SF6 density relay insulation state diagnosis method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of intelligent monitoring of power equipment, and discloses a density relay insulation state diagnosis method and system based on multi-source data fusion, and the diagnosis method comprises the steps: obtaining an undistorted ROI image, a measurement pressure value, a measurement density value and an effective pointer root point coordinate value based on an original ROI image; and obtaining a calculated density value based on the measured pressure value and the environment temperature, and obtaining a calculated pressure value based on the pointer starting point, the normalized radian value and the full scale angle. When the density difference value between the measured density value and the calculated density value is larger than a density difference value threshold value, a compensation mechanism is triggered to give a fault alarm, and the calculated density value is automatically switched to serve as a fault diagnosis basis. In addition, a diagnosis report in a preset format can be generated based on the calculated density value, the compensation mechanism state, the health state judgment result and the residual life prediction value. The device can adapt to the current market situation that single / double scales coexist, and the insulation diagnosis requirement of equipment in the whole life cycle is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of power equipment, and in particular to a density relay insulation state diagnosis method and system based on multi-source data fusion . BACKGROUND

[0002] With the rapid development of the power industry, power equipment containing gas such as Cubicle Gas-Insulated Switchgear (GIS) is widely used in substations, and its insulation state monitoring is highly dependent on density relay readings (gas density directly determines insulation performance). The current market presents the characteristics of "mainly double scale in stock, and single scale rising in increment": in the stock scenario, about 40% of the in-service substations still use double scale density relays; in the increment scenario, the single scale equipment usage rate of the newly built intelligent substations in 2025 has exceeded 65%.

[0003] The state monitoring of gas insulated equipment (such as GIS and circuit breakers) highly depends on density relays, but the traditional detection method has significant technical bottlenecks: first, temperature drift easily leads to misjudgment, and existing pure visual solutions only read the pressure value of the dial, without combining the nonlinear influence of temperature on density (density needs to be corrected by "pressure x temperature compensation coefficient"), which is difficult to accurately reflect the insulation capacity; second, there is a lack of hidden fault diagnosis, and general instrument recognition patents only output readings, and cannot identify hidden risks such as compensation mechanism failure and slow gas leakage; third, there is a blank in dynamic prediction ability, and there is a lack of quantitative analysis of the density decay trend, which makes it difficult to early warn insulation risks. Although the existing technology has constructed an "multi-model collaborative" instrument reading framework, it is limited by the pure visual solution design, single reading output logic and range-independent idea, which cannot solve the temperature compensation and double scale correlation diagnosis requirements specific to the equipment, and is also difficult to adapt to the market status of "single / double scale coexistence", and cannot meet the insulation diagnosis requirements of the equipment throughout its life cycle.

[0004] Although the existing technology has constructed an "multi-model collaborative" instrument reading framework, it is limited by the pure visual solution design, single reading output logic and range-independent idea, which cannot solve the temperature compensation and double scale correlation diagnosis requirements specific to the equipment, and is also difficult to adapt to the market status of "single / double scale coexistence", and cannot meet the insulation diagnosis requirements of the equipment throughout its life cycle. SUMMARY

[0005] The present application aims to provide a density relay insulation state diagnosis method and system based on multi-source data fusion ​​​​This invention relates to a method and system for diagnosing the insulation status of density relays. It utilizes a temperature compensation engine to convert visually recognized pressure values ​​into standard density values. A dual-scale verification mechanism compares the visual density values ​​with the calculated density values ​​to diagnose faults in the compensation mechanism. Insulation risk prediction is used to estimate the remaining lifespan based on density decay. The algorithm is adaptable to both single-scale and dual-scale density relays. The output is expanded from a single reading to a diagnostic report including density value, health status, and remaining lifespan, significantly improving efficiency. The equipment's ability to meet insulation condition assessment and predictive maintenance requirements.

[0006] Firstly, the present invention provides a method based on multi-source data fusion. The method for diagnosing the insulation condition of density relays includes the following steps: Based on the original ROI image, locate The density relay's dial area is analyzed, and the surface distortion of the dial area is corrected to obtain a distortion-free ROI image. Based on a distortion-free ROI image, dual-scale characters are identified to output measured pressure and measured density values; Based on the distortion-free ROI image, a primary localization method and a two-level backup localization method are used to determine the coordinates of the effective pointer root point. Using the coordinates of the root point of the valid pointer as the origin of space, and combining it with the ambient temperature, a space-temperature normalization binding is achieved through an affine transformation matrix. Calibration The full-scale angle of the density relay is adjusted, and a safe range-guided radian normalization is introduced to constrain the normalized radian value within the effective scale range of the dial. Input the measured pressure value and ambient temperature into the system. The gas law is used to calculate the equivalent standard density value at 20℃, which is the calculated density value; the calculated pressure value is obtained based on the pointer starting point, the normalized radian value, and the full-scale angle. The density difference between the measured density value and the calculated density value is calculated. When the density difference exceeds a threshold, a fault alarm is triggered on the compensation mechanism, and the calculated density value is automatically switched to be the basis for fault diagnosis. Simultaneously, the fault time and density deviation value are recorded in the maintenance database. A sequence of calculated density values ​​over multiple days is constructed, and the density decay slope is fitted using linear regression. Daily average leakage rate; known The density threshold of the density relay is calculated based on the calculated density value at the current moment, the density threshold, and the average daily leakage rate to obtain the remaining life prediction value; the health status judgment result is output according to whether the calculated density value at the current moment is within the normal density range. The left endpoint of the normal density range is the minimum density value, which is a known quantity, and the right endpoint is the density threshold. Based on the calculated density value, the compensation mechanism state, the health state determination result and the residual life prediction value, a diagnostic report in a preset format is generated and uploaded in real time to the substation monitoring system through a specified protocol, and a controllable sound and light alarm is triggered.

