A Method and System for Safety Authentication and Operation Management of Power Substation Maintenance Personnel Based on Iris Recognition

Abstract

This invention provides a method and system for security authentication and operation management of power substation maintenance personnel based on iris recognition, belonging to the field of power substation operation and maintenance management technology. It pre-constructs a spatiotemporal correlation network between the iris features of maintenance personnel and dynamic operation permissions; at the main entrance of the substation, iris images are collected and matched at the main iris recognition terminal to generate a first identity identifier, extracting corresponding permission parameters and valid time windows; based on this, tasks to be executed are filtered and an authorized task list is generated and pushed for display; upon receiving the selected target task identifier, target equipment information is queried and sent to the device-level iris recognition terminal; the terminal performs secondary identity verification and joint permission verification, generating an operation permission command to drive the device to execute standard operations. This invention improves the security and reliability of power substation operation and maintenance management.

Description

Technical Field

[0001] This invention relates to the field of power substation operation and maintenance management technology, and more specifically, to a method and system for safety authentication and operation management of power substation operation and maintenance personnel based on iris recognition. Background Technology

[0002] In the operation and maintenance management of power substations, ensuring the authenticity of the identities of maintenance personnel and the accuracy of their operating permissions is a crucial link in guaranteeing the safe and stable operation of the power system. Traditional methods of authenticating maintenance personnel, such as passwords and access cards, have numerous security vulnerabilities. Passwords are easily forgotten, leaked, or cracked, and access cards may be lost or misused by others. These situations could lead to unauthorized personnel entering the substation and operating it, posing serious security risks to the power system.

[0003] Meanwhile, existing operation management methods lack precise and dynamic control over the operational permissions of maintenance personnel. They typically assign fixed permissions based solely on job title, failing to consider the dynamic changes in these permissions across different time periods and task scenarios. Furthermore, the lack of effective secondary identity verification and procedural guidance during operations increases the risk of misoperation, further threatening the safety of the power system. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for security authentication and operation management of power substation maintenance personnel based on iris recognition, the method comprising:

[0005] Pre-construct a spatiotemporal correlation network between the iris features of maintenance personnel and their dynamic operation permissions;

[0006] At the main entrance and exit of the power substation, the first iris image of the maintenance personnel is collected by the main iris recognition terminal. The first iris image is matched with the iris feature code in the spatiotemporal association network to generate a first identity identifier that is successfully matched. Based on the first identity identifier, the corresponding first job role dynamic permission parameters and the first effective time window are extracted from the spatiotemporal association network.

[0007] Based on the first identity identifier and the first valid time window, a set of tasks to be executed that overlap with the first valid time window is extracted from the operation and maintenance task database. The set of tasks to be executed is filtered based on the dynamic permission parameters of the first job role to generate a first authorized task list. The first authorized task list is then pushed to the main iris recognition terminal for display.

[0008] The system receives the target task identifier selected by the maintenance personnel in the first authorized task list, queries the corresponding target device identifier and the standard operation procedure document of the target device from the device ledger database based on the target task identifier, and sends the target device identifier and the standard operation procedure document to the device-level iris recognition terminal deployed next to the target device.

[0009] The device-level iris recognition terminal collects the second iris image of the maintenance personnel, matches the second iris image with the iris feature code in the spatiotemporal correlation network to generate a successfully matched second identity identifier, performs joint permission verification based on the second identity identifier and the target task identifier, generates an operation permission instruction, and sends the operation permission instruction to the device controller connected to the target device to drive the target device to execute the operation sequence corresponding to the standard operation procedure document.

[0010] Furthermore, embodiments of the present invention also provide a security authentication and operation management system for power substation maintenance personnel based on iris recognition, comprising:

[0011] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned iris recognition-based power substation maintenance personnel security authentication and operation management method by executing the machine-executable instructions.

[0012] Based on the above, a spatiotemporal correlation network between the iris features of maintenance personnel and dynamic operation permissions was constructed, achieving deep integration and innovation in maintenance personnel identity authentication and operation permission management. At the main entrance of the substation, iris images are collected and matched using the main iris recognition terminal, which can quickly and accurately generate a first identity identifier and extract the corresponding first job role dynamic permission parameters and first valid time window. Then, based on the first identity identifier and valid time window, a set of tasks to be executed is extracted from the maintenance task database, filtered based on the job role dynamic permission parameters, and a first authorized task list is generated and pushed for display. This ensures that maintenance personnel can only obtain tasks that match their permissions and time, effectively preventing unauthorized operations. After maintenance personnel select a target task, secondary identity verification and joint permission verification are performed through the device-level iris recognition terminal, generating an operation permission command to drive the target device to execute standard operating procedures, further ensuring the security and accuracy of the operation, and greatly improving the security and reliability of power substation operation and maintenance management. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the execution flow of the iris recognition-based safety authentication and operation management method for power substation maintenance personnel provided in an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of exemplary hardware and software components of the iris recognition-based power substation maintenance personnel safety authentication and operation management system provided in an embodiment of the present invention. Detailed Implementation

[0015] Figure 1 This is a flowchart illustrating a method for safety authentication and operation management of power substation maintenance personnel based on iris recognition, provided in one embodiment of the present invention. A detailed description follows.

[0016] Step S110: Pre-construct a spatiotemporal correlation network between the iris features of maintenance personnel and dynamic operation permissions.

[0017] Step S111: Receive the input instruction of the substation safety manager on the basic identity information of the operation and maintenance personnel through the safety management terminal, and create a basic information record unit of the operation and maintenance personnel in the local database according to the input instruction.

[0018] In the safety management room of Substation A, the safety manager faces an industrial control computer equipped with dedicated configuration software and connected to the substation's industrial Ethernet network. The manager clicks the "Add Maintenance Personnel" button on the software interface and enters information item by item in the pop-up form: "Zhang Ming" in the "Full Name" text field, "20230548" in the "Employee ID Number" field, "Substation Maintenance Team 1" in the "Department Name" drop-down menu, and "Relay Protection Maintenance Specialist" in the "Standard Job Role Name" drop-down menu. After completing the data entry, the manager clicks "Submit." The industrial control computer combines these four fields into a structured data record and inserts it into the "Basic Information Table of Maintenance Personnel" in the local relational database using a structured query language. The database automatically generates a globally unique auto-incrementing primary key value for this record, which serves as the primary internal identifier for the maintenance personnel.

[0019] Step S112: Call the dedicated iris acquisition device that is connected to the security management terminal to perform continuous iris image capture and processing on the eyes of the maintenance personnel to obtain a sequence of original iris image pairs covering the entire texture area of ​​the iris of both eyes.

[0020] After successfully creating the basic information, the industrial control computer automatically displayed an iris acquisition prompt. Maintenance personnel Zhang Ming walked to the dedicated iris acquisition device connected to the industrial control computer via a universal serial bus and aligned his eyes with the device's acquisition window. The device has two built-in independent near-infrared imaging modules, one for the left eye and one for the right eye. The device automatically triggered continuous shooting mode, acquiring 30 frames of images for the left eye and 30 frames for the right eye within 2 seconds, generating a total of 60 raw iris images with a resolution of 640*480 pixels and a grayscale depth of 8 bits. These images were named in the format "Employee ID_Eye_Frame Number," for example, "20230548_L_0001," and temporarily stored in the industrial control computer's temporary storage buffer, forming a sequence of raw iris images covering the entire texture area of ​​both irises.

[0021] Step S113: Perform image sharpness index calculation processing on each frame of the original iris image in the original iris image pair sequence, extract the image gradient amplitude distribution features and the proportion features of high frequency components in the image spectrum, generate image quality score parameters, remove blurred iris image frames whose image quality score parameters are lower than a preset quality threshold from the original iris image pair sequence, and retain the valid iris image frame sequence that meets the feature extraction requirements.

[0022] The image preprocessing software on the industrial control computer starts to evaluate the quality of each frame of the image in the sequence. Taking the first frame of the left eye image as an example, the software first calculates the gradient of the image using the Sobel operator. A 3×3 pixel horizontal convolution kernel is convolved with the image to obtain the horizontal gradient map Gx; a 3×3 pixel vertical convolution kernel is convolved with the image to obtain the vertical gradient map Gy. For each pixel point, calculate its gradient magnitude G which is the square root of the sum of the squares of Gx and Gy. Statistically calculate the gradient magnitudes of all pixel points in the entire image, calculate the average value μ_G and standard deviation σ_G of the gradient magnitudes, and take the product of μ_G and σ_G as the gradient magnitude distribution eigenvalue A = μ_G×σ_G. At the same time, perform a two-dimensional discrete Fourier transform on this frame of image, converting the image from the spatial domain to the frequency domain to obtain a spectrogram. Calculate the total energy E_high of all frequency components within the high-frequency annular region where the frequency is higher than 30 line pairs per millimeter in the spectrogram, and then calculate the total energy E_total of the entire spectrogram. Take the ratio of E_high to E_total as the high-frequency component ratio eigenvalue B = E_high / E_total. Multiply the eigenvalue A by the preset first weight coefficient 0.6, and add it to the eigenvalue B multiplied by the preset second weight coefficient 0.4 to obtain the quality scoring parameter Q of this frame of image, Q = 0.6×A + 0.4×B. Compare Q with a preset quality threshold Q_th. If Q < Q_th, then determine that this frame of image is a blurred image and delete it from the original sequence. Repeat the above processing process for all 60 frames of images, and finally retain all the image frames with qualified quality scoring parameters to form a sequence of valid iris image frames.

[0023] Step S114, perform iris region localization processing on each frame of valid iris image in the sequence of valid iris image frames. Respectively fit the pupil boundary circular parameters and iris outer boundary circular parameters through the iris inner edge detection algorithm and iris outer edge detection algorithm, and calculate the radius range and center coordinates of the iris annular region according to the pupil boundary circular parameters and iris outer boundary circular parameters.

[0024] For the first frame of the left eye in the valid sequence, an integral-differential detection operator is used for iris localization. First, every pixel in the image is traversed within a preset pupil radius as a candidate circle center. For each candidate circle center and radius, the sum of grayscale values ​​of all pixels on the circumference of that circle center and radius is calculated. As the radius changes, the derivative of this sum with respect to the radius is calculated; the position with the largest absolute value of the derivative corresponds to the circumference with the most drastic grayscale value change, which is the pupil boundary. The corresponding pupil center coordinates and pupil radius are recorded. After pupil localization, the search range is limited to an annular region centered on the pupil. The integral-differential detection operator is applied again, but the search direction and radius range are adjusted to detect the outer boundary between the iris and sclera, obtaining the outer boundary circle center coordinates and outer boundary radius. Based on the inner and outer boundary circle parameters, the radial range of the annular region of the iris is from the inner edge to the outer edge, and the center point coordinates of the annular region are taken as the weighted average of the inner and outer circle centers.

[0025] Step S115: Extract the iris annular region image from the effective iris image according to the radius range and center coordinates, perform image normalization processing on the iris annular region image, and unfold the iris annular region image from polar coordinate space into a rectangular normalized iris image of fixed size to obtain an iris texture unfolded map with uniform geometric size.

