Distributed edge computing system and method based on electric power intelligent station building scene

Abstract

The invention belongs to the technical field of power distribution network information, and particularly relates to a distributed edge computing system and method based on an electric power intelligent station building scene. A distributed edge computing method based on an electric power intelligent station building scene comprises the following steps: S10, deploying an edge computing node in each station building, collecting sensor monitoring data of each electric power device as first data, collecting monitoring image data as second data, and sending the first data and the second data to the edge computing node; and S20, identifying a preset labeling region in the second data by the edge computing node, extracting pixel color information in the labeling region as first information, and comparing the first information extracted from the current labeling region with the first information extracted from other labeling regions. According to the scheme, under the condition of limited operation and maintenance and inspection cost, the problems of inaccurate station building electric leakage monitoring data and low uploading speed are solved, and meanwhile, the operation and maintenance efficiency is also optimized.

Description

Technical Field

[0001] This solution belongs to the technical field of distribution network information management, and specifically relates to a distributed edge computing system and method based on the scenario of a power intelligent substation building. Background Art

[0002] With the acceleration of the urbanization process, the scale of urban construction has been continuously expanding, and the number of substation buildings under the jurisdiction of power supply companies has also shown a rapid growth trend. These substation buildings have the characteristics of a large number, wide distribution, complex operating environment, and relatively high safety risks. The internal equipment of some substation buildings is aging, and the surrounding environment is harsh. During the inspection tasks, the inspection personnel face safety hazards such as electric shock. This makes the traditional method of relying on manpower for operation and maintenance and inspection face many challenges, including high labor costs, difficult inspection, long cycle, and low efficiency, and it is difficult to timely and accurately grasp the operating environment conditions of the substation buildings.

[0003] In response to the above problems, the existing technology has set sensors in the substation buildings to collect the monitoring data of each power equipment and upload the sensor monitoring data to the cloud. The cloud performs statistics and calculations on the monitored data. When the monitored data is abnormal, an abnormal signal is generated and sent to the operation and maintenance terminal. This method reduces the cost of manual inspection and improves the efficiency of operation and maintenance and inspection. However, the electric leakage of power equipment and the electromagnetic signals generated during operation will not only interfere with the normal detection of current by the sensor (affecting the accuracy of detection data), but also affect the speed of the sensor uploading the monitoring data (reducing the uploading speed of the monitoring data). If a special device for combating electromagnetic interference is equipped for each substation building, it will significantly increase the economic cost of operation and maintenance and inspection. Summary of the Invention

[0004] The purpose of this solution is to provide a distributed edge computing system and method based on the scenario of a power intelligent substation building to solve the problems of inaccurate leakage monitoring data and low uploading speed of substation buildings under the condition of limited operation and maintenance and inspection costs.

[0005] To achieve the above purpose, this solution provides a distributed edge computing method based on the scenario of a power intelligent substation building, including the following steps: S10: Deploy edge computing nodes inside each substation building, collect the sensor monitoring data of each power equipment as the first data, collect each monitoring image data as the second data, and send the first data and the second data to the edge computing nodes; S20: The edge computing node identifies the preset marked area in the second data, extracts the pixel color information in the marked area as the first information, compares the first information extracted from the current marked area with the first information extracted from other marked areas, and calculates the pixel color similarity value between the current marked area and other marked areas; S30: The edge computing node performs statistics on the first data, generates an anomaly signal based on the statistical result. After the edge computing node generates the anomaly signal, it obtains several labeled regions corresponding to the first data, obtains the pixel color similarity values corresponding to each labeled region, generates a maintenance instruction based on the pixel color similarity value and the position of the labeled region, and uploads the maintenance instruction to the cloud; S40: The edge computing node obtains the data of other edge computing nodes with the minimum communication delay as assisting nodes, splits the first data and the second data corresponding to the anomaly signal into multiple data packets according to the number of assisting nodes, and sends the data packets to the assisting nodes; after receiving the data packets, the assisting nodes upload the data packets to the cloud.

[0006] And, a distributed edge computing system based on a power intelligent substation scenario using a distributed edge computing method based on a power intelligent substation scenario.