[0007] As a possible implementation manner, the positioning The dial area of the density relay specifically includes the following steps: Taking CSPDarknet53 as a backbone network, at least two scale feature maps are extracted from the input original ROI image; Cross-scale feature fusion is realized through a PANet path aggregation network to obtain a cross-scale feature fusion map; The cross-scale feature fusion map is input into a decoupling head to predict the target class, the boundary box coordinates and the confidence respectively; The optimal boundary box is selected through non-maximum suppression to realize the positioning of the dial area of the density relay in a complex background. The dial area of the density relay.

[0008] As a possible implementation manner, the curved surface distortion of the dial area is corrected by the following method: A pre-trained GAN network model is configured, the original ROI image is input into the pre-trained GAN network model, and a distortion-free ROI image is generated, the character distortion rate of the distortion-free ROI image is less than 1%, and the character width error is less than 2 pixels; As a possible implementation manner, based on the distortion-free ROI image, double-scale characters are recognized to output the measured pressure value and the measured density value, specifically including the following steps: The character foreground is extracted through Otsu threshold segmentation, and the noise is eliminated by combining the inflation and corrosion operations; the pressure scale area and the density scale area are segmented according to the character height and width; The segmented character area is input into a CRNN model, the segmented character area is sequentially subjected to character spatial feature extraction, mapping of the character spatial feature to a sequence probability distribution, and alignment-free character decoding based on a CTC loss function to output the measured pressure value and the measured density value.

[0009] As a possible implementation manner, the first-level main positioning method specifically includes: The left scale area, the right scale area and the main scale area are extracted, and at least one effective scale point coordinate is collected for each scale area; A target function is constructed by using a least square circle fitting algorithm, and the target function is solved to obtain the dial center coordinates and the radius; The pointer area tip key point is extracted to obtain a candidate tip point, the Euclidean distance from the candidate tip point to the dial center is calculated, and the candidate tip point with the Euclidean distance less than or equal to an empirical threshold value is selected as an effective pointer root point.

[0010] As a possible implementation manner, the two-stage backup positioning method comprises a Hough straight line detection pointer contour method and an ROI geometric center point positioning method; wherein the Hough straight line detection pointer contour method comprises the following steps: Performing Hough straight line detection in the distortion-free ROI image, screening the longest straight line segment and outputting the two end point values; Calculating the Euclidean distances from the two end points to the center of the dial; Selecting the end point corresponding to the smaller Euclidean distance as the effective pointer root point; The ROI geometric center point positioning method comprises the following steps: Calculating the geometric center of the region where the distortion-free ROI image is located, and outputting the detection frame center point coordinates; Selecting the detection frame center point coordinates as the effective pointer root point.

[0011] As a possible implementation manner, the full-scale angle of the density relay is calibrated by the following method: Before the density relay is put into operation, the rotation angle of the pointer from the minimum scale to the maximum scale is measured by a high-precision angle gauge, and is recorded as the full-scale angle; During the operation of the density relay, the pixel coordinates of the scale disc from the minimum scale line to the maximum scale line are extracted, the angle of the detection scale line is detected by the Hough circle-straight line joint calibration of the circumferential scale, and the included angle is calculated as the full-scale angle.

[0012] As a possible implementation manner, the full-scale angle of the density relay is calibrated by the following method: The full-scale angle of the density relay: taking the effective pointer root point coordinate value as the origin, constructing a polar coordinate system, and calculating the real-time angle of the effective pointer root point by the following formula: wherein, is the real-time angle of the effective pointer root point, , is the ROI region geometric center coordinate value, , is the effective pointer root point coordinate value.

[0013] As a possible implementation manner, the radian normalization is specifically as follows: wherein, is the real-time angle of the effective pointer root point, is the full-scale angle.

[0014] In a second aspect, the present application provides a multi-source data fusion-based ​A density relay insulation state diagnosis system comprises: A GIS scene multi-interference target detection module, based on the original ROI image, locates The dial area of the density relay, and corrects the curved surface distortion of the dial area to obtain a distortion-free ROI image; A dial character semantic analysis module, based on the distortion-free ROI image, identifies double-scale characters to output a measured pressure value and a measured density value; A dial indicating element spatial positioning module, based on the distortion-free ROI image, determines effective pointer root point coordinate values by using a primary positioning and a two-level backup positioning method; A normalized radian calculation module, calibrates The full-scale angle of the density relay, while introducing a safety interval guided radian normalization to constrain the normalized radian value within the effective scale range of the dial; the measured pressure value and the ambient temperature are input to A gas state equation to calculate an equivalent standard density value at 20 DEG C, which is the calculated density value; the calculated pressure value is obtained based on the pointer starting point, the normalized radian value and the full-scale angle; A reading conversion module, calculates the density difference value between the measured density value and the calculated density value, and when the density difference value is greater than a density difference value threshold, triggers a compensation mechanism fault alarm, and automatically switches the calculated density value as the fault diagnosis basis, while recording the fault time and the density deviation value to an operation and maintenance database; a sequence of multi-day calculated density values is constructed, and a density decay slope is fitted through linear regression, which is The daily average leakage rate; the Locking density threshold of the density relay, based on the current time calculated density value, the locking density threshold and the daily average leakage rate to calculate the remaining life prediction value; according to whether the current time calculated density value is located in the normal density interval, output the health state judgment result, wherein the left end point value of the normal density interval is the minimum density value, which is a known quantity, and the right end point value is the locking density threshold; And a diagnosis report generation module, based on the calculated density value, the compensation mechanism state, the health state judgment result, the remaining life prediction value, generates a diagnosis report in a preset format, and uploads it to the substation monitoring system in real time through a specified protocol, and controllable sound and light alarms are triggered.