[0026] Based on the parameters obtained in step S114, a rubber sheet model is used for normalization. The iris annular region is mapped from Cartesian coordinates to polar coordinates, where the radial coordinates linearly change from points on the pupil boundary to points on the outer boundary of the iris, and the angular coordinates range from 0 degrees to 360 degrees. The radial resolution is set to 64 sampling points, and the angular resolution to 512 sampling points. For each normalized coordinate point, the corresponding pixel grayscale value in the original image is calculated using a bilinear interpolation algorithm. After this transformation, the entire iris annular region is unfolded into a rectangular normalized iris image with 64 rows and 512 columns. The rows of this image correspond to different radii, and the columns correspond to different angular positions, eliminating the effects of pupil scaling and image rotation.

[0027] Step S116: The iris texture feature point extraction algorithm is used to perform multi-scale two-dimensional Gabor filter convolution processing on the iris texture unfolded map, and the filter response amplitude and phase information of each pixel in the iris texture unfolded map in multiple directions and multiple frequencies are extracted to generate the original iris texture feature point distribution map.

[0028] A two-dimensional Gabor filter bank is applied to the normalized rectangular iris image. In this embodiment, Gabor filters with 4 scales and 8 directions are used, for a total of 32 filters. For each pixel in the image, a convolution operation is performed on it using these 32 filters. The convolution result of each filter with the image is a complex number. Taking the modulus of this complex number yields the filter response amplitude of the pixel at that scale and direction, and taking the principal argument value of this complex number yields the phase information. Therefore, for each pixel, a multidimensional feature vector consisting of 32 amplitude values ​​and 32 phase values ​​is generated. The multidimensional feature vectors of all pixels are combined to form a three-dimensional original iris texture feature point distribution map, with dimensions of 64 rows by 512 columns by 32 directional scales.

[0029] Step S117: Perform feature point filtering processing on the original iris texture feature point distribution map, remove noise feature points with response amplitudes lower than a preset response threshold, obtain the filtered core texture feature points, and generate feature descriptor vectors of the core texture feature points based on the coordinate positions of the core texture feature points in the iris texture unfolding map and the corresponding phase information.

[0030] The 3D distribution map generated in step S116 is filtered. An amplitude threshold is set. For each pixel location, the maximum value among its 32 filtered response amplitudes is checked. If this maximum value is lower than the amplitude threshold, the texture information at that pixel is considered weak and easily affected by noise, and it is discarded. Only pixels with a maximum amplitude value higher than the amplitude threshold are retained as core texture feature points. For each retained core point, its radial and angular coordinates in the normalized image are recorded, and the filtered phase information of that point in the 32 directions is quantized into 2 bits and combined into a 64-bit feature descriptor vector.

[0031] Step S118: The feature descriptor vectors of all the core texture feature points corresponding to the same maintenance personnel are concatenated and combined in a preset order to generate a unique binary format iris feature code for the maintenance personnel.

[0032] The feature descriptor vectors of all core points obtained in step S117 are concatenated in order of radial coordinates as the primary sequence and angular coordinates as the secondary sequence in the normalized image. Assuming a total of 200 core points are selected, each feature descriptor is 64 bits, and the concatenation results in a binary string of length 12800 bits. This binary string is the unique iris feature code of maintenance personnel Zhang Ming.

[0033] Step S119: Encrypt the iris feature encoding of the maintenance personnel using a secure hash algorithm to generate an encrypted iris feature ciphertext string. Then, associate the encrypted iris feature ciphertext string with the standard job role name text field in the maintenance personnel basic information record unit and store it on a dual-machine hot standby server.

[0034] The industrial control computer uses a secure hash algorithm to calculate the 12,800-bit iris feature code generated in step S118, resulting in a 256-bit hash value. This hash value is then converted into a 64-character hexadecimal string, which serves as the encrypted iris feature ciphertext string. The industrial control computer associates this ciphertext string with the standard job title text field in the previously created maintenance personnel basic information record unit via industrial Ethernet, encapsulates it into a data packet, and sends it to the server located in the substation's computer room through an encrypted transmission channel. The server employs a dual-machine hot standby architecture, with the two servers serving as backups for each other. The database management system writes this record into the iris feature permission association table.

[0035] Step S1110: Based on the electronic map of functional areas within the power substation and the responsibility scope description document corresponding to the text field of the standard job role name, parse out the access permission status that the standard job role name should have at the entrance of each functional area.

[0036] The permission resolution engine on the server loads an electronic map of Substation A. This map divides the substation into multiple functional areas, such as the main control room, relay protection room, and 10kV high-voltage room, each associated with a unique area code. Simultaneously, the engine reads the responsibility scope description document bound to the role of relay protection maintenance specialist. This responsibility scope description document is a structured Extensible Markup Language (XML) file. The permission resolution engine iterates through each functional area on the electronic map, performing text matching between the description in the responsibility scope description document and the area name. If a match is successful, an access permission status (represented by binary "1") is generated for that role in that area; if a match fails, an access prohibition status is generated (represented by binary "0").

[0037] Step S1111: Extract the functional area name text and area code number corresponding to the allowed binary status identifier to generate a list of area codes that the standard job role name can pass through. Sort each area in the area code list from high to low according to the area security level value to obtain the spatial access priority sorting sequence in the job role dynamic permission parameter.

[0038] Based on the access permission status generated in step S1110, the permission parsing engine filters out all areas with a status of "1". For each of these areas, its area name text and area code number are extracted to form a list of accessible areas. Then, the security level value of each area is queried from the equipment ledger database; for example, the main control room has a security level of 1 (highest), and the relay protection room has a security level of 2. The engine sorts this list from high to low according to the security level value, generating a spatial access priority sorting sequence in the dynamic permission parameters for job roles.

[0039] Step S1112: Obtain the unique identification code of each key equipment in the power substation, as well as the planned operation and maintenance time window and temporary operation and maintenance time window of each key equipment. After merging and deduplicating the planned operation and maintenance time window and the temporary operation and maintenance time window, generate a list of valid time windows for the key equipment.

[0040] Information on all critical equipment within Substation A is extracted from the production management system. Taking main transformer No. 1 as an example, its unique equipment identifier is "EQP_MAIN_001". All planned maintenance time windows for this equipment are retrieved, such as the annual preventative testing window, as well as temporary maintenance time windows generated due to defect handling. The time window management module on the server merges these two time windows, taking the union if there is overlap, and finally generates a unique, ordered list of time windows, which serves as the valid time window list for main transformer No. 1.

[0041] Step S1113 involves performing a four-dimensional association binding operation on the standard job title text field, the spatial access priority sorting sequence, the list of effective time windows for key equipment, and the encrypted iris feature ciphertext string to form a complete node record unit in the spatiotemporal association network.

[0042] The association binding module on the server creates a new data structure containing four core fields: the first field stores the standard job title "Relay Protection Maintenance Specialist"; the second field stores the space access priority sorting sequence generated in step S1111; the third field is a set of key-value pairs where the key is the device's unique identifier and the value is a list of valid time windows for that device; and the fourth field stores the encrypted iris feature ciphertext string generated in step S119. Linking these four parts of information constitutes a complete node record unit describing the relationships of maintenance personnel Zhang Ming across four dimensions: iris feature, job title, space permissions, and time permissions.

[0043] Step S1114: Organize the data storage structure of the multiple complete node record units according to the ascending order of the employee ID number field of the operation and maintenance personnel, construct the complete graph data structure of the spatiotemporal correlation network, and synchronously store the complete graph data structure to the dual-machine hot standby server and the remote cloud database.

[0044] For all maintenance personnel within substation A, steps S111 to S1113 are repeated to generate their respective corresponding node record units. The graph database engine on the server organizes these node record units, using the maintenance personnel's employee ID as the unique identifier for each node. Each node contains attribute information across the four dimensions mentioned above. Simultaneously, the engine creates logical edges between nodes based on job roles, shared equipment, and other relationships, ultimately forming a multi-relational graph data structure—a spatiotemporal relational network. After construction, a complete copy of this graph data structure is written to another server forming a dual-machine hot standby, and simultaneously transmitted to a remote cloud database via an encrypted 5G virtual private network.

[0045] Step S120: At the main entrance and exit of the power substation, the first iris image of the maintenance personnel is collected by the main iris recognition terminal. The first iris image is matched with the iris feature code in the spatiotemporal association network to generate a first identity identifier that has been successfully matched. Based on the first identity identifier, the corresponding first job role dynamic permission parameters and the first effective time window are extracted from the spatiotemporal association network.

[0046] Step S121: Monitor the sensing level signal output by the infrared proximity sensor built into the main iris recognition terminal. When the sensing level signal changes from a low level to a high level, it is determined that an operation and maintenance personnel are approaching the main iris recognition terminal, triggering the main iris recognition terminal to switch from standby power saving mode to full-function operation mode and generate a first iris recognition trigger signal.

[0047] The main iris recognition terminal installed at the main entrance of Substation A has a built-in infrared proximity sensor that continuously emits infrared light and monitors reflected signals. When no one is nearby, the sensor outputs a low-level signal, and the terminal is in standby mode. When maintenance personnel Zhang Ming walks to within about 0.5 meters of the terminal, the infrared light reflected from his body is received by the sensor, and the sensor output level jumps from low to high. The terminal's main controller detects this rising edge-triggered interrupt signal, immediately executes the wake-up program, increases the processor frequency to the normal operating frequency, lights up the touch screen, and simultaneously generates a Boolean flag in memory and sets it to true, indicating that the first iris recognition trigger signal has been generated, and the terminal enters full-function operation mode.

[0048] Step S122: In the full-function operation mode, the near-infrared light-emitting diode array inside the main iris recognition terminal is controlled to emit near-infrared light to the eye area of ​​the maintenance personnel with a preset pulse width modulation waveform to illuminate the iris tissue.

[0049] Upon entering full-function mode, the main controller immediately activates the near-infrared LED driver circuit connected to the general-purpose input / output interface. This driver circuit operates according to the pulse width modulation signal output by the controller, with preset parameters of a 1kHz frequency and a 50% duty cycle, causing the LEDs to flash at a frequency of 1000 times per second, conducting for half a cycle and turning off for the other half. Near-infrared light with a wavelength of 850nm is emitted from the LED array, uniformly illuminating Zhang Ming's eye area, resulting in a clear image of the iris tissue.

[0050] Step S123: Drive the high-resolution complementary metal-oxide-semiconductor image sensor inside the main iris recognition terminal to continuously capture multiple frames of original images containing the eye regions of the maintenance personnel, and obtain a continuous time series of original image frame sequences.

[0051] While illuminated by near-infrared light, the main controller sends a continuous shooting command to the high-resolution complementary metal-oxide-semiconductor image sensor inside the terminal. This sensor has a resolution of 1280*1024 pixels and a frame rate of 30fps. The sensor begins exposing at fixed intervals and transmits the read pixel grayscale data to the main controller via the mobile industrial processor interface bus. The main controller temporarily stores this data in a circular buffer in dynamic random access memory, forming a raw image frame sequence of 90 frames lasting 3 seconds.

[0052] Step S124: Perform human eye detection processing based on gray-scale integral projection on each frame of the original image in the original image frame sequence. Determine the vertical coordinate interval of the human eye region through the horizontal gray-scale integral projection curve, and determine the horizontal coordinate interval of the human eye region through the vertical gray-scale integral projection curve. Based on the vertical coordinate interval and the horizontal coordinate interval, crop out a local image block containing only the human eye region from the original image to obtain a human eye local image sequence.

