, This represents the semantic generation model for operations and maintenance; The parameters representing the operation and maintenance semantic generation model; This represents the semantic representation of the component's state structure. This indicates the state recognition result; This indicates business rules and prompt templates; This indicates the results of electrical quantity verification and historical trend analysis; This refers to the operation and maintenance prompt text generated by the operation and maintenance semantic generation model.

[0097] The probability of semantic generation in operations and maintenance can be expressed as:

[0098]

[0099] in, Indicates the first One generated word, Indicates the first The text sequence preceding the first r-1 words (i.e., the first r-1 generated words); R is the total length of the generated text; Indicates that in a given , , as well as Under the given conditions, generate operation and maintenance prompt text. The probability of. Indicates that in a given , , , as well as Under the condition of generating the r-th word The probability of.

[0100] In one possible embodiment, the overall loss function is obtained by weighted summation of the joint alignment loss, local structure association loss, state recognition loss, and operational semantic generation loss, and its functional expression is as follows:

[0101]

[0102] in, This represents the overall loss value. For joint alignment loss; This represents the loss due to local structural correlation. State recognition loss; Loss is generated for operational semantics; , , , This represents the weight of each loss. The following sections describe each loss individually.

[0103] (1) Joint alignment loss: obtained by weighted summation of global visual semantic alignment loss and structural semantic alignment loss.

[0104] Joint alignment loss The expression is as follows:

[0105]

[0106] in, This represents the global visual semantic alignment loss; This represents the structural semantic alignment loss; , This indicates the weight of each item.

[0107] Global visual semantic alignment loss The function used to measure the deviation between electrical cabinet image samples and their corresponding business status labels is expressed as follows:

[0108]

[0109] Structural semantic alignment loss The function expression for measuring the deviation between the structure-enhanced map sample extracted from the electrical cabinet image sample and the corresponding business status label is as follows:

[0110]

[0111] in, Represents the similarity function; Indicates the temperature coefficient; Indicates the first The original visual features of each electrical cabinet image sample Indicates from the first Structural features of structural enhancement maps extracted from image samples of electrical cabinets; Indicates the first Business semantic features corresponding to each electrical cabinet image sample.

[0112] To enable the model to learn both image appearance information and device structure information simultaneously, a structure-enhanced multimodal alignment mechanism is established in this embodiment. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the structure-enhanced multimodal alignment process provided in an embodiment of this application. This process is achieved by jointly optimizing two alignment losses: one is through a global visual-semantic alignment loss. The model learns the global correspondence between original visual features and business semantic features, enabling it to understand the mapping between the overall image and the business state; on the other hand, it uses structural semantic alignment loss. By learning the structural correspondence between structural features and business semantic features, the model can understand the mapping between component structural information and business state.

[0113] (2) Local structure association loss: used to measure the deviation between the local structure region and the corresponding local business status label.

[0114] The electrical cabinet image samples are divided into multiple local structural regions, forming a set of local structural regions. :

[0115]

[0116] in, Indicates the first K is the total number of local structural regions.

[0117] Please refer to Figure 4 This is a schematic diagram of local structural region association learning provided in the embodiments of this application. Different local structural region extraction methods are used for different types of components: For pointer instruments, the center of the dial and the scale arc are determined by Hough circle detection, the pointer direction is determined by straight line detection or skeleton extraction, and the angle-reading mapping relationship is established by combining the preset range and the zero and full scale positions to calculate the reading; For digital instruments, the digital display area is extracted by character region segmentation, and the digital content in the digital display area is identified by combining the optical character recognition (OCR) model or the recurrent convolutional neural network (CRNN) model, while the identification of the digital content is verified by the integrity of the digital tube strokes; For indicator lights, the on / off state, color, and flashing state are identified by color space segmentation and light-emitting area detection; For protective pressure plates, the body contour, rotation center, and end connection points of the pressure plate are extracted by edge detection and key point positioning, and the pressure plate deployment / retraction state is determined by the angle difference between the pressure plate main axis direction and the standard deployment / retraction direction.