[0015] Compared with the prior art, the present application has the following advantages: 1. For the problem of high ROI false detection rate caused by strong light / curved surface variation in the prior art, the present application uses a pre-trained GAN network model to correct the curved surface distortion of the dial area, which can improve the detection rate from 85% to 99.2% in a strong light scene.

[0016] 2. The prior art only has two-dimensional positioning, and does not combine temperature with The nonlinear influence of density is introduced into the safety interval-oriented arc normalization, the space and temperature are coupled in three dimensions based on the Clapeyron gas state equation, and the density calculation error is less than 0.5% after temperature compensation, which greatly improves the detection accuracy.

[0017] 3. The output of the prior art only has a numerical value, and the implicit risks such as compensation mechanism failure and slow gas leakage cannot be identified, while the present application generates a preset format of diagnosis report by fusing the calculation density value, compensation mechanism state, health state judgment result and residual life prediction value, and can early warn the insulation risk.

[0018] 4. The prior art adopts a pure visual scheme design and single reading output logic, and it is difficult to adapt to the market status of'single / double scale coexistence', and the present application adopts double scale semantic analysis and power dedicated character set, which greatly improves the analysis accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings, which are included to provide a further understanding of the application and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings: Figure 1 The density relay insulation state diagnosis system structure schematic diagram based on multi-source data fusion in the embodiments of the present application. The density relay insulation state diagnosis system structure schematic diagram based on multi-source data fusion in the embodiments of the present application. DETAILED DESCRIPTION

[0020] In order to clearly describe the technical scheme of the embodiments of the present application, in the embodiments of the present application, the same items or similar items with basically the same function and role are distinguished by using "first", "second" and the like. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and do not limit the order. Those skilled in the art can understand that "first", "second" and the like do not limit the number and execution order, and "first", "second" and the like do not necessarily mean different.

[0021] It should be noted that in the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or advantageous than other embodiments or designs. Rather, the use of "exemplary" or "for example" is intended to present concepts in a concrete manner. The words "first", "second", and the like are used to distinguish different items or similar items.

[0022] In the present application, "at least one" means one or more, and "multiple" means two or more. The "and / or" describes the association relationship of the associated objects, which means that there can be three kinds of relationships, for example, A and / or B, which can represent the following three cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. The following at least one or similar expressions mean any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent: a, b, c, 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 can be single or multiple.

[0023] The embodiment of the present application aims to provide a multi-source data fusion-based Density relay insulation state diagnosis method and system. Single scale reverse compensation and double scale positive verification double track design are adopted to realize full scene coverage of two types of density relays. For the mainstream single scale equipment of newly built intelligent substations, the pressure value is simulated by temperature compensation reverse deduction to realize double scale logic, so as to meet the intelligent operation and maintenance demand of "unattended and automatic diagnosis"; for the double scale equipment with a market share of about 40%, the visual density and calculated density value are directly compared to accurately locate the compensation system hidden fault and maximize the intelligent diagnosis value of the stock assets.

[0024] In the first aspect, the embodiment of the present application provides a multi-source data fusion-based Density relay insulation state diagnosis method, comprising the following steps: Based on the original ROI image, the dial area of the density relay is located, and the curved surface distortion of the dial area is corrected to obtain a distortion-free ROI image; As a possible implementation manner, the dial area of the density relay is located, and the curved surface distortion of the dial area is corrected to obtain a distortion-free ROI image, which specifically comprises the following steps: Taking CSPDarknet53 as the backbone network, at least two scale feature maps are extracted from the input original ROI image; As an example, the original ROI image has a rate range of 480p to 4K, and five scale feature maps are extracted from the input original ROI image by taking CSPDarknet53 as the backbone network, and the sizes are , , , , , , Cross-scale feature fusion is achieved through the Path Aggregation Network (PANet), which enhances the robustness of small target detection.

[0025] The cross-scale feature fusion map is input into the decoupling head to predict the target category, bounding box coordinates, and confidence score separately; the optimal bounding box is selected through non-maximum suppression to achieve prediction in complex backgrounds. Positioning of the dial area of ​​the density relay.

[0026] As an example, a decoupled head is used to predict the target category, bounding box coordinates, and confidence score separately. The optimal bounding box is then selected using a non-maximum suppression (NMS) algorithm. For instance, the confidence threshold is set to 0.5, the intersection-over-union (IoU) threshold is set to 0.45, and the optimal bounding box is selected using (…). , , , This indicates that, under complex backgrounds, Rapid positioning of the dial area of ​​the density relay. In this embodiment, the positioning accuracy is greater than or equal to 99.5%, and the inference delay is less than 50 milliseconds.