[0053] The digital signal processor on the main controller begins processing each frame of the image in the circular buffer. Taking the first frame as an example, it is first converted into a grayscale image. The horizontal grayscale integral projection is calculated by summing the grayscale values ​​of all pixels in each row of the image, resulting in a curve. Since the grayscale value of the human eye region is lower than the surrounding area, two distinct troughs appear on the curve, thus determining the vertical coordinate range of the human eye region. Within this vertical range, the vertical grayscale integral projection is calculated by summing the grayscale values ​​of all pixels in each column, resulting in another curve. By analyzing the trough positions, the horizontal coordinate ranges of the left and right eyes are determined. Finally, based on these coordinate ranges, local image patches containing the left and right eyes are cropped from the original image. This process is repeated for each frame in the sequence, generating a sequence of 90 pairs of left and right eye images.

[0054] Step S125: Perform image sharpness evaluation processing based on the Laplacian operator on each frame of the human eye local image sequence, calculate the Laplacian operator response value of each pixel in the human eye local image, sum the Laplacian operator response values ​​of all pixels to generate an image sharpness score, remove blurry human eye local images with image sharpness scores lower than a preset sharpness threshold from the human eye local image sequence, and retain valid human eye local image sequences that meet the sharpness requirements.

[0055] Sharpness is evaluated for each frame of the local human eye image. The Laplacian operator is used, with a 3x3 pixel Laplacian kernel convolved with the image. Each pixel in the convolution result receives a Laplacian response value, which reflects the intensity of gray-level abrupt changes in its neighborhood. The absolute values ​​of the Laplacian response values ​​of all pixels are summed to obtain the total sharpness score of the image. This score is compared with a preset sharpness threshold; if it is below the threshold, the image is considered blurry and is removed from the local human eye image sequence. The images for the left and right eyes are processed separately, and the remaining frames constitute the valid local human eye image sequence.

[0056] Step S126: Perform motion feature analysis processing based on optical flow method on each frame of the effective human eye local image sequence, calculate the optical flow vector of the pixels between two adjacent effective human eye local images, and generate eye micro-motion feature curves based on the direction consistency parameter and amplitude stability parameter of the optical flow vector.

[0057] The sparse optical flow method is used to calculate motion between consecutive images. Taking the first two frames of the effective sequence for the left eye as an example, a set of corner features are detected in the first frame. Assuming that the positions of these corner points remain unchanged in the second frame, the positions with the smallest gray-level changes are searched within the neighborhood centered on their original coordinates. The new coordinates of these corner points in the second frame are then found, and the displacement vector of each corner point is the optical flow vector. Statistical analysis is performed on the optical flow vectors of all matched corner points to calculate the average direction and average amplitude. Then, the deviation of each vector direction from the average direction and the deviation of each vector amplitude from the average amplitude are calculated to obtain the direction consistency parameter and amplitude stability parameter. The direction consistency parameter and amplitude stability parameter calculated between consecutive frames are connected in chronological order to form the eye micro-motion feature curve.

[0058] Step S127: Input the eye micro-motion feature curve into a pre-trained liveness detection classifier for real / fake detection processing. The liveness detection classifier outputs the probability value of the effective human eye local image sequence belonging to a real live iris. When the probability value is greater than a preset liveness probability threshold, a liveness detection pass result is generated.

[0059] The eye micro-motion feature curve generated in step S126 is input into a pre-trained liveness detection classifier deployed within the main iris recognition terminal. This pre-trained liveness detection classifier employs a support vector machine model and is pre-trained using micro-motion feature data containing tens of thousands of real liveness samples and tens of thousands of spoofing attack samples. The classifier maps the input feature vector sequence to a high-dimensional feature space using a kernel function, calculates the confidence score of the sequence belonging to a real liveness sample based on the trained classification hyperplane, and converts it into a probability value between 0 and 1 using a logistic function. This probability value is compared with a preset liveness probability threshold; if it is greater than the threshold, it is determined to be a real liveness sample, and the terminal generates a liveness detection flag that is set to true; otherwise, it is determined to be a spoofing attack, and the recognition process terminates.

[0060] Step S128: Based on the liveness detection result, select the effective human eye local image with the highest image quality score from the effective human eye local image sequence as the first iris image.

[0061] After generating the liveness detection result in step S127, the effective local image sequences of the left and right eyes are traversed. For each frame, the Laplacian sharpness score calculated in step S125 is called. The scores of all left-eye images are compared, and the frame with the highest score is selected. Similarly, the frame with the highest score is selected for the right eye. These two frames are used together as the first iris image for subsequent feature extraction and matching.

[0062] Step S129: Perform iris region localization, normalization processing and feature extraction on the first iris image, and generate the first iris feature code using the same iris texture feature point extraction algorithm as when constructing the spatiotemporal correlation network.

[0063] For the left-eye image in the first iris image selected in step S128, the processing flow from steps S114 to S118 is repeated. First, iris region localization is performed to obtain the pupil boundary and iris outer boundary parameters. Then, normalization processing is performed to expand the iris annular region into a rectangular normalized iris image. Next, the same 4-scale, 8-direction two-dimensional Gabor filter bank is used for convolution processing to extract the filter response amplitude and phase information. Then, feature point selection is performed to retain the core texture feature points. Finally, the feature descriptor vectors of the core points are concatenated to generate the iris feature code of the left-eye image. The same processing flow is performed on the right-eye image to generate the iris feature code of the right-eye image. The iris feature codes of the left and right eyes are concatenated in a preset order to obtain the final first iris feature code.

[0064] Step S1210: Calculate the Hamming distance between the first iris feature code and the iris feature code of each maintenance personnel in the spatiotemporal association network to obtain a first similarity score set. Select the maintenance personnel iris feature code corresponding to the smallest Hamming distance from the first similarity score set as the benchmark feature code for successful matching.

[0065] The main iris recognition terminal sends the first iris feature code generated in step S129 to the server via industrial Ethernet. The iris feature comparison module on the server compares this code bit-by-bit with the iris feature codes of each maintenance personnel stored in the spatiotemporal correlation network, calculating the Hamming distance. The Hamming distance is defined as the number of different bits at corresponding positions between two equal-length binary strings; the smaller the distance, the more similar the two iris features. After calculating the similarity for all stored iris feature codes, a first similarity score set is obtained, containing the Hamming distance value corresponding to each code. The code with the smallest Hamming distance is selected from the set as the baseline feature code that successfully matches the currently acquired first iris feature code.

[0066] Step S1211: Perform a reverse index query in the spatiotemporal correlation network based on the successfully matched baseline feature code to obtain a first identity identifier uniquely bound to the baseline feature code.

[0067] The server uses the successfully matched baseline feature code found in step S1210 as the query condition and performs a reverse index query in the iris feature permission association table of the spatiotemporal association network. Since the baseline feature code has a one-to-one binding relationship with the first identity identifier in the basic information record unit of the operation and maintenance personnel, the query result directly returns the first identity identifier corresponding to the code, that is, the internal database primary key value of the operation and maintenance personnel Zhang Ming.

[0068] Step S1212: Extract the corresponding first job role dynamic permission parameters and the first valid time window from the spatiotemporal association network according to the first identity identifier, and compare the first valid time window with the current system time. If the current system time falls within the first valid time window, retain the first job role dynamic permission parameters; otherwise, mark the first job role dynamic permission parameters as invalid and trigger a permission expiration alarm.

[0069] Based on the first identity identifier obtained in step S1211, the server locates the complete node record unit corresponding to the maintenance personnel Zhang Ming in the spatiotemporal correlation network. From this node record unit, the server extracts the dynamic permission parameters for his job role, including the spatial access priority sorting sequence and the list of effective time windows for key equipment. Simultaneously, it extracts the first effective time window associated with this node, which is typically a globally valid working time period. The server obtains the current system timestamp and compares it with the start and end times of the first effective time window. If the current system timestamp is greater than or equal to the start time and less than or equal to the end time, it is determined that the current time is within the effective window, and the extracted dynamic permission parameters for job role are retained for later use. If the current time is not within the window, the dynamic permission parameters for the job role are marked as invalid, and an expired permission alarm message is simultaneously sent to the main iris recognition terminal and the monitoring center via industrial Ethernet, triggering the terminal to display an expired permission prompt.

[0070] Step S130: Extract a set of tasks to be executed that overlap with the first effective time window from the operation and maintenance task database based on the first identity identifier and the first effective time window; filter the set of tasks to be executed based on the dynamic permission parameters of the first job role; generate a first authorized task list; and push the first authorized task list to the main iris recognition terminal for display.

[0071] Step S131: Access the operation and maintenance task database deployed on the dual-machine hot standby server, and read all operation and maintenance task records for the day stored in the operation and maintenance task database.

[0072] The maintenance task management module on the server accesses the maintenance task database deployed on the dual-machine hot standby server. This maintenance task database uses a relational database management system. The module executes a structured query language statement to select all records whose task schedule start timestamps are between midnight of the current day and midnight of the next day. Each returned maintenance task record contains multiple fields: task number (e.g., "TASK_20231029_001"), task type (e.g., "periodic inspection"), task schedule start timestamp and task schedule end timestamp, task-associated device identifier (e.g., "EQP_MAIN_001"), and a comma-separated list of roles allowed to execute the task (e.g., "relay maintenance specialist, primary maintenance specialist").

[0073] Step S132: parse the first effective time window, obtain the start time and end time of the first effective time window, and determine the time interval overlap between the task plan start time stamp and the task plan end time stamp and the start time and end time of the first effective time window, respectively.

[0074] The first valid time window extracted from the spatiotemporal correlation network is analyzed to obtain its start time T_start and end time T_end. For each maintenance task record read in step S131, its task plan start timestamp T_task_start and task plan end timestamp T_task_end are extracted. The logic for determining whether two time intervals overlap is as follows: if T_task_start is less than or equal to T_end and T_task_end is greater than or equal to T_start, then the two intervals overlap. This condition is used to determine each task record.

[0075] Step S133: For each maintenance task record, if the task plan start timestamp is earlier than or equal to the end time of the first effective time window and the task plan end timestamp is later than or equal to the start time of the first effective time window, then it is determined that the time interval of the maintenance task record overlaps with the first effective time window, and the maintenance task record is included in the candidate task set.

[0076] For each maintenance task record, the overlap judgment condition from step S132 is applied. If the condition is met, the task record is added to the candidate task set. If the condition is not met, i.e., the two time intervals do not overlap at all, the task record is skipped. After traversing all task records for the day, a candidate task set is obtained, in which all tasks have an intersection in execution time with the first effective time window of maintenance personnel Zhang Ming.

[0077] Step S134: Extract the list of roles that the task can execute for each operation and maintenance task record in the candidate task set, and perform an intersection operation on the list of roles that the task can execute for and the standard job role names contained in the dynamic permission parameters of the first job role.

[0078] For each task record in the candidate task set, extract the "role list field" for which the task is allowed to execute. This role list field is a string, which is parsed into a set of role names. Simultaneously, obtain the standard job role name "Relay Maintenance Specialist" for maintenance personnel Zhang Ming from the first job role dynamic permission parameters extracted in step S1212. Perform an intersection operation on these two sets to determine whether "Relay Maintenance Specialist" appears in the role list for which the task is allowed to execute.

[0079] Step S135: If the result of the intersection operation is not empty, it is determined that the role and permission matching of the operation and maintenance task record is successful, and the operation and maintenance task record is retained; otherwise, the operation and maintenance task record is removed from the candidate task set, and a set of tasks to be executed after dual filtering by time window and role and permission is obtained.