[0118] Local structural correlation loss The expression is as follows:

[0119]

[0120] in, Indicates the first A local structural region; This represents the corresponding local business status label; K is the total number of local structural regions; This represents classification loss, angular regression loss, or character recognition loss. This represents a local structure identification function; Indicates from local structural region The corresponding local structural information is extracted, and the local state recognition result is generated based on the local structural information.

[0121] (3) State recognition loss: This measures the deviation between the component state recognition result output by the state recognition model and the corresponding business state label. The expression is as follows:

[0122]

[0123] in, Indicates the first Component status recognition results for a sample of electrical cabinet images; Indicates the first The business status label corresponding to each electrical cabinet image sample; N represents the total number of electrical cabinet image samples; This represents classification loss, regression loss, or matching loss.

[0124] (4) Operation and maintenance semantic generation loss This is used to measure the deviation between the operation and maintenance (O&M) prompt text output by the O&M semantic generation model and the corresponding O&M prompt sample. The expression for the O&M semantic generation loss is as follows:

[0125]

[0126] in, Indicates the first One generated word, Indicates the first The text sequence preceding the first r-1 words (i.e., the first r-1 generated words); R is the total length of the generated text; Indicates that in a given , , , as well as Under the condition of generating the r-th word The probability of.

[0127] In one possible embodiment, after performing S2, the method further includes:

[0128] Based on the component status identification results of the target electrical cabinet, the visual credibility score, electrical quantity consistency score, and historical trend anomaly score are calculated and weighted to obtain the comprehensive status evaluation value of the target electrical cabinet; if the comprehensive status evaluation value is greater than or equal to the preset alarm threshold, an anomaly alarm is triggered.

[0129] Among them, the visual credibility score is used to measure the reliability of the component status identification result itself; the electrical quantity consistency score is used to measure the deviation between the component status identification result and the SCADA electrical quantity data; and the historical trend anomaly score is used to measure the deviation between the component status identification result and historical statistical data.

[0130] The formula for calculating the comprehensive condition assessment value is as follows:

[0131]

[0132] in, This is the comprehensive condition assessment value; The visual credibility score is used to measure the reliability of the component state recognition results themselves. Electrical quantity consistency score is used to measure the deviation between component condition identification results and SCADA electrical quantity data; Historical trend anomaly score is used to measure the deviation between the component status identification results and historical statistical data; , , These are the weight parameters for each item.

[0133]

[0134] in, This indicates that an abnormal alarm has been triggered; This indicates that no abnormal alarm will be triggered; This indicates the alarm threshold.

[0135] Visual credibility score The calculation formula is as follows:

[0136]

[0137] in, This represents the confidence level of the component state recognition result output by the state recognition model; The structural consistency score represents the relationship between the structural reinforcement diagram of the target electrical cabinet and the component status identification results. Image quality score for the original image of the target electrical cabinet; , , The weights of each item.

[0138] Confidence level of the state recognition model output It can be represented as:

[0139]

[0140] in, This represents the probability that the nth electrical cabinet image sample is identified as the kth state.

[0141] Structural consistency score It can be represented as:

[0142]

[0143] in, Indicates the first In the first electrical cabinet image sample Each structural benchmark feature This represents the corresponding structural features identified by the state recognition model; Represents the structural deviation function. Indicates the number of structural features involved in the verification.

[0144] Image quality score It can be represented as:

[0145]

[0146] in, This indicates the image sharpness score; This indicates the score for uniformity of illumination; Indicates the noise suppression score; , , This indicates the weight of each item.

[0147] Electrical quantity consistency score The calculation formula is as follows:

[0148]

[0149] in, The instrument readings in the component status identification results; ϵ represents the corresponding measurement point data in the SCADA electrical quantity data, where ϵ represents a minimum constant to prevent the denominator from being zero; max() represents taking the maximum value.