[0027] As one possible approach, the surface distortion of the dial area can be corrected using the following method: Configure a pre-trained GAN network model, input the original ROI image into the pre-trained GAN network model, and generate a distortion-free ROI image. The character distortion rate of the distortion-free ROI image is less than 1%, and the character width error is less than 2 pixels.

[0028] As an example, collect 1000+ images with distortion. Images of the density relay dial were used, and the glass curvature and refraction angle were obtained using a laser scanner to generate a "distortion-free reference image" training set. A GAN network model was configured, with the generator G using a U-Net structure and the discriminator D using a PatchGAN structure. The GAN network model was pre-trained using the aforementioned "distortion-free reference image" training set, and optimized using a cycle consistency loss to minimize the distortion of the input image. With the generated distortion-free image G ( The structural similarity (SSIM) of the original ROI image is greater than or equal to 95%. The original ROI image is then input into a pre-trained GAN network model to generate a distortion-free ROI image. The character distortion rate is less than 1%, and the character width error is less than 2 pixels.

[0029] Based on the undistorted ROI image, double-scale characters are recognized to output the measured pressure value and the measured density value. As a possible implementation, based on the undistorted ROI image, double-scale characters are recognized to output the measured pressure value and the measured density value, specifically including the following steps: The character foreground is extracted by Otsu threshold segmentation, and noise is eliminated by combining dilation and erosion operations; the pressure scale area and the density scale area are segmented according to the character height and width; As an example, the character height is about the character width is about the character height, the pressure scale area is located at the outer circle of the dial, the density scale area is located at the inner circle of the dial, and the minimum character area is set to 10x10 pixels. This segmentation method is suitable for dense scale dial.

[0030] The segmented character area is input into the CRNN model, which sequentially processes the segmented character area through character spatial feature extraction, mapping of character spatial features to sequence probability distribution, and non-aligned character decoding based on the CTC loss function, to output the measured pressure value and the measured density value.

[0031] As an example, the CRNN model includes a convolution layer, a recurrent layer, and a transcription layer. The convolution layer is a 7-layer convolution (including BatchNorm and ReLU), which is used to extract character spatial features, and the output feature map size is 1xHxW, where H represents the sequence length and W represents the feature dimension. The recurrent layer is a 2-layer bidirectional long short-term memory network (LSTM), which is used to map the character spatial features to the sequence probability distribution. The transcription layer outputs the measured pressure value and the measured density value based on the CTC (Connectionist Temporal Classification) loss function. The unit of the measured pressure value supports intelligent conversion between MPa and kPa, such as detecting the "kPa" character to automatically convert the MPa value by 1000, and the recognition accuracy is greater than or equal to 99.5%. The measured density value locks the "20℃ equivalent density" identification area, focuses on the dense scale characters through the Transformer attention mechanism, and the scale line positioning error is less than 2 pixels.

[0032] Based on the undistorted ROI image, a primary positioning method and two backup positioning methods are used to determine the effective pointer root point coordinate value. As a possible implementation, the primary positioning method is as follows: Extract the left tick area, right tick area, and main tick area, and collect at least one valid tick point coordinate for each tick area; The objective function is constructed using the least squares circle fitting algorithm, and the coordinates of the center and radius of the dial are obtained by solving the objective function. As an example, key points of three types of objects—left tick area, right tick area, and main tick area—output by the CRNN model are extracted, and the coordinates of at least three valid tick points are collected. The objective function is constructed using the least squares circle fitting algorithm as follows, where the coordinates are derived from the key points of the scale-like object output by the CRNN model: in, This represents the first output of the CRNN model. The coordinates of key points of each tick mark object, corresponding to The coordinates of the center of the dial can be obtained by solving the scale markings on the density relay dial. With radius When the fit is successful, the center positioning error is less than 1 pixel. If there are fewer than 3 scale points, the backup positioning method will be automatically triggered.

[0033] Extract key points at the tip of the pointer area to obtain candidate tip points. Calculate the Euclidean distance from the candidate tip point to the center of the dial. Select candidate tip points whose Euclidean distance is less than or equal to an empirical threshold as valid pointer root points.

[0034] As an example, extract the key points at the tip of the pointer region to obtain candidate tip points. Calculate the distance from the candidate tip to the center of the dial. The Euclidean distance constrains the geometric relationship between the root of the pointer and the dial: For example, the experience threshold is set to 0.8. This empirical threshold is adapted to The typical ratio of the pointer length to the dial radius in a density relay. If... If the candidate tip is determined to be a valid pointer root point, then output... The positioning error is less than 1 pixel. It is the Euclidean distance from the tip of the pointer to the center of the dial, used to determine whether the base of the pointer satisfies the mechanical characteristic of being "close to the center of the dial". This indicates the coordinates of the key point at the tip of the pointer area.