[0080] For each candidate task, if the intersection operation result in step S134 is not empty, meaning "Relay Maintenance Specialist" is one of the roles allowed to execute the task, then the role permission matching for the task is deemed successful, and the task record is retained in the set. If the intersection operation result is empty, meaning the task is not allowed to be executed by the Relay Maintenance Specialist, then the task record is removed from the candidate task set. After all candidate tasks have been processed, the remaining task records constitute a set of tasks to be executed, filtered by both time windows and role permissions.

[0081] Step S136: Perform task priority parsing on each maintenance task record in the set of tasks to be executed, extract the value of the task priority field, sort the set of tasks to be executed in descending order of task priority value, and generate the first authorized task list.

[0082] Each task record in the set of tasks to be executed contains a task priority field, where the value is an integer, such as 1 for highest priority, 2 for medium priority, and 3 for low priority. The sorting module sorts the set of tasks to be executed in ascending order according to this priority value, with the task with the lowest priority value at the beginning of the list. After sorting, a first authorized task list is generated, where each item in the list is, in order, the task number, task type, task start time stamp, and task end time stamp.

[0083] Step S137: Encapsulate the first authorized task list into a data format conforming to the main iris recognition terminal display protocol, and send the encapsulated data to the main iris recognition terminal via industrial Ethernet.

[0084] The first authorized task list generated in step S136 is encapsulated according to the communication protocol agreed upon with the main iris recognition terminal. It uses a custom binary format and includes a message header, message length, number of tasks, and various fields for each task. After encapsulation, the data packet is sent to the main iris recognition terminal at the main entrance / exit via industrial Ethernet.

[0085] In step S138, after receiving the first authorized task list, the main iris recognition terminal displays a summary of each task in a list format on the touch screen.

[0086] After receiving the data packet, the main iris recognition terminal parses it according to the agreed protocol to reconstruct the first authorized task list. The user interface rendering engine on the terminal presents the task information in the list as a vertical list on its touch screen. Each line displays a summary of the task information, including the task number (e.g., "TASK_20231029_001"), the task type (e.g., "periodic inspection"), the task start time stamp (e.g., "2024-05-20 09:00"), and the task end time stamp (e.g., "2024-05-20 12:00").

[0087] Step S139: Set a touch selection button for each task on the touch screen of the main iris recognition terminal. When the maintenance personnel click the touch selection button, a task selection signal containing the corresponding task number is generated and the task selection signal is sent back to the server.

[0088] To the right of each task summary displayed on the touchscreen, the interface rendering engine draws a touch selection button with the word "Select" displayed on it. When maintenance personnel Zhang Ming taps the selection button corresponding to one of the tasks, the touchscreen controller detects the touch event, calculates the coordinates of the clicked location, and determines the task number corresponding to the clicked button by comparing it with the interface layout. The terminal generates a task selection signal message containing the selected task number. The terminal then transmits this message back to the server via industrial Ethernet.

[0089] Step S140: Receive the target task identifier selected by the maintenance personnel in the first authorized task list, query the corresponding target device identifier and the standard operation procedure document of the target device from the device ledger database according to the target task identifier, and send the target device identifier and the standard operation procedure document to the device-level iris recognition terminal deployed next to the target device.

[0090] Step S141: The main iris recognition terminal receives the task selection signal generated after the maintenance personnel click the touch selection button through the communication module, and parses the task number of the selected target task from the task selection signal as the target task identifier.

[0091] The server's communication module continuously listens to the data ports of each terminal. When it receives the task selection signal message returned by the main iris recognition terminal, it parses the data payload of the message and extracts the task number field contained therein. The value of the task number field is the identifier of the target task selected by the maintenance personnel Zhang Ming, for example, "TASK_20231029_001".

[0092] Step S142: Based on the target task identifier, perform an index query in the operation and maintenance task database to extract the task-associated device identifier corresponding to the target task identifier, and obtain the target device identifier.

[0093] Using the target task identifier "TASK_20231029_001" parsed in step S141 as the query condition, an index query is performed on the task table in the operation and maintenance task database. Since a unique index is established on the task number field, the query efficiency is extremely high. From the records returned by the query, the task-associated device identifier field is extracted. The value of this field is the unique identifier of the device bound to the target task, thus obtaining the target device identifier, for example, "EQP_MAIN_001".

[0094] Step S143: Query the equipment ledger database according to the target equipment identifier, and read the equipment type, equipment model, equipment installation location coordinates and the standard operation procedure document corresponding to the target equipment identifier stored in the equipment ledger database.

[0095] Using the target equipment identifier "EQP_MAIN_001" obtained in step S142 as the query condition, a query is performed in the equipment ledger database. The equipment ledger database stores detailed information on all equipment within the substation. The records returned by the query contain multiple fields: equipment type field, such as "main transformer"; equipment model field, such as "SFZ11-50000 / 220"; equipment installation location coordinate field, such as a structure containing longitude, latitude, or substation grid coordinates; and standard operating procedure document field, which stores a Uniform Resource Locator (URL) pointing to the document's storage location or the document itself as a binary large object. The standard operating procedure document is a structured Extensible Markup Language (EXPLAIN) file containing a list of pre-operation safety checks for the target equipment, a sequence of operating steps, the specified range of operating parameters for each step, and a list of post-operation status confirmation items.

[0096] Step S144: Based on the target device identifier, retrieve the list of valid time windows for the target device in the spatiotemporal correlation network, extract the planned maintenance time window and temporary maintenance time window for the target device, and compare the current system time with the list of valid time windows for the target device.

[0097] Returning to the spatiotemporal correlation network, the target device identifier "EQP_MAIN_001" is used as the key to search the list of valid time windows for key devices constructed in step S1112. The list of valid time windows corresponding to this device is obtained. This list contains one or more time windows, each with a start and end time. The current system timestamp is obtained, and each time window in the list is traversed to determine whether the current timestamp falls within the interval of any window.

[0098] Step S145: If the current system time is not within any time window in the list of valid time windows of the target device, a device operation time violation alarm is generated and pushed to the main iris recognition terminal and the monitoring center.

[0099] If the iteration result in step S144 finds that the current system time does not match any of the time windows in the list, meaning the current time is not within any time period during which the device is allowed to be operated, it is determined that the device operation time is in violation. An alarm message is immediately generated, containing the alarm type "Device Operation Time Violation," the target device identifier "EQP_MAIN_001," the current system timestamp, and a description of the violation. This alarm is pushed to the main iris recognition terminal via industrial Ethernet, a red alarm window pops up on the terminal display, and simultaneously pushed to the large screen display system and audible alarm in the monitoring center via a separate alarm channel.

[0100] Step S146: If the current system time is within the list of valid time windows of the target device, then based on the installation location coordinates of the target device and the location coordinates of the main iris recognition terminal, generate navigation path planning data from the main entrance / exit to the installation location of the target device.

[0101] If the comparison in step S144 is successful and the current time is within the valid window, continue with navigation path planning. Obtain the target equipment installation location coordinates from the equipment ledger database and the fixed installation location coordinates of the main iris recognition terminal from the configuration information. Call the built-in path planning engine, which is based on the substation's vector electronic map and traffic network model. The engine uses the A* algorithm to search for the shortest feasible path from the starting point coordinates to the ending point coordinates in the traffic network map, considering factors such as whether the passage is unobstructed and whether there are temporary blockages. After a successful search, navigation path planning data is generated, where each point contains longitude, latitude, or grid coordinates, as well as turning instructions.

[0102] Step S147: Package the navigation path planning data, the target device identifier, and the standard operation procedure document into a data package to generate a combined data package containing navigation information and operation instructions.

[0103] The navigation path planning data generated in step S146, the target device identifier "EQP_MAIN_001" obtained in step S142, and the standard operation procedure document obtained in step S143 are integrated. The data is then encapsulated according to a custom combined data packet format, which includes a data packet type identifier, device identifier field length, device identifier field value, navigation data length, navigation data sequence, document data length, and document binary data.

[0104] In step S148, the combined data packet is sent to the device-level iris recognition terminal deployed next to the target device via industrial Ethernet. After receiving the packet, the device-level iris recognition terminal displays the navigation path and operation guidance information on its display screen.

[0105] The combined data packet generated in step S147 is sent via industrial Ethernet to the device-level iris recognition terminal deployed next to the target device. This device-level iris recognition terminal has a unique network address. Upon receiving the data packet, the terminal parses it according to the same protocol, reconstructing the navigation path planning data, the target device identifier, and the standard operating procedure document. The graphical user interface on the terminal displays the navigation path as a dynamic arrow and path line overlaid on the substation floor plan, while simultaneously displaying a summary of the standard operating procedure document, such as the total number of operation steps and the current step, in the lower area of ​​the screen.

[0106] Step S150: The second iris image of the maintenance personnel is acquired by the device-level iris recognition terminal. The second iris image is matched with the iris feature code in the spatiotemporal correlation network to generate a successfully matched second identity identifier. Joint permission verification is performed based on the second identity identifier and the target task identifier to generate an operation permission instruction. The operation permission instruction is sent to the device controller connected to the target device to drive the target device to execute the operation sequence corresponding to the standard operation procedure document.

[0107] Step S151: Monitor the proximity sensor built into the device-level iris recognition terminal. When the terminal detects that the maintenance personnel have arrived at the operation area next to the target device, wake up the device-level iris recognition terminal to enter the working state and display the navigation path planning data and the summary of the standard operation procedure document on its display screen.

[0108] The equipment-level iris recognition terminal deployed next to main transformer No. 1 has a built-in infrared proximity sensor. When maintenance personnel Zhang Ming arrives within approximately 0.5 meters of the equipment following the navigation path, the sensor's output level changes. Upon detecting this signal, the terminal's main controller wakes up from standby mode. The terminal automatically illuminates the touchscreen display and displays the navigation path planning data received and stored locally in step S148 as a thumbnail in a corner of the screen. Simultaneously, a summary of the standard operation procedure document is displayed in the main area of ​​the screen, including the total number of operation steps and a brief description of the first step to be performed.

[0109] Step S152: The iris acquisition module of the device-level iris recognition terminal acquires real-time iris images of the eyes of the maintenance personnel, and performs liveness detection and image quality screening on the real-time iris images to obtain a second iris image that meets the quality requirements.

[0110] The device-level iris recognition terminal activates its internal near-infrared light-emitting diodes and image sensors, repeating the process identical to steps S122 to S128. First, near-infrared light is emitted to illuminate Zhang Ming's eyes, and multiple frames are captured consecutively. Then, each frame undergoes human eye detection, cropping, Laplacian sharpness evaluation, and liveness detection based on optical flow. After liveness detection, the frame with the highest quality score for both the left and right eyes is selected from the image frames that meet the sharpness requirements, and used as the second iris image.

[0111] Step S153: Perform iris region localization, normalization processing and feature extraction on the second iris image, and generate the second iris feature code using the same iris texture feature point extraction algorithm as when constructing the spatiotemporal correlation network.

[0112] For the second iris image selected in step S152, repeat the complete processing flow from steps S114 to S118. Perform iris localization, normalization, Gabor filtering, feature point selection, and feature descriptor concatenation on the left-eye image to generate the left-eye iris feature code. Perform the same processing on the right-eye image to generate the right-eye iris feature code. Concatenate the left and right eye codes to obtain the final second iris feature code.