[0007] The principle and technical effect of this solution are as follows: First, this solution deploys edge computing nodes inside the power intelligent substation to achieve the proximity collection of power equipment sensor data (the first data) and monitoring image data (the second data), and complete the local processing of the data. This mechanism effectively avoids the delay problem caused by long-distance data transmission, significantly improves the system response speed, and ensures the timeliness of data processing.

[0008] Secondly, when the key equipment in the substation generates induced electromotive force, compared with traditional sensors, the monitoring image acquisition is less affected by the electromagnetic signal generated by the induced electromotive force, and the data accuracy is higher. Although the outer shell of the substation is equipped with a grounding wire to avoid potential safety hazards caused by electric leakage, the grounding wire is easily disconnected due to environmental influences. From the appearance, the grounding wire is still normally connected to the outer shell, but in fact, it has been disconnected from the outer shell, resulting in the problems of the grounding wire being easily undetected. This solution accurately identifies the preset labeled regions (such as key parts like the outer shell of the substation) in the monitoring image, extracts the pixel color information, and calculates the pixel color similarity values between the labeled regions to quantify the image differences. The reason is that when the substation leaks electricity, the outer shells of some equipment will carry charges, forming different forms of capacitors, causing the outer shells of some equipment to adsorb dust in the environment. By quantitatively analyzing the image differences at different positions of the equipment outer shell, this solution can accurately lock the labeled regions with more adsorbed dust. At the same time, by means of multi-region comparison and analysis, it effectively excludes the interference of environmental factors such as light, thereby accurately determining the abnormal charged area of the substation. This technology not only greatly reduces the interference of electromagnetic signals on data acquisition, but also significantly improves the accuracy of leakage monitoring through image recognition technology, greatly shortens the time for manual troubleshooting of the leakage location, and significantly optimizes the operation and maintenance efficiency.

[0009] Furthermore, this solution makes full use of the redundant communication capabilities of other edge nodes (assistant nodes) to split data packets and upload them in parallel. This method effectively avoids the high latency problem caused by excessive data volume in a single node. With the collaborative effect of the assistant nodes, the data upload speed is significantly improved, the transmission loss rate is greatly reduced, and the reliability of data transmission is strongly guaranteed. In addition, this collaborative transmission mode reduces the hardware performance requirements for a single node, effectively saving hardware procurement and operation and maintenance costs.

[0010] In summary, under the condition of limited operation and maintenance and inspection costs, this solution solves the problems of inaccurate leakage monitoring data and low upload speed in the substation building, and at the same time, optimizes the operation and maintenance efficiency.

[0011] Furthermore, when the edge computing node extracts the first information, it performs color space conversion on the pixels in the marked area, converts the RGB values to the HSV space, takes the maximum value among the three components of red, green, and blue as the brightness value, calculates the saturation with the ratio of the difference between the brightness and the minimum component to the brightness, and determines the hue according to the component corresponding to the maximum value among the three RGB components. The specific formula is shown as formula (1) below: (1).

[0012] By converting the RGB color space to the HSV color space, not only can the color information in the image be processed and analyzed more intuitively and effectively, but also by calculating the maximum value among the three components of red, green, and blue as the brightness value, the brightness information of the image can be represented more accurately, which is of great significance for subsequent image analysis and processing. In addition, using the ratio of the difference between the brightness and the minimum component to the brightness to calculate the saturation can more accurately reflect the purity or intensity of the color, and determining the hue according to the component corresponding to the maximum value among the three RGB components can more accurately represent the color types in the image. The precise calculation of these color features not only optimizes the color space conversion and color feature extraction process of the edge computing node, reduces the data transmission volume, and reduces the dependence on cloud computing resources, thus improving the efficiency and real-time performance of data processing.