[0035] As one possible implementation, the two-level backup localization method includes the Hough line detection pointer profile method and the ROI geometric center point localization method; among them, the Hough line detection pointer profile method includes the following steps: Perform Hough line detection in the undistorted ROI image, filter the longest straight line segment and output the two endpoint values; Calculate the Euclidean distance of the two endpoints to the center of the dial circle; Select the endpoint corresponding to the smaller Euclidean distance as the effective pointer root point; As an example of a Hough line detection pointer contour method, perform Hough line detection in the undistorted ROI image, filter the longest straight line segment and output the two endpoint values (converted to the original ROI image coordinate system): 、 , respectively, calculate the Euclidean distance of the two endpoints to the center of the dial circle : wherein, represents the Euclidean distance of the endpoint to the center of the dial circle, represents the Euclidean distance of the endpoint to the center of the dial circle, and , both in pixels, are used to determine which end is more consistent with the mechanical characteristic of "the pointer root close to the center of the dial". If , then , otherwise . The effective pointer root point is less than away from the center, with a positioning error of less than 2 pixels.

[0036] The ROI geometric center point positioning method includes the following steps: Calculate the geometric center of the region where the undistorted ROI image is located, and output the detection frame center point coordinates; Select the detection frame center point coordinates as the effective pointer root point.

[0037] As an example of an ROI geometric center point positioning method, calculate the geometric center of the region where the undistorted ROI image is located ( , , , ), and output the detection frame center point coordinates as follows: The positioning error is less than 2 pixels, determined by the positioning accuracy of the detection bounding box, is the geometric center coordinates of the region where the undistorted ROI image is located, serving as a backup reference point for positioning in extreme conditions such as complete pointer occlusion and dial failure.

[0038] Using the coordinates of the root point of the valid pointer as the spatial origin, and combining it with the ambient temperature, a space-temperature normalization binding is achieved through an affine transformation matrix. The normalization binding provides standardized input for all subsequent calculations that depend on spatial location and temperature, ensuring that multi-source data are fused under a unified benchmark.

[0039] As an example, using the effective pointer root point coordinates The origin is defined by pixels, and the data is combined with ambient temperature data collected by a temperature sensor. The unit is ℃. The space-temperature normalization binding is achieved using the affine transformation matrix as follows: Among them, the normalized three-dimensional coordinates , , Dimensionless, representing the temperature normalization coefficient, and the normalized coordinates ( , This is used for subsequent radian normalization to ensure that the pointer angle calculation only reflects the actual mechanical deflection and avoids over-range error. , This is a spatial scaling factor, dimensionless, applied when the dial diameter is 80mm. , All values ​​are set to 0.01, converting pixel coordinates to physical dimensions, achieving unit conversion from pixels to mm, eliminating lens distortion and mounting tilt errors, and adapting to... Industrial design dimensions of density relays. This is the temperature normalization coefficient, dimensionless, with 293K corresponding to an absolute temperature of 20℃, ensuring the physical consistency of the subsequent "temperature compensation engine". , This is the spatial translation coefficient, measured in pixels. It is used to move the origin of the coordinate system to the geometric center of the dial through dial center calibration, thus eliminating installation errors.

[0040] Temperature normalization coefficient 1 / 293 represents the actual temperature Mapping to a 20°C reference temperature scale (293K) ensures that the temperature compensation terms of the subsequent gas law (Clapeyron equation) have physical consistency.

[0041] Calibration The full-scale angle of the density relay is adjusted, and a safe range-guided radian normalization is introduced to constrain the normalized radian value within the effective scale range of the dial. As one possible implementation, it is calibrated using the following method. Full-scale angle of density relay: Before the density relay is put into operation, the rotation angle of the pointer from the minimum scale to the maximum scale is measured by a high-precision angle gauge, and recorded as the full-scale angle. During the operation of the density relay, the pixel coordinates of the scale disc from the minimum scale line to the maximum scale line are extracted, the angle of the scale line is detected through the joint calibration of the circular scale Hough circle-straight line, and the included angle is calculated as the full-scale angle.

[0042] As a kind of calibration The full-scale angle of the density relay is an example. It is found through experiments that the radian value is not strictly in the interval, and the unity of physical meaning and calculation rigor is realized through full-scale angle calibration and radian value constraint. Through full-scale angle calibration combined with the mechanical characteristics of the scale disc of the density relay, a double-mode calibration method of manual annotation + template matching is adopted. The manual annotation method is used to Before the density relay is put into operation, the rotation angle of the pointer from the minimum scale to the maximum scale is measured by a high-precision angle gauge (accuracy ±0.1°), and recorded as the full-scale angle . For example, the full-scale angle of a certain type of density relay is 270°, with an error less than 0.5°. The template matching method is used to During the operation of the density relay, the pixel coordinates of the scale disc from the minimum scale line to the maximum scale line are extracted, the angle of the scale line is detected through the joint calibration of the circular scale Hough circle-straight line (adapted to the circular scale), and the included angle is calculated as the full-scale angle , with an angle error less than 1°.

[0043] As a possible implementation, the full-scale angle of the density relay is calibrated by the following method: taking the coordinate value of the effective pointer root point as the origin, constructing a polar coordinate system, and calculating the real-time angle of the effective pointer root point by the following formula: wherein, is the real-time angle of the effective pointer root point; (( , ) is the geometric center coordinate value of the ROI region; (( , ) is the coordinate value of the effective pointer root point.

[0044] The real-time angle reflects the real-time deflection angle of the pointer root point relative to the geometric center of the dial, and the full-scale angle (examples: fixed constant 270°) is the total angle range of the effective scale of the dial.