[0150] Historical trend anomaly score The calculation formula is as follows:

[0151]

[0152] in, This indicates the instrument readings in the component status identification results; This represents the mean of historical statistical data. The standard deviation of historical statistical data; This represents a minimal constant to prevent the denominator from being zero.

[0153] Please refer to Figure 5 This diagram illustrates the electrical quantity linkage verification and anomaly warning provided in this application embodiment. By integrating visual confidence assessment, electrical quantity consistency verification, and historical trend deviation analysis for comprehensive status evaluation, it achieves multi-dimensional credibility verification of the identification results. If normal, it records the result normally; if abnormal, it issues an anomaly warning, outputting the alarm level, cause of the anomaly, and maintenance prompts, significantly improving the credibility of the identification results under complex operating conditions.

[0154] Please refer to Figure 6 This is another flowchart illustrating the method for identifying the status of electrical cabinet components based on structural enhancement provided in this application, including processes such as electrical cabinet image acquisition, image preprocessing, structural enhancement map generation, feature encoding, multimodal alignment and fusion, multi-task status identification, operation and maintenance semantic generation, electrical quantity linkage verification, and abnormal warning output.

[0155] In summary, this application provides a structural enhancement-based method for identifying the state of electrical cabinet components. By introducing structural enhancement maps and a multimodal alignment mechanism, stable structural information such as instrument outlines, pointer skeletons, scale distributions, digital character boundaries, pressure plate edges, and insertion / retraction angles are transformed into independent supervisory signals during model training. This allows the model to no longer rely solely on color, texture, and brightness features under complex operating conditions, thereby improving target localization, reading recognition, and state discrimination capabilities in environments such as glass reflection, oil and dust accumulation, uneven lighting, and shooting angle shifts.

[0156] This application utilizes joint modeling of original visual features, structural features, and business semantic features to enable the model to simultaneously learn image appearance information and equipment structural information, and establish a mapping relationship between the two and the power business status. For pointer-type meters, reading recognition is achieved by combining the dial center, scale arc, and pointer tip position; for digital meters, the accuracy of digital recognition is improved by combining character area boundaries and digital tube stroke features; for protective pressure plates, the engagement / disengagement status is determined by combining the pressure plate edge, rotation angle, and connection point position, thereby effectively improving the generalization ability and fine-grained recognition accuracy in mixed recognition scenarios of multiple component types.

[0157] This application further transforms the recognition results into textual prompts for operation and maintenance personnel, generating readable and practical operation and maintenance instructions based on the current operating status, abnormal risks, trend changes, and handling suggestions. By linking and verifying with SCADA electrical quantity data and historical statistical data, the reliability of the visual recognition results can be assessed, and alarm prompts can be automatically triggered for situations such as readings exceeding limits, inconsistent statuses, and abnormal trends.

[0158] Based on the same inventive concept, please refer to Figure 7 This application also provides a structurally reinforced electrical cabinet component status identification system, the system comprising:

[0159] The image acquisition module is used to acquire raw images of the target electrical cabinet and input the raw images into the trained joint model; the joint model includes a state recognition model.

[0160] The state recognition module is used to extract stable structural information from the original image through the state recognition model, generate a structure enhancement map, encode and fuse the original image and the structure enhancement map, construct a semantic representation of the component state structure of the target electrical cabinet, and generate the component state recognition result of the target electrical cabinet based on the semantic representation of the component state structure.

[0161] The joint model is trained using a training sample set, which includes electrical cabinet image samples and their corresponding business status labels. The overall loss function used by the joint model includes global visual semantic alignment loss and structural semantic alignment loss. Global visual semantic alignment loss is used to measure the deviation between electrical cabinet image samples and their corresponding business status labels. Structural semantic alignment loss is used to measure the deviation between structural enhancement map samples extracted from electrical cabinet image samples and their corresponding business status labels.