[0013] Furthermore, when calculating the pixel color similarity value between the current marked area and other marked areas, for the current marked area , compare it one by one with each other marked area to calculate the color feature distance between the two areas , The calculation formula of is shown as formula (3) below: Among them, , , and , , are the mean values of hue, saturation, and value in the HSV color space for regions and respectively; , , are the weight coefficients for each color component, and satisfy ; According to the pixel coordinates of region and the current region , calculate the spatial distance , and set the weight according to , . The calculation formula of is shown in formula (4) below: where is the distance attenuation coefficient; By comprehensively considering the similarity and weight between all other labeled regions and the current region, calculate the overall similarity value of the current region , . The calculation formula of is shown in formula (5) below:

[0014] Through the per-region color similarity calculation and spatial distance weighting mechanism, not only the accuracy and anti-interference ability of power equipment anomaly detection are significantly improved, but also because of the use of HSV hue distance calculation and neighboring region weight assignment, it can effectively distinguish real equipment failures (such as color changes caused by leakage) from environmental interferences (such as light fluctuations), thus greatly reducing the false alarm rate; At the same time, since the weight coefficients and distance attenuation coefficients can be adjusted according to different substation scenarios, this solution has dynamic adaptability and can be compatible with diverse equipment layouts and monitoring conditions; In addition, since only the abnormal candidate regions are analyzed pixel by pixel, the computational redundancy of global image processing is avoided, so it particularly meets the resource limitation requirements of edge nodes. Moreover, this solution can also construct a baseline for equipment color changes by long-term recording of region similarity values, and then assist in predicting potential failures; Furthermore, through cross-substation similarity value comparison, systematic failures (such as the collective deterioration of equipment in the same batch) can ultimately be identified.

[0015] Further, in step S30, the edge computing node also extracts the motion trajectory of the target device in the second data. After generating the abnormal signal, the edge computing node combines the motion trajectory with the pixel color similarity value to determine whether the marked area is abnormally charged, and then generates a maintenance instruction according to the judgment result and the position of the marked area; the target device includes a suspension rod and a number of feathers, the suspension rod is fixedly connected to the substation building, and multiple feathers are successively suspended on the suspension rod; when the edge computing node identifies the motion trajectory of the target device, it first calculates the swing arc of the feather according to the relative position between the feather and the target area in the second data, and then calculates the swing trajectory of the feather according to the swing arc in combination with the acquisition time of the second data, and takes the swing trajectory of the feather as the motion trajectory of the target device.

[0016] Further, when the edge computing node identifies the motion trajectory of the target device, it includes the following steps: S31: For suspended feathers, a collaborative detection model is established through the following formula (6): (6), where represents the swing angle of the th feather at moment, is the allowable angle deviation threshold; S32: Reconstruct the three-dimensional motion trajectory based on the three-dimensional coordinates of the feathers by binocular vision, and the reconstruction formula is shown in the following formula (7): (7), where is the baseline distance, is the focal length, is the parallax; S33: Calculate the dynamic characteristic parameters of the swing trajectory, and the calculation formula of the characteristic parameters is shown in the following formula (8): (8); S34: Use a support vector machine classifier for anomaly determination, and the determination formula is shown in the following formula (9): (9), where the feature vector ; S35: Establish a spatio-temporal correlation matrix between feathers, and the matrix is shown in the following formula (10): (10), When the rank of the matrix triggers an area anomaly alarm.

[0017] ​Through the multimodal fusion of mechanical signals and visual signals, this solution not only significantly improves the accuracy and reliability of live anomaly detection, but also, due to the introduction of high-precision three-dimensional motion modeling and dynamic feature analysis, can achieve sub-millimeter-level micro motion detection and millisecond-level transient response. At the same time, through multi-feather collaborative detection and probability fusion mechanism, it not only effectively suppresses environmental interference and single-point false triggering, but also further enhances the power frequency interference resistance and fault location accuracy by combining frequency domain energy entropy analysis and spatio-temporal correlation matrix. In addition, since this solution supports long-term trajectory feature recording and self-learning optimization, it can not only meet the real-time monitoring requirements, but also provide data support for predictive maintenance. Furthermore, thanks to the standardized parameter design and flexible expansion architecture, this solution can ultimately adapt to diverse power equipment monitoring scenarios while reducing operation and maintenance costs.