[0113] Step S154: Calculate the Hamming distance between the second iris feature code and the iris feature code of each maintenance personnel in the spatiotemporal association network to obtain a second similarity score set. Select the maintenance personnel iris feature code corresponding to the smallest Hamming distance from the second similarity score set as the second benchmark feature code for successful matching.

[0114] The device-level iris recognition terminal sends the second iris feature code generated in step S153 to the server via industrial Ethernet. The iris feature comparison module on the server performs a comparison operation again, calculating the Hamming distance between this code and all iris feature codes stored in the spatiotemporal correlation network to obtain a second similarity score set. The code with the smallest Hamming distance is selected from this set as the second benchmark feature code for successful matching.

[0115] Step S155: Perform a reverse index query in the spatiotemporal correlation network based on the second reference feature code to obtain a second identity identifier uniquely bound to the second reference feature code.

[0116] The server uses the second baseline feature code found in step S154 as the query condition, performs a reverse index query in the iris feature permission association table, and returns the second identity identifier bound to the code, namely the internal database primary key value of the maintenance personnel Zhang Ming, which should be the same value as the first identity identifier obtained in step S1211.

[0117] Step S156: Extract the corresponding second job role dynamic permission parameters and the second effective time window from the spatiotemporal association network based on the second identity identifier, and verify whether the current system time falls within the second effective time window.

[0118] Based on the second identity identifier obtained in step S155, the node record unit corresponding to the maintenance personnel Zhang Ming is located in the spatiotemporal correlation network, and his second job role dynamic permission parameters and second effective time window are extracted. The current system timestamp is obtained and compared with the start and end times of the second effective time window to verify whether the current time is within the effective window.

[0119] Step S157: Based on the target task identifier, query the list of roles that the target task is allowed to execute in the operation and maintenance task database, and compare the standard job role name contained in the second job role dynamic permission parameter with the list of roles that the task is allowed to execute.

[0120] Using the target task identifier "TASK_20231029_001" as the query condition, search the operation and maintenance task database for the record corresponding to the task, and extract the role list field that the task is allowed to execute. Parse this list into a set of role names. Compare the standard job role name "Relay Maintenance Specialist" contained in the dynamic permission parameters of the second job role with this set to determine whether the former exists in the latter.

[0121] Step S158: If the standard job role name contained in the dynamic permission parameters of the second job role exists in the role list that the task is allowed to execute, a role matching pass signal is generated; otherwise, a role matching failure signal is generated and the operation process is terminated.

[0122] If the comparison result matches, meaning "Relay Protection Maintenance Specialist" is in the list of roles allowed to execute the task, a Boolean signal "role_match" is generated and set to true, indicating that the role match is successful and the process continues. If the comparison result does not match, "role_match" is generated and set to false. The server immediately sends an operation termination command to the device-level iris recognition terminal. The terminal display shows the message "Insufficient permissions, unable to execute this task," and the entire operation process is terminated.

[0123] Step S159: Based on the target task identifier, query the standard operation procedure document of the target device in the device ledger database, and parse the sequence of operation steps and the range of operation parameter specifications for each step in the standard operation procedure document.

[0124] Using the target task identifier "TASK_20231029_001" and the associated target device identifier "EQP_MAIN_001" as the query criteria, the standard operating procedure document for that device is retrieved from the device ledger database. This standard operating procedure document is a structured file, which is parsed into memory to extract the sequence of operating steps. Each item contains a step number, step description, expected operating instruction string, and operating parameter specification ranges, such as the opening angle range and the maximum allowable current value.

[0125] Step S1510: The operation step sequence is presented on the display screen of the device-level iris recognition terminal in an interactive interface, waiting for the operation and maintenance personnel to input operation instructions and operation parameters one by one according to the operation step sequence.

[0126] The server transmits the parsed sequence of operation steps to the device-level iris recognition terminal via industrial Ethernet. The interactive interface on the terminal presents the sequence of steps in a list format, highlighting the current first step. The interface includes command input boxes and parameter input boxes, waiting for Zhang Ming to input the corresponding commands and parameters according to the requirements of the first step.

[0127] Step S1511: Receive the first step operation instruction and first step operation parameters input by the operation and maintenance personnel on the interactive interface; perform string matching between the first step operation instruction and the expected operation instruction of the corresponding step in the standard operation process document; and perform numerical range matching between the first step operation parameters and the operation parameter specification range of the corresponding step.

[0128] Zhang Ming inputs the first step's command, such as "open the circuit breaker," and parameters, such as the target angle "45 degrees," onto the terminal touchscreen. The terminal packages the input command string and parameter value and sends it to the server. The server precisely compares the received command string with the expected first step's command string in the standard operating procedure document, requiring complete consistency. Simultaneously, it matches the received parameter value with the specified range for the first step's operation, such as the angle range "40 degrees to 50 degrees," determining whether the input value falls within this closed range.

[0129] In step S1512, if the string matching is successful and the numerical range matching is successful, the operation permission instruction is generated and sent to the device controller connected to the target device. The device controller drives the operating mechanism of the target device to execute the action corresponding to the operation instruction in the first step according to the operation permission instruction.

[0130] If the string comparison in step S1511 is completely consistent and the parameter values ​​are within the specified range, then the operation verification for this step is deemed successful. The server generates an operation permission instruction message, which includes the target device identifier "EQP_MAIN_001" and the specific operation instruction code. This instruction is sent via industrial Ethernet to the device controller connected to main transformer No. 1. This controller is an industrial-grade programmable logic controller. After parsing the instruction, the device controller sends the corresponding level signal and pulse sequence to the operating mechanism of the main transformer through its output module, driving the operating mechanism to perform the tripping action.

[0131] Step S1513: During the operation of the target device's operating mechanism, the status sensor data of the target device is collected in real time. The status sensor data is dynamically compared with the expected status change curve of the corresponding step in the standard operation procedure document to generate an operation execution process deviation index.

[0132] During the tripping action of the operating mechanism, the equipment controller, through its analog and digital input modules, collects real-time data from various status sensors installed on the target equipment at a sampling frequency of 100Hz. This includes the current angle value fed back by the angle sensor, the drive motor current value measured by the current transformer, and the equipment vibration amplitude measured by the vibration sensor. The controller transmits this real-time data stream back to the server via industrial Ethernet. The server compares the received real-time angle change curve with the expected angle change curve for that step pre-stored in the standard operating procedure document, point by point. It calculates the absolute value of the difference between the actual angle value A_t and the expected angle value A_exp_t at each sampling time t, and calculates the average and maximum values ​​of all absolute differences over the entire execution cycle to obtain the operation execution process deviation index D_proc. This operation execution process deviation index is a dimensionless value that reflects the degree of deviation between the actual execution process and the standard procedure.

[0133] Step S1514: If the deviation index of the operation execution process exceeds the preset deviation threshold, an emergency stop command is immediately sent to the device controller to suspend the operation of the target device and an abnormal alarm message is displayed on the device-level iris recognition terminal.

[0134] The server compares the calculated deviation index D_proc with a preset deviation threshold D_th. If D_proc is greater than D_th, for example, if the actual angle curve deviates too much from the expected curve, it may indicate that the operating mechanism is jammed or there is another fault. The server immediately generates an emergency stop command message and sends it to the equipment controller via industrial Ethernet with the highest priority. Upon receiving the emergency stop command, the equipment controller immediately cuts off the power supply to the operating mechanism and activates the mechanical brake, pausing the equipment's operation. Simultaneously, the server pushes an abnormal alarm message to the equipment-level iris recognition terminal, and a red alarm window pops up on the terminal's display screen, showing "Operation execution abnormal, emergency stop" and specific deviation information.

[0135] Step S1515: If the deviation index of the operation execution process does not exceed the preset deviation threshold, after the action corresponding to the operation instruction of the first step is completed, the interactive interface is updated on the display screen of the device-level iris recognition terminal, the completed steps are marked as completed, and the next step is automatically highlighted, waiting for the operation and maintenance personnel to input the operation instructions and operation parameters of the next step.

[0136] During the execution of step S1513, if the real-time calculated deviation index D_proc never exceeds the preset deviation threshold D_th, and the operating mechanism positioning signal fed back by the device controller (e.g., the digital input signal triggered by the trip limit switch) becomes high, the server determines that the first step operation has been completed. The server sends a step completion confirmation message to the device-level iris recognition terminal. After receiving the message, the terminal changes the background color of the list item corresponding to the first step to green on the interactive interface and displays a "Completed" checkmark icon on its right. Subsequently, the interface automatically scrolls and highlights the second step operation, while clearing the instruction input box and parameter input box, waiting for Zhang Ming to input the operation instruction and operation parameters for the second step. For each subsequent step defined in the standard operation procedure document, the processing logic of steps S1511 to S1515 is repeated until the entire sequence of operation steps is completed.

[0137] Step S1516: After all the operation steps in the standard operation procedure document have been executed, an overall quality evaluation report for this operation is generated based on the deviation index of each step recorded in the entire operation process. The overall quality evaluation report is then associated with the second identity identifier and the target task identifier and stored in the operation and maintenance history database.

[0138] After the final step is completed and the device controller sends a confirmation signal, the server summarizes the deviation index D_proc_i recorded for each step during the operation, where i is the step number. The server calculates the average value D_avg of all step deviation indices and finds the maximum value D_max. Based on the values ​​of D_avg and D_max, and combined with preset evaluation rules (e.g., D_avg less than 0.1 and D_max less than 0.2 is "Excellent", D_avg less than 0.2 and D_max less than 0.3 is "Good", otherwise "Average"), a text-based overall quality evaluation label is generated. The server associates the overall quality evaluation report, operation start timestamp, operation end timestamp, deviation index sequence for each step, secondary identity identifier, target task identifier, and target device identifier into a structured record unit, and writes it into the "Operation Quality Record Table" in the operation and maintenance history database using a structured query language insertion statement.

[0139] In step S160, during the execution of the operation sequence by the operating mechanism of the target device, multiple status monitoring sensors deployed on the target device are used to collect the time series of the operating parameters of the target device in real time, and the time series of the operating parameters are analyzed and processed to generate derivative data related to the operation quality.

[0140] For example, in step S161, during the execution of the operation sequence by the operating mechanism of the target device, the operating parameters time series of the target device are collected in real time by multiple status monitoring sensors deployed on the target device. The operating parameter time series includes voltage waveform, current waveform, temperature change curve, mechanical vibration spectrum and displacement trajectory of the operating mechanism.

[0141] During the tripping operation of the No. 1 main transformer, the voltage transformer, current transformer, fiber optic temperature sensor, accelerometer, and displacement sensor deployed on it continuously operate. The voltage transformer outputs an analog signal from 0 to 100V, which is converted into a digital signal by the analog input module of the equipment controller at a sampling rate of 10kHz, forming a voltage waveform time series U(t). The current transformer outputs an analog signal from 0 to 5A, which is also converted into a current waveform time series I(t) at a sampling rate of 10kHz. The fiber optic temperature sensor outputs the winding temperature value at a frequency of 1Hz through a serial communication interface, forming a temperature change curve T_temp(t). The accelerometer outputs vibration acceleration values ​​in three axes at a sampling rate of 5kHz. The server performs a Fast Fourier Transform on every 1024 sampling points to obtain the mechanical vibration spectrum sequence Freq(t,f). The displacement sensor outputs the linear or angular displacement values ​​of the operating mechanism at a sampling rate of 100Hz, forming the operating mechanism displacement trajectory sequence S_pos(t). All of these time-series data are stamped with a uniform timestamp and are packaged in real time and sent to the server via Industrial Ethernet.