[0045] ​If the effective pointer rotates clockwise, the angle takes negative value, and if the effective pointer rotates counterclockwise, the angle takes positive value, and the angle sign is corrected according to the direction of the dial scale, for example, 0 density → rated density counterclockwise increment, which ensures consistent with the growth direction of the scale.

[0046] As a possible implementation, the radian normalization is specifically: wherein, is the real-time angle of the effective pointer root point, is the full-scale angle. represents that the pointer is below the minimum scale, represents that the pointer is within the effective scale range, represents that the pointer is above the maximum scale. The normalized radian value , the radian calculation error is less than 0.5%, which solves the calculation ambiguity caused by the non-strict interval of radian value.

[0047] The measured pressure value and the ambient temperature are input into the Clapeyron gas state equation to calculate the 20℃ equivalent standard density value, i.e., the calculated density value; and the calculated pressure value is obtained based on the pointer starting point, the normalized radian value and the full-scale angle; As an example, the calculated density value is as follows: wherein, is the measured pressure value (unit: kPa), is the ambient temperature (unit: ℃); in the above formula, is equivalent to , so that the calculated density value is always the 20℃ equivalent standard density value, eliminating the temperature drift error. In addition, it is also the basis input for fault diagnosis, health state determination and residual life prediction of the compensation mechanism.

[0048] Dimension verification: × temperature compensation coefficient (dimensionless) → , the physical meaning and unit conversion are consistent; the density calculation error after temperature compensation is less than 0.5%, meeting the accuracy requirement.

[0049] The density difference between the measured density value and the calculated density value is calculated, and when the density difference value is greater than the density difference threshold value, the compensation mechanism fault alarm is triggered, and the calculated density value is automatically switched to the fault diagnosis basis, and the fault time and density deviation value are recorded to the operation and maintenance database; As an example, the density difference threshold value is determined by 500 times of compensation element jamming, transmission connecting rod loosening fault simulation, for example, 5%, compared with the measured density value The density difference between the calculated density value When the density difference , the compensation mechanism fault alarm is triggered, and the is automatically switched as the basis for fault diagnosis, and the fault time and density deviation value are recorded to the operation and maintenance database.

[0050] A sequence of multi-day calculated density values is constructed, and the density decay slope is fitted by linear regression, that is, The daily average leakage rate; the known The remaining life prediction value is calculated based on the calculated density value at the current time, the lock density threshold value, and the daily average leakage rate; As an example, a long short-term memory network is used to construct the density decay prediction, and the standard density value sequence of the last 30 days is extracted The density decay slope is fitted by linear regression as follows: Among them, The daily average leakage rate is kg / m³·day.

[0051] The remaining life prediction value is calculated as follows: Remaining life prediction value= Among them, The lock density threshold value of the density relay, which is usually a known value, for example, the lock density threshold value of a certain type Density relay kg / m³, The calculated density value at the current time. According to whether the calculated density value at the current time is located in the normal density interval, the health state judgment result is output, wherein the left end point value of the normal density interval is the minimum density value, which is a known quantity, and the right end point value is the lock density threshold value.

[0052] As an example, the health state judgment result includes three types of "normal", "warning", and "lock". According to the relationship between the calculated density value at the current time And the normal density interval When , "normal" is output, When , "warning" is output, When

[0053] Based on the calculated density value, the status of the compensation mechanism, the health status assessment results, and the remaining life prediction value, a diagnostic report in a preset format is generated and uploaded to the substation monitoring system in real time through a specified protocol, triggering audible and visual alarms in a controllable manner.

[0054] As an example, integrating the calculation of density values The following is an example of a diagnostic report generated in JSON format, containing information such as the status of the compensation institution, the results of the health status assessment, and the predicted remaining life expectancy: { "density_comp": 6.8, / / Equivalent standard density value at 20℃ (kg / m³) "health_status": "Normal", / / Health status label "compensator_status": "Normal", / / Compensation agency status "remaining_life": 45.2, / / Remaining lifespan (days) "leakage_rate": 0.012, / / Average daily leakage rate (kg / m³·day) "timestamp": "2025-03-19T14:23:18" / / Data collection timestamp } The system uploads data in real time to the substation monitoring system via the Modbus-TCP protocol, with an upload cycle of less than or equal to 1 minute. It can also controllably trigger audible and visual alarms, such as: warning: a yellow warning light that stays on and a buzzer that sounds every 10 seconds; and interlock: a red warning light that stays on and a continuous buzzer.

[0055] This invention is based on the calculation of density values The system outputs monitoring results such as normal, early warning, and locked status based on whether the density is within the normal density range. Temperature normalization is used to ensure the accuracy of the calculated density value. The equivalent value is 20℃, making the judgment results more reliable.