[0018] Furthermore, when an anomaly alarm is triggered in the detection area, the edge computing node combines the feather swing trajectory with the pixel color similarity value to establish a joint decision function, as shown in the following formula (11): (11), where, is the Sigmoid activation function; is the normalized swing amplitude feature; is the pixel color similarity value; is the estimated electric field intensity vector, and is the trainable weight; Calculate the priority weight according to the position of the marked area, and the calculation formula is as shown in the following formula (12): (12), where, is the center coordinate of the th marked area, is the center of the feather swing; Define the anomaly level function according to the joint decision function, and the definition formula is as shown in the following formula (13): L = (13), where the threshold is obtained through training of historical data; Finally, generate the maintenance instruction according to the maintenance instruction parameters, and the calculation formula of the maintenance instruction parameters is as shown in the following formula (14): (14), where is the required response time threshold.

[0019] Through the multi-modal feature fusion and spatial position weighting mechanism, this solution not only significantly improves the accuracy and reliability of live anomaly detection, but also, due to the introduction of anomaly level division and dynamic weight adjustment, can achieve accurate fault grading and priority determination. At the same time, by combining the feather swing feature and visual similarity analysis, it not only effectively enhances the anti-interference ability in complex environments, but also, because of the adoption of an intelligent maintenance instruction generation strategy, greatly improves the operation and maintenance response efficiency. In addition, since this solution supports online parameter optimization and real-time response, it can not only adapt to changing on-site working conditions, but also provide decision-making support for resource scheduling. Furthermore, thanks to the modular instruction architecture and scalable fusion algorithm, this solution can ultimately be flexibly applied to the intelligent operation and maintenance scenarios of various power equipment while ensuring the detection accuracy.

[0020] Further, in step S40, when the edge computing node splits the abnormal data into multiple data packets according to the number of assisting nodes, the following steps are included: S41: Establish a node selection scoring function as shown in formula (15) below: (15), where, is the round-trip delay to node , is the available bandwidth, is the CPU load rate, is the configurable weight; S42: Calculate the optimal number of shards through formula (16) below: (16), where, is the total amount of abnormal data, is the network maximum transmission unit, is the benchmark bandwidth requirement; S43: Allocate the data packet size based on the node score, and the allocation formula is as shown in formula (17) below: (17); S44: Use fountain code for encoding, and the encoding formula is as shown in formula (18) below: (18), where, is the original data shard, is the randomly generated matrix element, .

[0021] Through the dynamic scoring model and the adaptive sharding algorithm, this solution not only significantly improves the efficiency and reliability of abnormal data transmission, but also, due to the adoption of the real-time load awareness mechanism, can intelligently optimize network resource allocation. At the same time, through fountain code encoding and dynamic sharding strategies, it not only ensures the transmission success rate in a high-concurrency environment, but also enhances the scalability of this solution because it supports elastic scaling of nodes. In addition, due to the introduction of periodic load balancing adjustment, it can not only adapt to the dynamic changes of the network state, but also effectively reduce the task processing delay. Furthermore, thanks to the flexible configuration of the scoring model and the adaptive characteristics of the sharding algorithm, this solution can ultimately be widely applied to the collaborative processing requirements of various edge computing scenarios while ensuring the transmission quality. Description of the Drawings

[0022] Figure 1 It is a flowchart of a distributed edge computing method based on the power intelligent substation scenario in an embodiment of the present invention. Detailed Embodiments

[0023] The following will clearly and completely describe the concept of the present invention and the technical effects produced in combination with the embodiments to fully understand the purpose, features, and effects of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative efforts shall fall within the scope of protection of the present invention: As Figure 1 shown, a distributed edge computing method based on the power intelligent substation scenario includes the following steps: S10: Deploy edge computing nodes inside each substation, collect the sensor monitoring data of each power device as the first data, collect each monitoring image data as the second data, and send the first data and the second data to the edge computing nodes; S20: The edge computing node identifies the preset annotation area in the second data, extracts the pixel color information in the annotation area as the first information, compares the first information extracted from the current annotation area with the first information extracted from other annotation areas, and calculates the pixel color similarity value between the current annotation area and other annotation areas; S30: The edge computing node statistically analyzes the first data, generates an abnormal signal according to the statistical result. After the edge computing node generates the abnormal signal, it obtains several annotation areas corresponding to the first data, obtains the pixel color similarity values corresponding to each annotation area, generates a maintenance instruction according to the pixel color similarity value of the annotation area and the position of the annotation area, and uploads the maintenance instruction to the cloud; S40: The edge computing node obtains the data of other edge computing nodes with the minimum communication delay as assisting nodes, splits the first data and the second data corresponding to the abnormal signal into multiple data packets according to the number of assisting nodes, and sends the data packets to the assisting nodes; after receiving the data packets, the assisting nodes upload the data packets to the cloud.