[0142] Step S162: Compare the time series of the operating parameters with the preset standard operating parameter curve in the standard operating procedure document point by point, calculate the difference between the operating parameter value and the standard value at each sampling time, and generate an operating parameter deviation time series diagram.

[0143] After receiving the real-time operating parameter time series, the server reads the preset standard operating parameter curves for the current operation step from the standard operating procedure document. For example, for a circuit breaker tripping operation, the standard operating procedure document presets a standard operating mechanism displacement trajectory curve S_std(t). The server aligns the real-time acquired displacement trajectory sequence S_pos(t) with S_std(t) on the time axis. Since there may be slight differences in the operation start time, the server first calculates the time delay between the two sequences using a cross-correlation function. After alignment, for each identical sampling time t_k, the server calculates the displacement deviation value ΔS_k, which is equal to S_pos(t_k) minus S_std(t_k). The same point-by-point subtraction operation is performed for parameters such as voltage and current. The deviation values ​​ΔS_k calculated for all sampling times are connected in chronological order to generate an operating parameter deviation time series diagram. This operating parameter deviation time series diagram is a two-dimensional data sequence, with time on the horizontal axis and deviation value on the vertical axis.

[0144] Step S163: Associate and store the operation parameter deviation timing diagram with the second identity identifier, the target task identifier, and the current operation step identifier to generate a fine-grained operation record with an operation quality label.

[0145] The server associates the timing diagram of the operation parameter deviation for each operation step generated in step S162 with the second identity identifier obtained in step S155, the target task identifier obtained in step S141, and the step identifier of the current operation step (e.g., "STEP_01"). This data is then combined into a fine-grained operation record unit, which contains deviation information for each instant during the operation. The server uses a structured query language insertion statement to write this fine-grained operation record into the "Fine-Grained Operation Data Table" in the operation and maintenance history database.

[0146] Step S164: After completing all operation steps, calculate the overall quality score of the entire operation process based on the operation parameter deviation time sequence diagram, and display the overall quality score on the device-level iris recognition terminal.

[0147] After all operation steps are completed, the server iterates through the deviation timing diagrams generated for each operation parameter at each step. For each deviation timing diagram, the average of the absolute values ​​of all deviations is calculated to obtain the average deviation of that parameter for that step. Then, a weighted sum is performed on the average deviations of all parameters across all steps. The weights are pre-set based on the importance of each parameter to the safe operation of the equipment; for example, displacement deviation has a weight of 0.5, current deviation has a weight of 0.3, and temperature deviation has a weight of 0.2. The result of the weighted sum is a dimensionless comprehensive quality score Q_total. The smaller the Q_total value, the higher the operation quality. The server sends this comprehensive quality score Q_total to the device-level iris recognition terminal, which displays the comprehensive quality score of this operation on the operation completion interface in the form of a prominent number and a progress bar.

[0148] Step S165: The fine-grained operation record and the comprehensive quality score are encapsulated into an operation quality report, which is sent to the server via industrial Ethernet, and the server stores the operation quality report in the operation and maintenance history database.

[0149] The server packages all the fine-grained operation records generated in step S163 and the comprehensive quality score calculated in step S164 into an operation quality report file. The file format adopts the Extensible Markup Language (XML) format commonly used in the power industry. This operation quality report contains comprehensive information, from personnel identity, tasks performed, minor deviations in the operation process to the final score. The server stores this report file in a designated storage area of ​​the operation and maintenance history database via industrial Ethernet and creates an index for subsequent rapid retrieval.

[0150] Step S166: Based on the deviation information in the operation quality report, trigger the operation and maintenance knowledge base update operation, and add the operation steps, deviation types and handling measures that have deviations as new cases to the operation and maintenance knowledge base.

[0151] The operations and maintenance knowledge base management module on the server parses the operation quality report generated in step S165. If the time sequence diagram of the operation parameter deviation in the report shows areas that continuously exceed a preset threshold, such as a displacement deviation that is greater than 5 mm for more than 1 second, the module determines that an "execution deviation" has occurred in that step. The module extracts the operation step identifier, the type of deviation parameter (e.g., "displacement"), the magnitude of the deviation, and the corresponding standard operating procedure for that step. Then, the module automatically generates a case record for this deviation event. The case record includes a description of the deviation, a possible cause analysis (based on preset rules, such as displacement deviation possibly being related to insufficient lubrication of the mechanism), and suggested handling measures (e.g., "check the operating mechanism and add lubricant"). This case is added to the operations and maintenance knowledge base for auxiliary diagnosis of similar problems in the future and for training operations and maintenance personnel.

[0152] Step S167: Perform multi-source data fusion on the first iris image, the second iris image, the first identity identifier, the second identity identifier, the target task identifier, the operation parameter deviation time sequence diagram, and the comprehensive quality score to generate a complete operation and maintenance log record that conforms to the standard format of the power industry.

[0153] The operation and maintenance log generation module on the server retrieves data from various data sources: the first iris image generated in step S128 and the second iris image generated in step S152 from the temporary storage area; the first identity identifier generated in step S1211 and the second identity identifier generated in step S155 from memory variables; the target task identifier obtained in step S141 from the task processing flow; the operation parameter deviation time series diagram from the database generated in step S163; and the comprehensive quality score from the calculation results in step S164. The module organizes and encodes the above data according to the log format defined in the power industry standard "DL / T1664-2016 Power Equipment Operation and Maintenance Regulations" to generate a structured, complete operation and maintenance log record containing text, images, and numerical data.

[0154] Step S168: Encrypt the complete operation and maintenance log record using an encryption algorithm to generate an encrypted log data packet, and simultaneously store the encrypted log data packet in a local dual-machine hot standby server and a remote cloud database.

[0155] The server invokes the encryption module, employing the AES-256 symmetric encryption algorithm and a pre-generated key to encrypt the complete operation and maintenance log records generated in step S167. The encryption process converts the original data into ciphertext data, generating an encrypted log data packet. The server writes this encrypted data packet to the disk array of the local dual-machine hot standby server via its internal high-speed bus. Simultaneously, the server establishes a secure transmission channel with a remote cloud database through a dedicated encrypted 5G virtual private network, transmitting the same encrypted data packet to the cloud for off-site backup.

[0156] Step S169: After the encrypted log data packet is successfully stored, a storage success receipt is generated and sent to the monitoring center for confirmation and archiving by the monitoring center administrator.

[0157] After both the local dual-machine hot standby server and the remote cloud database return confirmation messages of successful write, the server generates a storage success receipt message. This storage success receipt message includes a unique identifier for this operation, a storage timestamp, storage location information, and a data integrity check value. The server sends this receipt to the large-screen display system in the monitoring center and the mobile terminal application of the monitoring administrator. After confirming that all data is correct, the monitoring center administrator clicks the confirm archive button to complete the final archiving process for this operation and maintenance operation.

[0158] Step S170: While sending the operation permission instruction to the device controller, start an independent operation monitoring thread to perform parallel monitoring and auxiliary analysis of the operation execution process.

[0159] For example, in step S171, while sending the operation permission instruction to the device controller, an independent operation monitoring thread is started. The operation monitoring thread polls the execution status register of the device controller at a fixed frequency to obtain the real-time operation progress of the target device.

[0160] In step S1512, while the server sends an operation permission command to the device controller, it simultaneously creates a separate lightweight thread at the server operating system level specifically for monitoring this operation. This thread sends a request message to read the status register to the device controller via industrial Ethernet at fixed time intervals of 100 milliseconds. The execution status register inside the device controller reflects the progress of the current operation in real time; for example, 0x00 indicates idle, 0x01 indicates execution of the first step, 0x02 indicates the first step completed and waiting for the next step, and 0x11 indicates an execution error. The device controller responds to the request by returning the value of the status register to the monitoring thread on the server. The monitoring thread continuously records these status values ​​and their timestamps.

[0161] Step S172: The real-time operation progress is displayed on the display screen of the device-level iris recognition terminal in the form of a graphical progress bar. The graphical progress bar includes a completed step identifier, a currently executing step identifier, and a pending step identifier.

[0162] The monitoring thread parses the status values ​​obtained from each polling into operation progress information. For example, a status value of 0x01 indicates that the first step is being executed. The monitoring thread pushes this progress information to the device-level iris recognition terminal in real time. The user interface on the terminal dynamically renders a graphical progress bar based on the received information. The progress bar is divided into multiple segments, each corresponding to an operation step. Segments corresponding to completed steps are displayed in green, segments corresponding to currently executing steps are displayed in yellow with a flashing animation, and segments corresponding to steps yet to be executed are displayed in gray. Text prompts are also displayed above the progress bar, such as "Step 1 of 5, performing the circuit breaker opening operation."

[0163] Step S173: During the operation of the target device, the audio signal of the operation site is collected by the microphone built into the device-level iris recognition terminal, and the audio signal is subjected to spectrum analysis to extract abnormal sound features during the operation.

[0164] During operation, the built-in microphone of the device-level iris recognition terminal continuously operates, acquiring audio signals from the scene at a sampling rate of 16kHz. The digital signal processor on the terminal processes the audio stream in real time. It divides the audio stream into data blocks of 50 milliseconds each, applies a Fast Fourier Transform to each frame to convert the time-domain signal into a frequency-domain signal, obtaining a spectrogram. The processor compares the real-time spectrogram with the sample spectral characteristics in a pre-stored "abnormal sound library." The abnormal sound library pre-stores spectral templates for various abnormal device sounds, such as metallic friction sounds, discharge sounds, and mechanical jamming sounds. By calculating the Euclidean distance or correlation coefficient between the real-time spectrum and the template spectrum, the presence of abnormal sounds is identified.

[0165] Step S174: Match the abnormal sound features with the sample features in the pre-stored abnormal sound library. If the match is successful, it is determined that there is an abnormal sound at the operation site, the operation is immediately suspended, and an on-site abnormal alarm is sent to the monitoring center.

[0166] If the matching result in step S173 shows that the similarity between the real-time spectrum and a certain abnormal sound template exceeds a preset threshold (e.g., a correlation coefficient greater than 0.85), the processor on the terminal determines that there is an abnormal sound at the operation site. The terminal immediately sends a high-priority abnormal alarm message to the server via the industrial Ethernet, and simultaneously sends a local emergency stop signal to the equipment controller (directly connected to the emergency stop input pin of the controller through the terminal's general input / output interface). After receiving the alarm, the server immediately pushes an alarm message containing "abnormal sound at the operation site, emergency stop" and an audio clip to the monitoring center. After receiving the emergency stop signal, the equipment controller immediately cuts off the equipment's power supply.

[0167] Step S175: The video stream of the operation site is captured by the built-in camera of the device-level iris recognition terminal. The video stream is analyzed in real time to detect whether the operation gestures of the maintenance personnel are consistent with the operation gestures specified in the standard operation procedure document.

[0168] During operation, the wide-angle camera built into the device-level iris recognition terminal captures a wide-angle video stream of the operation site at a frame rate of 30fps. The neural network processing unit on the terminal loads a pre-trained pose estimation model (such as a lightweight version of OpenPose). For each frame of video image, the model first detects the human body in the image, and then predicts the two-dimensional pixel coordinates of the main skeletal key points of the human body (such as the shoulder, elbow, and wrist). By analyzing the relative positional relationship between the wrist key points and specific button or knob areas on the target device's control panel, as well as the arm movement trajectory in consecutive frames, the operation gesture of the maintenance personnel can be determined, such as whether it is "pressing" or "rotating".