[0056] Secondly, this invention provides a method based on multi-source data fusion. Density relay insulation condition diagnostic system, see Figure 1 ,include: The GIS scene multi-interference target detection module, based on the original ROI image, locates... The density relay's dial area is analyzed, and the surface distortion of the dial area is corrected to obtain a distortion-free ROI image. The dial character semantic parsing module, based on a distortion-free ROI image, recognizes dual-scale characters to output measured pressure and measured density values; The dial indicating element spatial positioning module determines the effective pointer root point coordinate value based on the undistorted ROI image, using a first-level main positioning and two-level backup positioning method; The normalized radian calculation module calibrates The full-scale angle of the density relay, and the radian normalization guided by the safety interval, so as to constrain the normalized radian value in the effective scale range of the dial; the measured pressure value and the ambient temperature are input into The gas state equation, to obtain the 20℃ equivalent standard density value, that is, the calculated density value; based on the pointer starting point, the normalized radian value and the full-scale angle, the calculated pressure value is obtained; The reading conversion module calculates the density difference value between the measured density value and the calculated density value, when the density difference value is greater than the density difference value threshold, the compensation mechanism fault alarm is triggered, and the calculated density value is automatically switched as the fault diagnosis basis, and the fault time and the density deviation value are recorded to the operation and maintenance database; a sequence of multi-day calculated density values is constructed, and the density attenuation slope is fitted through linear regression, that is, The daily average leakage rate; the known The lockout density threshold of the density relay, based on the current time calculated density value, the lockout density threshold and the daily average leakage rate, the remaining life prediction value is calculated; according to whether the current time calculated density value is located in the normal density interval, the health state judgment result is output, wherein the left end point value of the normal density interval is the minimum density value, which is a known quantity, and the right end point value is the lockout density threshold; And the diagnosis report generation module, based on the calculated density value, the compensation mechanism state, the health state judgment result, the remaining life prediction value, generates a diagnosis report in a preset format, and uploads it to the substation monitoring system in real time through the specified protocol, and the controllable sound and light alarm is triggered.

[0057] Although the present application is described herein in conjunction with various embodiments, other variations of the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from an inspection of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit can implement several of the functions recited in the claims. Measures recited in mutually different dependent claims do not preclude their combination in a single claim.

[0058] Although the present application has been described in connection with the preferred embodiments thereof with reference to the specific content thereof, it will be apparent to those skilled in the art that various modifications and changes can be made thereto without departing from the spirit and scope of the application. Accordingly, it is intended that the present application cover all such modifications and changes as fall within the scope of the application, along with all equivalents thereof. It will be understood by those within the art that, in general, terms used herein, and especially to the immediately preceding description and claims attached hereto, are intended to be given their broadest interpretation consistent with the specification and the patent statutes.

Claims

1. A multi-source data fusion based A density relay insulation condition diagnosis method characterized by The method comprises the following steps: Based on the original ROI image, positioning the dial area of the density relay, and correcting the curved surface distortion of the dial area to obtain a distortion-free ROI image; Based on the undistorted ROI image, the double-scale characters are recognized to output the measured pressure value and the measured density value; Based on the undistorted ROI image, a primary main positioning method and a two-level backup positioning method are used to determine the coordinate value of the effective pointer root point; Taking the coordinate of the effective pointer root point as the spatial origin, and combining the ambient temperature, the spatial-temperature normalization binding is realized through the affine transformation matrix; Calibration The full-scale angle of the density relay is introduced with the arc normalization guided by the safety interval, so as to constrain the normalized arc value in the effective scale range of the dial. The measured pressure value and the ambient temperature are input to The gas state equation is used to calculate the equivalent standard density value at 20°C, i.e. the calculated density value; the calculated pressure value is obtained based on the pointer starting point, the normalized radian value and the full-scale angle; A density difference value between the calculated measurement density value and the calculated density value is calculated, and when the density difference value is greater than a density difference value threshold, a compensation mechanism fault alarm is triggered, and the calculated density value is automatically switched to be a fault diagnosis basis, and a fault time and a density deviation value are recorded to an operation and maintenance database; a sequence of multi-day calculated density values is constructed, a density decay slope is fitted through linear regression, and the density decay slope is The daily average leakage rate is known The lockout density threshold of the density relay, the remaining life prediction value is calculated based on the calculated density value at the current time, the lockout density threshold and the daily average leakage rate; and the health state determination result is output according to whether the calculated density value at the current time is located in a normal density interval, wherein the left end point value of the normal density interval is the minimum density value, which is a known quantity, and the right end point value is the lockout density threshold. Based on the calculated density value, the compensation mechanism state, the health state judgment result and the remaining life prediction value, a preset format diagnostic report is generated, which is uploaded to the transformer substation monitoring system in real time through the specified protocol, and a controllable sound and light alarm is triggered.

2. The multi-source data fusion based The density relay insulation state diagnosis method is characterized by comprising the steps of Positioning The dial area of the density relay specifically includes the following steps: Taking CSPDarknet53 as the backbone network, at least two scale feature maps are extracted from the input original ROI image; Cross-scale feature fusion is realized through a PANet path aggregation network to obtain a cross-scale feature fusion map; The cross-scale feature fusion map is input into a decoupling head to predict the target category, the bounding box coordinate and the confidence value respectively; The optimal bounding box is screened by non-maximum suppression to realize the target recognition in complex background Positioning of the dial area of the density relay.

3. The multi-source data fusion based system as claimed in claim 2, wherein The density relay insulation condition diagnosis method is characterized in that, The curved surface distortion of the dial area is corrected by the following method: A pre-trained GAN network model is configured, the original ROI image is input into the pre-trained GAN network model, and an undistorted ROI image is generated, the character distortion rate of the undistorted ROI image is less than 1%, and the character width error is less than 2 pixels.