[0162] Optionally, the joint model also includes an operations and maintenance semantic generation model; the system also includes an operations and maintenance semantic generation module, which is used for:

[0163] After generating the component status identification results of the target electrical cabinet, the semantic representation of the component status structure of the target electrical cabinet, the component status identification results, business rules and prompt templates are input into the operation and maintenance semantic generation model, and the operation and maintenance prompt text of the target electrical cabinet is output.

[0164] Optionally, the operations and maintenance semantic generation module is specifically used for:

[0165] Based on the deviation between the component status identification results of the target electrical cabinet and the SCADA electrical quantity data, the electrical quantity verification results are obtained;

[0166] Based on the deviation between the component status identification results of the target electrical cabinet and historical statistical data, historical trend analysis results are obtained;

[0167] Input the semantic representation of the component status structure of the target electrical cabinet, the component status identification results, the electrical quantity verification results, the historical trend analysis results, the business rules and the prompt template into the operation and maintenance semantic generation model, and output the operation and maintenance prompt text of the target electrical cabinet.

[0168] Optionally, the system also includes an anomaly alarm module, which is used for:

[0169] After generating the component status identification results for the target electrical cabinet, the visual reliability score, electrical quantity consistency score, and historical trend anomaly score are calculated and weighted to obtain the comprehensive status assessment value of the target electrical cabinet based on these results. The visual reliability score measures the reliability of the component status identification results themselves; the electrical quantity consistency score measures the deviation between the component status identification results and SCADA electrical quantity data; and the historical trend anomaly score measures the deviation between the component status identification results and historical statistical data.

[0170] If the overall status assessment value is greater than or equal to the preset alarm threshold, an abnormal alarm will be triggered.

[0171] It should be noted that each module in the structure-enhanced electrical cabinet component status identification system in this embodiment corresponds one-to-one with each step in the structure-enhanced electrical cabinet component status identification method in the aforementioned embodiment. Therefore, the specific implementation of this embodiment can refer to the implementation of the structure-enhanced electrical cabinet component status identification method described above, and will not be repeated here.

[0172] Based on the same inventive concept, this application also provides a computer device, which includes a processor, a memory, and a computer program stored in the memory. The computer program is executed by the processor to implement the aforementioned method for identifying the status of electrical cabinet components based on structural enhancement.

[0173] Based on the same inventive concept, this application also provides a computer storage medium storing a computer program, which is executed by a processor to implement the aforementioned method for identifying the status of electrical cabinet components based on structural enhancement.

[0174] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0175] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0176] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborative files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0177] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0178] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0179] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying the status of electrical cabinet components based on structural reinforcement, characterized in that, include: Obtain the original image of the target electrical cabinet and input the original image into the trained joint model; The joint model includes a state recognition model; Using the state recognition model, stable structural information is extracted from the original image to generate a structure enhancement map. The original image and the structure enhancement map are encoded and fused to construct a semantic representation of the component state structure of the target electrical cabinet. Based on the semantic representation of the component state structure, the component state recognition result of the target electrical cabinet is generated. The joint model is trained using a training sample set, which includes electrical cabinet image samples and corresponding service status labels. The overall loss function used by the joint model includes global visual semantic alignment loss and structural semantic alignment loss. The global visual semantic alignment loss is used to measure the deviation between the electrical cabinet image samples and the corresponding service status labels. The structural semantic alignment loss is used to measure the deviation between the structural enhancement map samples extracted from the electrical cabinet image samples and the corresponding service status labels.

2. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 1, characterized in that, The joint model also includes an operation and maintenance semantic generation model; after generating the component status identification results of the target electrical cabinet, the method further includes: The semantic representation of the component status structure of the target electrical cabinet, the component status identification results, business rules, and prompt templates are input into the operation and maintenance semantic generation model, and the operation and maintenance prompt text of the target electrical cabinet is output.

3. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 2, characterized in that, The step of inputting the semantic representation of the component status structure of the target electrical cabinet and the component status identification results, business rules and prompt templates into the operation and maintenance semantic generation model, and outputting the operation and maintenance prompt text of the target electrical cabinet, includes: Based on the deviation between the component status identification results of the target electrical cabinet and the SCADA electrical quantity data, the electrical quantity verification results are obtained; Based on the deviation between the component status identification results of the target electrical cabinet and historical statistical data, historical trend analysis results are obtained; The component status structure semantic representation and component status identification results of the target electrical cabinet, the electrical quantity verification results, the historical trend analysis results, business rules and prompt templates are input into the operation and maintenance semantic generation model, and the operation and maintenance prompt text of the target electrical cabinet is output.

4. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 2, characterized in that, The training sample set also includes maintenance prompt samples corresponding to the electrical cabinet image samples and local business status labels corresponding to local structural regions in the electrical cabinet image samples; the overall loss function is obtained by weighted summation of joint alignment loss, local structural association loss, state recognition loss, and maintenance semantic generation loss. The joint alignment loss is obtained by weighted summation of the global visual semantic alignment loss and the structural semantic alignment loss; The local structure association loss is used to measure the deviation between the local structure region and the corresponding local business status label. The state recognition loss is used to measure the deviation between the component state recognition result output by the state recognition model and the corresponding business state label; The operation and maintenance semantic generation loss is used to measure the deviation between the operation and maintenance prompt text output by the operation and maintenance semantic generation model and the corresponding operation and maintenance prompt sample.

5. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 1, characterized in that, The functional expression for the global visual semantic alignment loss is as follows: ； The functional expression for the structural semantic alignment loss is as follows: ； in, The global visual semantic alignment loss; The structural semantic alignment loss; Represents the similarity function; Indicates the temperature coefficient; Indicates the first The original visual features of each electrical cabinet image sample Indicates from the first Structural features of structural enhancement maps extracted from image samples of electrical cabinets; Indicates the first Business semantic features corresponding to each electrical cabinet image sample.

6. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 1, characterized in that, After generating the component status identification results of the target electrical cabinet, the method further includes: Based on the component status identification results of the target electrical cabinet, a visual reliability score, an electrical quantity consistency score, and a historical trend anomaly score are calculated and weighted to obtain a comprehensive status assessment value for the target electrical cabinet. The visual reliability score measures the reliability of the component status identification results themselves; the electrical quantity consistency score measures the deviation between the component status identification results and SCADA electrical quantity data; and the historical trend anomaly score measures the deviation between the component status identification results and historical statistical data. If the comprehensive status evaluation value is greater than or equal to the preset alarm threshold, an abnormal alarm is triggered.

7. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 6, characterized in that, The formula for calculating the visual credibility score is as follows: ； in, This represents the confidence level of the component state recognition result output by the state recognition model; This represents the structural consistency score between the structural enhancement diagram of the target electrical cabinet and the component status identification result; The image quality score is the original image of the target electrical cabinet. , , The weights of each item.

8. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 6, characterized in that, The formula for calculating the electrical quantity consistency score is as follows: ； in, The instrument readings in the component status identification results; ϵ represents the corresponding measurement point data in the SCADA electrical quantity data, where ϵ represents a minimum constant to prevent the denominator from being zero; max() represents taking the maximum value.

9. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 6, characterized in that, The formula for calculating the historical trend anomaly score is as follows: ； in, This indicates the instrument readings in the component status identification results; This represents the mean of historical statistical data. The standard deviation of historical statistical data; This represents a minimal constant to prevent the denominator from being zero.

10. The method for identifying the status of electrical cabinet components based on structural reinforcement according to claim 1, characterized in that, The stable structural information includes the dial center, pointer axis, scale arc, and pointer tip position of the pointer-type instrument; the character area boundary, digital tube strokes, and decimal point position of the digital instrument; and the edge, rotation center, end connection point, and engagement / disengagement angle of the protective pressure plate.

Images