[0169] Step S176: If an operation gesture is detected to be inconsistent with the prescribed gesture, a prompt window will pop up on the display screen of the device-level iris recognition terminal, requiring the maintenance personnel to correct the operation gesture and record the inconsistent gesture in the operation log.

[0170] The terminal compares the operation gesture identified in step S175 with the operation gesture specified for the current step in the standard operation procedure document. For example, if the current step specifies "rotate the knob 90 degrees clockwise," but the identified gesture is "press," the terminal immediately displays a semi-transparent prompt window in the center of the screen, showing the text "Operation gesture error, please follow the prompts to rotate" and a dynamic illustration. Simultaneously, the terminal records this gesture inconsistency event, including the timestamp, current step identifier, and the type of incorrect gesture identified, in its local operation log cache, and finally reports it to the server, storing it in the operation and maintenance history database.

[0171] Step S177: After the operation is completed, the key segments corresponding to the operation steps in the video stream are extracted to generate a short video of the operation process, and the short video of the operation process is associated with and stored in the complete operation and maintenance log record.

[0172] After all operation steps are completed, the server sends instructions to the device-level iris recognition terminal based on the start and end timestamps of the operation recorded in step S171, as well as the start and end timestamps of each step, requesting the terminal to return video stream segments for the corresponding time period. The terminal extracts the video data of these key segments from its local cache, compresses and encodes them, and generates several short MP4 format video files, such as "Step 1 Switch Opening Operation.mp4". The server associates these short video files with the complete operation and maintenance log record generated in step S167, using the file's storage path as a field in the log record, thus achieving associated storage of video data and text log data.

[0173] Step S178: Based on the operation posture of the maintenance personnel in the short video of the operation process, the operation posture features are extracted using the human skeleton key point detection algorithm. The operation posture features are compared with the postures in the standard operation posture library to generate an operation standardization evaluation score.

[0174] The offline analysis module on the server performs in-depth processing on the short video of the operation process captured in step S177. For each frame of the video, a high-precision human skeleton keypoint detection algorithm is called again to extract the precise three-dimensional spatial coordinates of all joints of the upper limbs of the maintenance personnel (through multi-view geometry or monocular depth estimation technology). The joint coordinate sequence of the above consecutive frames is combined to form an operation posture feature vector, which describes the changes in the body posture of the personnel throughout the operation. The server performs dynamic time warping comparison on this feature vector with the "standard posture" feature vectors pre-stored in the standard operation posture library and demonstrated and annotated by senior experts, and calculates the similarity between the two sequences. The higher the similarity score, the more standardized the operation posture. Finally, an operation standardization evaluation score S_std between 0 and 100 is generated.

[0175] Step S179: The operation standardization assessment score and the comprehensive quality score are weighted and integrated to generate a comprehensive capability evaluation index for the operation and maintenance personnel's operation, and the comprehensive capability evaluation index is stored in the operation and maintenance personnel's personal skill file.

[0176] The server merges the operational standardization assessment score S_std generated in step S178 with the comprehensive quality score Q_total calculated in step S164. Since the two indicators have different dimensions, Q_total is first normalized, mapping it to the range of 0 to 100 to obtain Q_norm, where Q_norm equals 100 minus Q_total (because a smaller Q_total indicates higher quality). Then, a weighted average method is used to calculate the comprehensive capability evaluation index E_total, which is equal to alpha multiplied by S_std plus (1 minus alpha) multiplied by Q_norm, where alpha is a preset weighting coefficient, for example, 0.6, indicating a greater emphasis on operational standardization. The calculated E_total value reflects the overall performance of maintenance personnel Zhang Ming in this operation. The server appends this E_total value, along with information such as the date, task type, and equipment type of this operation, to the maintenance personnel's personal skill profile database, which records the evaluation results of the personnel's past operations for tracking and evaluating their skill development.

[0177] For example, the method may also include:

[0178] In step S180, while the device-level iris recognition terminal generates an operation permission instruction and sends it to the device controller, the operation process mirror reconstruction engine deployed on the server is started to perform in-depth reconstruction and analysis of the operation process.

[0179] Step S181: While the device-level iris recognition terminal generates an operation permission instruction and sends it to the device controller, the operation process image reconstruction engine deployed on the server is started. The operation process image reconstruction engine receives in real time the operation video stream collected by the device-level iris recognition terminal and the device response data stream returned by the device controller.

[0180] In step S1512, while the server sends the operation permission instruction, it also starts a dedicated operation process mirror reconstruction engine service, which establishes connections through two independent data channels: one channel receives the operation video stream (compressed and encoded in H.264 format) transmitted in real time from the device-level iris recognition terminal; the other channel receives the device response data stream transmitted in real time from the device controller, which contains high-frequency data such as device status register values ​​and real-time sensor sampling values.

[0181] Step S182: The operation video stream is decomposed frame by frame according to the time axis to generate an operation video frame sequence with timestamps, and the device response data stream is sampled and aligned according to the same time axis to generate a device response data point sequence corresponding to each operation video frame in time.

[0182] The mirror reconstruction engine first decodes the received operational video stream, restoring it into frame-by-frame images. Each frame is precisely labeled with a system timestamp indicating the moment of reception, generating an operational video frame sequence F_frames, where each element is a tuple (timestamp_f, image_data). Simultaneously, the received device response data stream is also processed according to its own sampling timestamp, generating a device response data point sequence D_points, where each element is a tuple (timestamp_d, device_status, sensor_values). Because the video frame rate of 30fps and the data sampling rate of 100Hz are inconsistent, the engine uses nearest-neighbor interpolation for time axis alignment. For each video frame's timestamp_f, the engine finds the closest data points in D_points and calculates the device response data point that precisely corresponds to that video frame in time through interpolation.

[0183] Step S183: Perform human posture skeletal key point detection processing on each frame of the operation video frame sequence, extract the three-dimensional spatial coordinates of the upper limb joints of the operation and maintenance personnel, and generate the arm movement trajectory curve and hand operation posture change sequence of the operator.

[0184] For each operation video frame (image_data) generated in step S182, the mirror reconstruction engine calls a high-precision human pose estimation model. This model first detects the bounding box of the maintenance personnel in the image, and then predicts the two-dimensional coordinates of upper limb joints, including shoulders, elbows, wrists, and palms, within that region. Combining pre-calibrated camera intrinsics and scene geometry information, the two-dimensional coordinates are converted into device-centric three-dimensional spatial coordinates (x, y, z) using triangulation or monocular depth estimation methods. For each frame, a set of three-dimensional joint coordinates is obtained. The three-dimensional coordinates of the right wrist joint in all frames are connected sequentially over time to form a three-dimensional arm movement trajectory curve. Simultaneously, by analyzing the relative positional changes of the palm and wrist joints between consecutive frames, the hand's operational posture, such as "gripping," "pinching," or "rotating," can be inferred, forming a time-varying sequence of operational postures.

[0185] Step S184: Perform numerical change rate analysis on each device response data point in the device response data point sequence, calculate the response delay time, action execution speed and action positioning accuracy of the device operating mechanism after receiving the operation permission instruction, and generate a set of device execution characteristic parameters.

[0186] The mirror reconstruction engine analyzes the device response data point sequence generated in step S182. First, it finds the timestamp t_cmd for sending the operation permission command. Then, in the device response data point sequence, it finds the first time point t_resp where the operating mechanism status register changes from "idle" to "execute," and calculates the response delay time T_delay, which is equal to t_resp minus t_cmd. Next, it analyzes the displacement sensor data, from the start of movement to the target position, and calculates the average speed V_avg of the action execution, which is equal to the total displacement minus the total time. Finally, it analyzes the steady-state value of the sensor data after reaching the target position, compares it with the target value, and calculates the positioning accuracy P_acc, which is equal to the difference between the actual steady-state value and the target value. The parameters T_delay, V_avg, and P_acc are combined to form a parameter set describing the characteristics of this device execution.

[0187] Step S185: The operator's arm movement trajectory curve and the set of equipment execution characteristic parameters are superimposed and fused on the time axis to construct a mirror map of the collaborative operation process with time as the horizontal axis and personnel action and equipment response as the dual vertical axes.

[0188] The mirror reconstruction engine creates a new data visualization structure, namely the collaborative operation process mirror map, with a unified time axis as the horizontal axis. On the first vertical axis, a certain dimension of the personnel arm movement trajectory curve generated in step S183 is plotted, such as the curve of the wrist point's vertical coordinate changing over time. On the second vertical axis, the equipment execution characteristics calculated in step S184 are plotted, such as the curve of the real-time feedback value of the displacement sensor changing over time. By overlaying these two curves on the same time axis, the temporal relationship and interaction pattern between personnel actions and equipment responses can be intuitively observed.

[0189] Step S186: Mark the time point when the operation permission command is issued, the time point when the equipment starts to respond, the key turning point of the personnel operation action, and the time point when the equipment action is completed on the mirror map of the collaborative operation process, and generate a sequence of key nodes of the operation collaboration timing.

[0190] The mirror reconstruction engine automatically marks several key nodes on the graph. First, it marks t_cmd (command issuance point) and t_resp (device response point). Then, it performs second-order derivative analysis on the personnel's arm movement trajectory curve to identify the point with the largest curvature change, i.e., the point where the velocity or acceleration direction changes abruptly, serving as a key turning point in the personnel's operation, such as the point where the hand begins to rotate or stops moving. Finally, it marks the time point t_complete when the equipment feedback data reaches a steady-state value, as the equipment action completion point. Arranging these nodes in chronological order forms a sequence of key nodes for operational coordination, recording the crucial timing information of the human-equipment collaborative work.

[0191] Step S187: Calculate the time difference sequence between personnel operation actions and equipment response actions based on the key node sequence of operation coordination timing, and perform statistical analysis on the time difference sequence to generate an operation coordination efficiency index between personnel and equipment.

[0192] The mirror reconstruction engine calculates the time difference between the turning point of a human action and the subsequent change in the device response within a key node sequence. For example, it calculates the difference Δt between the moment the human's hand begins to rotate (t_human_turn) and the moment the device displacement curve begins to change (t_device_move). Time differences are calculated for all similar, paired human-device action nodes during the operation, forming a time difference sequence. The mean and standard deviation of this sequence are calculated; a smaller mean indicates better coordination and faster response between the human and device, while a smaller standard deviation indicates higher stability of the coordination. The reciprocal of the mean and the reciprocal of the standard deviation are weighted and combined to generate a dimensionless operational coordination efficiency index, E_coop; a larger E_coop value indicates higher coordination efficiency.

[0193] Step S188: The collaborative operation process mirror map, the operational collaboration time sequence of key nodes, and the operational collaboration optimization prompt information are encapsulated and processed to generate an operational process mirror analysis report, and the operational process mirror analysis report is pushed to the monitoring center's large screen for display.

[0194] The mirror reconstruction engine encapsulates the graph generated in step S185, the node sequence generated in step S186, and the optimization prompts generated based on the analysis results of step S187 (such as "It is recommended to pause briefly after issuing the command and wait for the device to fully respond before proceeding to the next step") to generate a graphical and textual operation process mirror analysis report in PDF format. The engine pushes this report file to the large-screen display system in the monitoring center via industrial Ethernet. Monitoring center managers can zoom in to view the graph details on the large screen and analyze the collaborative process of this operation.