4. The multi-source data fusion based system as claimed in claim 1, wherein the system is configured to perform the steps of: determining the data source based on the data source priority; and determining the data source priority based on the data source priority table. The density relay insulation condition diagnosis method is characterized by comprising the steps of: Based on the undistorted ROI image, the double-scale characters are recognized to output the measured pressure value and the measured density value, which comprises the following steps: The character foreground is extracted by Otsu threshold segmentation, and the noise is eliminated by combining the inflation and corrosion operation; the pressure scale area and the density scale area are divided according to the character height and width; The segmented character area is input into a CRNN model, the CRNN model sequentially performs character spatial feature extraction, mapping of character spatial features to sequence probability distribution, and non-aligned character decoding based on a CTC loss function, to output the measured pressure value and the measured density value.

5. The multi-source data fusion based method according to claim 1 The method for diagnosing the insulation condition of density relays is characterized by, The primary main positioning method specifically comprises: Extracting the left scale area, the right scale area and the main scale area, and collecting at least one effective scale point coordinate for each scale area; A target function is constructed by using a least square circle fitting algorithm, and the target function is solved to obtain the dial center coordinate and the radius; The pointer region tip key point is extracted to obtain a candidate tip point, the Euclidean distance of the candidate tip point to the dial center is calculated, and the candidate tip point with the Euclidean distance less than or equal to the empirical threshold value is selected as the effective pointer root point.

6. The multi-source data fusion based method of claim 5, The density relay insulation condition diagnosis method is characterized in that, The two-level backup positioning method comprises a Hough line detection pointer contour method and a ROI geometric center positioning method; wherein the Hough line detection pointer contour method comprises the following steps: Hough line detection is performed in the undistorted ROI image, the longest straight line segment is selected and the two end point values are output; The Euclidean distances of the two end points to the dial center are calculated; The end point corresponding to the smaller Euclidean distance is selected as the effective pointer root point; The ROI geometric center positioning method comprises the following steps: The geometric center of the region where the undistorted ROI image is located is calculated, and the detection frame center point coordinate is output; The detection frame center point coordinate is selected as the effective pointer root point.

7. The multi-source data fusion based system of claim 1, wherein the system is further configured to: The density relay insulation condition diagnosis method is characterized in that, Calibration is performed by the following method Full scale angle of the density relay: Before the density relay is put into operation, the rotation angle of the pointer from the minimum scale to the maximum scale is measured by a high-precision angle gauge, and is recorded as the full-scale angle. In the operation of the density relay, the pixel coordinates of the minimum scale line to the maximum scale line of the dial are extracted, the angle of the scale line is detected through the joint calibration of the Hough circle-line of the circumferential scale, and the included angle is calculated as the full-scale angle.

8. The multi-source data fusion based system of claim 1, wherein the system is further configured to: The density relay insulation condition diagnosis method is characterized in that, Calibration by the following method Full-scale angle of density relay: taking the coordinate value of the effective pointer root point as the origin, an polar coordinate system is constructed, and the real-time angle of the effective pointer root point is calculated by the following formula: wherein, is the real-time angle of the effective pointer root point; , is the ROI region geometric center coordinate value; , is the effective pointer root point coordinate value.

9. The multi-source data fusion based system as claimed in claim 1, wherein The density relay insulation condition diagnosis method is characterized in that, The radian normalization specifically comprises: wherein, is the real-time angle of the effective pointer root point, is the full-scale angle.

10. A multi-source data fusion based A density relay insulation condition diagnosis system characterized by, It comprises: The GIS scene multi-interference target detection module is based on an original ROI image, positions a dial area of the density relay, and corrects a curved surface distortion of the dial area to obtain a distortion-free ROI image; The dial character semantic analysis module, based on the undistorted ROI image, identifies the double-scale characters to output the measured pressure value and the measured density value; The dial indicating element spatial positioning module, based on the undistorted ROI image, adopts a primary positioning and two-stage backup positioning method to determine the effective pointer root point coordinate value; Normalization radian calculation module, calibration Full-scale angle of the density relay, while introducing the radian normalization guided by the safety interval, to constrain the normalized radian value in the effective scale range of the dial; input the measured pressure value and the ambient temperature to Gas state equation, calculate the equivalent standard density value at 20℃, which is the calculated density value; based on the pointer starting point, the normalized radian value and the full-scale angle, obtain the calculated pressure value; The readout conversion module calculates a density difference value between the measured density value and the calculated density value, triggers a compensation mechanism fault alarm when the density difference value is greater than a density difference value threshold, and automatically switches the calculated density value as a fault diagnosis basis, while recording the fault time and the density deviation value to an operation and maintenance database; a sequence of multi-day calculated density values is constructed, and a density decay slope is fitted through linear regression, that is The daily average leakage rate is known The lockout density threshold of the density relay, the remaining life prediction value is calculated based on the current time calculated density value, the lockout density threshold and the daily average leakage rate; according to whether the current time calculated density value is located in the normal density interval, the health state judgment result is output, wherein the left end point value of the normal density interval is the minimum density value, which is a known quantity, and the right end point value is the lockout density threshold; And the diagnostic report generation module, based on the calculated density value, the compensation mechanism state, the health state determination result and the remaining life prediction value, generates a diagnostic report in a preset format, which is uploaded to the substation monitoring system in real time through a specified protocol and can controllably trigger an audible and light alarm.

Images