[0024] Among them, in step S20, when the edge computing node extracts the first information, it performs color space conversion on the pixels in the marked area, converts the RGB values to the HSV space, takes the maximum value among the red, green, and blue components as the lightness value, and calculates the saturation with the ratio of the difference between the lightness and the minimum component to the lightness. The calculation formula of the saturation is shown in formula (2) below; the hue is determined according to the component corresponding to the maximum value among the three RGB components, as shown in formula (1) below: (1), (2).

[0025] Among them, R, G, and B are the red, green, and blue component values of the pixel point (range 0 - 255), V takes max(R, G, B), that is, the lightness value (brightness). When V = R (dominated by red), calculate the hue with (G - B); when V = G (dominated by green), calculate and add 120° with (B - R) (2 in the formula corresponds to 120°); when V = B (dominated by blue), calculate and add 240° with (R - G) (4 in the formula corresponds to 240°). V - min(R, G, B) represents the color purity to avoid division by zero (when V = 0, the saturation is 0).

[0026] Among them, when calculating the pixel color similarity value between the current marked area and other marked areas, for the current marked area , compare it with each other marked area one by one, and calculate the color feature distance , The calculation formula of is shown in formula (3) below: (3), Among them, , , and , , are the means of the hue, saturation, and lightness of the regions and in the HSV color space respectively; , , are the weight coefficients of each color component, and satisfy ; According to the region Calculate the spatial distance from the pixel coordinates of the current region and set the weight according to the following formula (4): Set the weight , The calculation formula of (4), where is the distance attenuation coefficient, which is used to adjust the influence degree of the spatial distance on the weight; Integrate the similarity and weight between all other labeled regions and the current region, and calculate the overall similarity value of the current region , The calculation formula of (5), where maps the distance to a similarity (value range [0,1]), and the weight is used to emphasize the influence of adjacent regions.

[0027] Among them, in step S30, the edge computing node also extracts the motion trajectory of the target device in the second data. After generating an abnormal signal, the edge computing node combines the motion trajectory with the pixel color similarity value to determine whether the labeled area is abnormally charged, and then generates a maintenance instruction according to the judgment result and the position of the labeled area; the target device includes a suspension rod and a number of feathers, the suspension rod is fixedly connected to the substation building, and multiple feathers are suspended on the suspension rod one by one; when the edge computing node identifies the motion trajectory of the target device, it first calculates the swing arc of the feathers according to the relative position between the feathers and the target area in the second data, and then calculates the swing trajectory of the feathers according to the swing arc in combination with the acquisition time of the second data, and takes the swing trajectory of the feathers as the motion trajectory of the target device.

[0028] Specifically, when the edge computing node identifies the motion trajectory of the target device, it includes the following steps: S31: For suspended feathers, establish a collaborative detection model through the following formula (6): (6), where represents the swing angle of the th feather at time, is the allowable angle deviation threshold; S32: Reconstruct the three-dimensional motion trajectory based on the three-dimensional coordinates of the feathers by binocular vision. The reconstruction formula is as shown in the following formula (7): (7), where is the baseline distance, is the focal length, is the parallax; and are the pixel coordinates of the feather in the left camera image (unit: pixel); , and are the principal point coordinates of the left camera (image center point, unit: pixel), used to correct the optical center offset.

[0029] S33: Calculate the dynamic characteristic parameters of the swing trajectory. The calculation formula of the characteristic parameters is shown in the following formula (8): (8); where is the Fourier transform of the swing angle signal , is the amplitude spectrum; is the frequency corresponding to the maximum value of the amplitude spectrum (reflecting the main swing frequency); is the frequency 's energy proportion ( ); is the energy entropy, quantifying the uncertainty of the motion (the higher the entropy value, the more disordered the swing); is the small angle perturbation at the initial moment; is the perturbation amplitude at the moment; is the Lyapunov exponent. If > 0, it indicates chaotic motion (abnormal swing).

[0030] S34: Use a support vector machine classifier for anomaly determination. The determination formula is shown in the following formula (9): ) (9), where the feature vector (the four parameters are the swing amplitude, energy entropy, main frequency component, and Lyapunov exponent extracted from formula (8)); is the Lagrange multiplier (obtained through training); is the sample label (+1 normal, -1 abnormal); is the kernel function (such as Gaussian kernel), calculating the similarity between samples; is the bias term (obtained through training); is the sign function, and the output +1 or -1 represents the classification result.

[0031] S35: Establish the spatio-temporal correlation matrix between feathers. The matrix is shown in the following formula (10): (10), is the motion correlation coefficient (such as the cross-correlation coefficient) between the th feather and the th feather, where both i and j are positive integers; When the rank of the matrix triggers an abnormal area warning. ​

[0032] More specifically, when an abnormal warning is triggered in the trigger area, the edge computing node combines the feather swing trajectory with the pixel color similarity value to establish a joint decision function, and the joint decision function is shown in the following formula (11): (11), where is the Sigmoid activation function; is the normalized swing amplitude feature; is the pixel color similarity value; is the estimated electric field strength vector, and is the trainable weight; Calculate the priority weight according to the position of the marked area, and the calculation formula is shown in the following formula (12): (12), where is the center coordinate of the th marked area, is the center of the feather swing; is the standard deviation of the Gaussian function (controlling the weight decay rate), The larger it is, the flatter the weight distribution (the higher the weight of the distant area); The smaller it is, the more concentrated the weight is around the feather center nearby; Define the abnormal level function according to the joint decision function, and the definition formula is shown in the following formula (13): L = (13), where the threshold is obtained by training historical data; Finally, generate a maintenance instruction according to the maintenance instruction parameters, and the calculation formula of the maintenance instruction parameters is shown in the following formula (14): (14), where is the required response time threshold, is the index of the marked area (such as numbers 1 to m); indicates the selection · the area with the largest As the primary inspection point, is the joint decision value for the area (output of formula (11)), is the priority weight for the area (output of formula (12)).

[0033] Among them, in step S40, when the edge computing node splits the abnormal data into multiple data packets according to the number of assisting nodes, the following steps are included: S41: Establish a node selection scoring function as shown in the following formula (15): (15), where is the round-trip delay to node , is the available bandwidth, is the CPU load rate, is the configurable weight; S42: Calculate the optimal number of shards through the following formula (16): (16), where is the total amount of abnormal data, is the network maximum transmission unit, is the benchmark bandwidth requirement; S43: Allocate the data packet size based on the node score, and the allocation formula is as shown in the following formula (17): (17); where is the assisting node index; is the score of node (calculated by formula (15)); is the sum of the scores of all assisting nodes; is the total amount of abnormal data to be transmitted; is the sum of the CPU load rates of all nodes; is the CPU load rate of node . The higher the score (better performance), the more data is allocated; the lower the load , the more data is allocated; S44: Encode using fountain code, and the encoding formula is as shown in the following formula (18): (18), where is the original data shard index (1 ≤ ≤ k, k is the total number of shards); is the coding packet index (the j-th coding packet); is the th original data shard, is a randomly generated 0 / 1 matrix element (the coefficient corresponding to the packet for the shard), .

[0034] This embodiment also includes a distributed edge computing system based on a distributed edge computing method for a power intelligent substation scenario.

[0035] In specific implementation, in the power supply network of a certain city, multiple power intelligent substations are widely distributed. According to the plan, edge computing nodes are deployed inside each substation. These nodes are like "intelligent housekeepers", continuously collecting sensor monitoring data of power equipment, such as first data like equipment temperature, current and voltage values, and at the same time collecting monitoring image data, i.e., second data, and quickly processing this information.

[0036] When the substation is running, the edge computing node quickly starts working. It accurately identifies the preset marked areas in the monitoring images, such as areas of key parts of equipment, insulating handles of the substation, the outer shell of the substation, etc., and carefully extracts the pixel color information of these areas. By comparing the color information of different marked areas, it can keenly detect subtle changes.

[0037] During the data processing process, the edge computing node is also responsible for counting the sensor monitoring data. Once it finds that the data is abnormal, such as a sudden large fluctuation in the current value, it immediately generates an abnormal signal. Then, it comprehensively combines the pixel color similarity value and the position of the marked area, quickly generates a maintenance instruction, and uploads it to the cloud in a timely manner. This process is efficient and accurate, greatly shortening the time from problem discovery to maintenance arrangement.

[0038] It is worth mentioning that in the plan, a target device composed of a suspension rod and a feather is used to assist in judgment. Under normal circumstances, the feather will swing naturally in the air. Once there is a live abnormal area in the substation, the electric field change will cause the swing trajectory of the feather to change. The edge computing node uses binocular vision technology to accurately capture the swing situation of the feather, and through a series of intelligent algorithm analyses, it can more accurately judge the live abnormal area, making the detection result more reliable.

[0039] When an abnormal data needs to be uploaded in a certain substation, the edge computing node will quickly find other edge computing nodes with the minimum communication delay as assisting nodes. Then, according to the number of assisting nodes, it cleverly splits the abnormal data into multiple data packets and sends them to the assisting nodes. These assisting nodes work together to quickly upload the data packets to the cloud, ensuring timely and accurate data transmission and avoiding information loss caused by excessive data volume or transmission failures.

[0040] The above are only embodiments of the present invention, and common knowledge such as specific structures and characteristics known in the art are not described in detail herein. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application shall be subject to the content of its claims, and the specific implementation manners and the like recorded in the specification can be used to interpret the content of the claims.

Claims

1. A distributed edge computing method based on the scenario of a power intelligent substation, characterized in that It includes the following steps: S10: Deploy edge computing nodes inside each station building, collect the sensor monitoring data of each power equipment as the first data, collect each monitoring image data as the second data, and send the first data and the second data to the edge computing nodes; S20: The edge computing node identifies the preset annotation areas in the second data, extracts the pixel color information in the annotation areas as the first information, compares the first information extracted from the current annotation area with the first information extracted from other annotation areas, and calculates the pixel color similarity value between the current annotation area and other annotation areas; S30: The edge computing node conducts statistics on the first data, generates an abnormal signal according to the statistical result. After the edge computing node generates the abnormal signal, it obtains several annotation areas corresponding to the first data, obtains the pixel color similarity values corresponding to each annotation area, generates a maintenance instruction according to the pixel color similarity value of the annotation area and the position of the annotation area, and uploads the maintenance instruction to the cloud; S40: The edge computing node obtains the data of other edge computing nodes with the minimum communication delay as assisting nodes, splits the first data and the second data corresponding to the abnormal signal into multiple data packets according to the number of assisting nodes, and sends the data packets to the assisting nodes; After receiving the data packets, the assisting nodes upload the data packets to the cloud.

2. The distributed edge computing method based on the power intelligent substation scenario according to claim 1, wherein: When the edge computing node extracts the first information, it performs color space conversion on the pixels in the annotation area, converts the RGB value to the HSV space, takes the maximum value among the red, green, and blue components as the lightness value, calculates the saturation with the ratio of the difference between the lightness and the minimum component to the lightness, and determines the hue according to the component corresponding to the maximum value among the RGB three components. The specific formula is as shown in formula (1) below: （1）。 3. The distributed edge computing method based on the power intelligent substation scenario according to claim 2, wherein: When calculating the pixel color similarity value between the current labeled area and other labeled areas, for the current labeled area , compare it one by one with each of the other labeled areas to calculate the color feature distance between the two areas , The calculation formula is as shown in formula (3) below: （3）， Among them, , , and , , are respectively the mean values of hue, saturation and lightness of regions and in the HSV color space; , , are the weight coefficients of each color component, and satisfy ; According to the region Calculate the spatial distance based on the pixel coordinates of the current region According to Set the weight The calculation formula of is as shown in formula (4) below:​ （4）， Among them, is the distance attenuation coefficient; Calculate the overall similarity value of the current region by integrating the similarities and weights of all other marked regions and the current region The overall similarity value is calculated according to the following formula (5): （5）。 4. The distributed edge computing method based on the power intelligent substation scenario according to claim 3, wherein: In step S30, the edge computing node also extracts the movement trajectory of the target device in the second data. After the edge computing node generates an abnormal signal, it combines the running trajectory with the pixel color similarity value to judge whether the annotation area is abnormally charged, and then generates a maintenance instruction according to the judgment result and the position of the annotation area; The target device includes a suspension rod and several feathers, the suspension rod is fixedly connected to the station building, and multiple feathers are hung on the suspension rod one by one; When the edge computing node identifies the movement trajectory of the target device, it first calculates the swing arc of the feather according to the relative position between the feather and the target area in the second data, and then calculates the swing trajectory of the feather according to the swing arc of the feather in combination with the acquisition time of the second data, and takes the swing trajectory of the feather as the movement trajectory of the target device.

5. The distributed edge computing method based on the power intelligent substation scenario according to claim 4, wherein: When the edge computing node identifies the movement trajectory of the target device, it includes the following steps: S31: For hanging feathers, a collaborative detection model is established through the following formula (6): （6）， Among them, represents the th feather's swing angle at the moment, and is the allowable angle deviation threshold; S32: Reconstruct the three-dimensional movement trajectory based on the three-dimensional coordinates of the feather by binocular vision. The reconstruction formula is as shown in formula (7) below: （7）， wherein is the baseline distance, is the focal length, is the parallax; S33: Calculate the dynamic characteristic parameters of the swing trajectory. The calculation formula of the characteristic parameters is as shown in formula (8) below: （8）； S34: Use a support vector machine classifier for abnormal determination. The determination formula is as shown in formula (9) below: （9）， Among them, the feature vector ; S35: Establish a spatio-temporal correlation matrix between feathers. The matrix is as shown in formula (10) below: （10）， When the matrix has a rank an area anomaly alarm is triggered.

6. The distributed edge computing method based on the power intelligent substation scenario according to claim 5, wherein: When a trigger area anomaly alarm occurs, the edge computing node combines the feather swing trajectory with the pixel color similarity value to establish a joint decision function, and the joint decision function is shown in the following formula (11): （11）， wherein, is the Sigmoid activation function; is the normalized swing amplitude feature; is the pixel color similarity value; is the estimated electric field intensity vector, and is the trainable weight; Calculate the priority weight according to the position of the marked area, and the calculation formula is shown in the following formula (12): （12）， Among them, is the central coordinate of the th marked area, is the center of the feather swing; Define the anomaly level function according to the joint decision function, and the definition formula is shown in the following formula (13): L= （13）， wherein the threshold value is obtained by training with historical data; Finally, generate a maintenance instruction according to the maintenance instruction parameters, and the calculation formula of the maintenance instruction parameters is shown in the following formula (14): （14）， wherein is the required response time threshold value.

7. The distributed edge computing method based on the power intelligent substation scenario according to claim 6, wherein: In step S40, when the edge computing node splits the abnormal data into multiple data packets according to the number of assisting nodes, the following steps are included: S41: Establish a node selection scoring function shown in the following formula (15): （15）， Among them, is the round-trip delay to node , is the available bandwidth, is the CPU load rate, is the configurable weight; S42: Calculate the optimal number of shards through the following formula (16): （16）， wherein, is the total amount of abnormal data, is the maximum network transmission unit, is the benchmark bandwidth requirement; S43: Allocate the data packet size based on the node score, and the allocation formula is shown in the following formula (17): （17）； S44: Use fountain code for encoding, and the encoding formula is shown in the following formula (18): （18）， Among them, is the original data shard, is the randomly generated matrix element, .

8. A distributed edge computing system based on the power intelligent substation scenario, characterized in that, The distributed edge computing method based on the power intelligent substation scenario described in any one of claims 1-7 is used.

Images