[0195] Step S189: The operation process mirror analysis report is associated and stored with the second identity identifier and the target task identifier as case material for subsequent operation training, and a trend chart of the change in the operation and collaboration capabilities of the operation and maintenance personnel is generated after multiple operation data accumulations.

[0196] The server associates the image analysis report generated in step S188 with the second identity identifier obtained in step S155 and the target task identifier obtained in step S141, and stores them in the "Training Case Library" of the operation and maintenance history database. Over time, Zhang Ming performed many different operation tasks. The server periodically (e.g., monthly) extracts the collaboration efficiency index E_coop of all of Zhang Ming's historical operations from the database, plots the operation date on the horizontal axis and the E_coop value on the vertical axis, and draws a trend chart of Zhang Ming's personal operation collaboration ability. This chart can show the progress of his skills and can be anonymously compared with the data of other employees in the same position for targeted skills training.

[0197] For example, the method may also include:

[0198] Step S190: After the target device completes the operation sequence, the operation semantic reconstruction engine deployed on the server is triggered to perform in-depth semantic analysis and knowledge extraction on this operation.

[0199] Step S191: After the target device completes the operation sequence, the operation semantic reconstruction engine deployed on the server is triggered. The operation semantic reconstruction engine extracts all the original record data of this operation from the operation and maintenance history database.

[0200] After the semantic reconstruction engine is automatically activated, it initiates a joint query to the operation and maintenance history database using the combination of the target task identifier "TASK_20231029_001" and the operation start timestamp "2024-05-20-10-15-30" as the query key. The query involves the "Iris Recognition Record Table", "Task Execution Record Table", "Fine-grained Operation Data Table", "Operation Quality Evaluation Table", and "Mirror Analysis Report Table". The engine extracts the storage path of the first iris image, the second iris image, the first identity identifier, the second identity identifier, the target task identifier, the operation parameter deviation time series data sequence, and the operation process mirror analysis report file from these tables, and loads the above raw data into memory for subsequent analysis.

[0201] Step S192: Perform depth feature re-extraction processing on the first iris image and the second iris image in the original record data to generate a highly robust iris feature condensation code for traceability, and bind the highly robust iris feature condensation code with the first identity identifier and the second identity identifier to an immutable traceability-dedicated storage area.

[0202] A pre-trained ResNet-152 deep convolutional neural network model is loaded, and the first and second iris images are scaled to 256*256 pixels before being input into the model. After 152 layers of convolution, pooling, and fully connected computation, a 2048-dimensional floating-point feature vector is extracted from the last fully connected layer, which summarizes the deep texture features of the iris. This vector is compressed into a 512-byte highly robust iris feature condensation code through principal component analysis and product quantization. Then, the condensation code is bound to the first identity identifier, the second identity identifier, and the operation timestamp, and written to a dedicated traceability storage area based on distributed ledger technology via an application programming interface. This dedicated traceability storage area uses digital signatures to ensure that the data is tamper-proof.

[0203] Step S193: Perform waveform feature recognition processing on the operation parameter deviation time sequence diagram, extract the number of peaks, depth of valleys, fluctuation frequency and convergence speed in the operation parameter deviation curve, and generate operation quality waveform feature vector.

[0204] Taking the displacement deviation time series as an example, the engine applies a five-point cubic smoothing filter to the deviation value sequence, then uses first-order derivative zero-crossing detection to determine all peaks and troughs, counts the number of peaks N_peak, and calculates the average trough depth D_valley_mean. A fast Fourier transform is performed on the smoothed sequence, and the frequency with the largest amplitude is extracted from the spectrum as the fluctuation frequency F_dom. The time from the start of the action to the moment when the deviation value first enters the ±0.5 mm steady-state range and no longer exceeds it is calculated as the convergence speed T_settle. N_peak, D_valley_mean, F_dom, and T_settle are combined into a four-dimensional vector. The above process is repeated for parameters such as current and vibration. Finally, the waveform feature vectors of all parameters are concatenated into a comprehensive operational quality waveform feature vector V_waveform_total.

[0205] Step S194: Input the operation quality waveform feature vector into the pre-trained operation skill level evaluation model, and the operation skill level evaluation model outputs the operation skill level label and operation skill defect type label corresponding to this operation.

[0206] The `V_waveform_total` is input into a multi-layer gradient boosting decision tree model based on XGBoost. This model is trained using a large amount of expert-annotated historical operation waveform data. The annotation dimensions include four skill levels: "beginner," "intermediate," "advanced," and "expert," as well as sixteen defect types, such as "excessive start-up impact," "overshoot at position," "undershoot at position," and "delayed action response." The decision trees in the model perform binary tree judgments based on the values ​​of each dimension of the vector. The prediction results from all trees are weighted and summed to output the probability values ​​for the four skill levels and the probability values ​​for the sixteen defect types. The skill level with the highest probability is taken as the skill level label, and defect types with a probability value greater than 0.5 are filtered out to form a defect type label set.

[0207] Step S195: Combine the operation skill level label and operation skill defect type label with the equipment type information carried in the target task identifier to generate updated operation skill profile data for the operation and maintenance personnel of that equipment type.

[0208] The skill level label "Intermediate" and the defect type label set (e.g., "overshoot in position," "action response delay") output in step S194 are combined with the equipment type "Main Transformer" associated with the target task identifier into a four-tuple data structure. This four-tuple data structure contains a second identity identifier, an equipment type string, a skill level label, and a defect type label set. This four-tuple serves as incremental update data for the operation skill profile of maintenance personnel Zhang Ming for the main transformer equipment type and is stored in the maintenance personnel's personal skill file database for subsequent skill assessments and targeted training.

[0209] Step S196: Based on the operational skill defect type tag, perform a matching search in the preset training resource library to extract standardized training course videos and simulation practice scenario configuration files corresponding to the operational skill defect type.

[0210] Using each tag in the defect type tag set, such as "overshoot" and "action response delay," as keywords, the corresponding defect type code, such as "FLT-008" and "FLT-012," is searched in the "Defect Type Code Table" of the training resource library. These codes are then used to search the "Training Resource Table" to obtain matching training resource records. The search results return two resources: an HTTP link to the training course video on "High-Voltage Circuit Breaker Opening and Closing Accuracy Control Techniques," and a download address for the JSON format configuration file of the simulation exercise scenario on "Relay Protection Action Response Time Testing and Adjustment."

[0211] Step S197: Push the link address of the standardized training course video and the simulation exercise scenario configuration file to the personal mobile terminal of the maintenance personnel, and remind the maintenance personnel through the main iris recognition terminal the next time they enter the substation.

[0212] The server's push notification service sends the training course link and simulation configuration file download address to Zhang Ming's personal mobile terminal application via an encrypted channel. Simultaneously, the "has_pending_training" field is set to true in Zhang Ming's personal operations profile. When Zhang Ming next completes iris recognition at the main entrance / exit, the server adds a "reminder flag" field to true in the response message returning the first authorized task list. After parsing by the main iris recognition terminal, a yellow notification bar appears above the task list on the display screen: "You have a new targeted training course; please view it on the learning terminal."

[0213] Step S198: Semantically encapsulate all the original recorded data of this operation, the operation quality waveform feature vector, the operation skill level label, and the operation skill defect type label to generate an operation process summary text described in natural language and an operation quality radar chart displayed in a graphical manner.

[0214] The natural language generation module is invoked to fill in the personnel name "Zhang Ming", operation time "May 20, 2024, 10:15 AM", equipment type "main transformer", operation type "shutdown", skill level "intermediate", and defect type "overshoot at position, action response delay" based on a preset template to generate summary text. Simultaneously, six indicators—start-up impact, stable operation, positioning accuracy, response time, power matching, and vibration suppression—are normalized to 0 to 100, and a hexagonal radar chart is drawn using a graphics rendering library. The summary text and radar chart are then encapsulated in HTML format to generate a semantic summary page of the operation.

[0215] Step S199: The operation process summary text and operation quality radar chart are compared horizontally with the historical operation records of the same equipment in the substation to generate a heat map of the operation quality distribution of the equipment under different periods and different operators.

[0216] Using the device identifier "EQP_MAIN_001", the comprehensive waveform feature vector of 1500 historical operations over the past three years is retrieved. The high-dimensional data is then reduced to a two-dimensional plane using a t-distributed random neighborhood embedding algorithm, with each point representing one operation. Points are colored according to their overall quality score, with higher quality points leaning towards blue and lower quality points towards red. The two-dimensional plane is divided into a grid, and the number of points within each grid is counted and filled with color depth to generate a heatmap of operation quality distribution, visually displaying quality clusters and outliers in historical operations.

[0217] Step S1910: Mark the quality location and deviation characteristics of this operation on the operation quality distribution heat map, store them in the equipment ledger database as dynamic update content of the equipment operation and maintenance file, and display them as risk warning information when the next planned operation and maintenance task is generated.

[0218] Locate the corresponding 2D coordinates of this operation on the heatmap, mark it with a red pentagram, and pop up a text box to indicate "Operation: 2024-05-20, Skill Level: Intermediate, Deviation Characteristics: Overshoot, Action Response Delay". Convert the annotated heatmap to PNG format and write it to the "Operation and Maintenance History Analysis" field of the "EQP_MAIN_001" record in the equipment ledger database using an update statement. When the production management system generates the next planned operation and maintenance task for this equipment, read the image from this field and display it in the "Risk Warning" area of ​​the task details page, along with the text: "Note: The previous batch of operations had risks of overshoot and action response delay. It is recommended to check the mechanism lubrication and control loop parameters."

[0219] In one exemplary embodiment, an iris recognition-based security authentication and operation management system for power substation maintenance personnel is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the iris recognition-based power substation maintenance personnel safety authentication and operation management system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements an iris recognition-based power substation maintenance personnel safety authentication and operation management method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the casing of the power substation operation and maintenance personnel safety authentication and operation management system based on iris recognition, or an external keyboard, touchpad, or mouse, etc.

[0220] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for power substation operation and maintenance personnel safety authentication and operation management based on iris recognition, characterized in that, The method comprises: Pre-constructing a space-time association network of operation and maintenance personnel iris features and dynamic operation permissions; Collecting a first iris image of the operation and maintenance personnel through a main iris recognition terminal at the main entrance of the power substation, matching the first iris image with the iris feature codes in the space-time association network, generating a first identity tag with a matching success, and extracting corresponding first post role dynamic permission parameters and a first valid time window from the space-time association network according to the first identity tag; Extracting a set of tasks to be executed that overlap with the first valid time window from an operation and maintenance task database according to the first identity tag and the first valid time window, filtering the set of tasks to be executed based on the first post role dynamic permission parameters, generating a first authorized task list, and pushing the first authorized task list to the main iris recognition terminal for display; Receiving a target task tag selected by the operation and maintenance personnel in the first authorized task list, querying a corresponding target device tag and a standard operation procedure document of the target device from a device account database according to the target task tag, and sending the target device tag and the standard operation procedure document to a device-level iris recognition terminal deployed beside the target device; Collecting a second iris image of the operation and maintenance personnel through the device-level iris recognition terminal, matching the second iris image with the iris feature codes in the space-time association network, generating a second identity tag with a matching success, performing joint permission verification according to the second identity tag and the target task tag, generating an operation permission instruction, and sending the operation permission instruction to a device controller connected to the target device to drive the target device to execute an operation sequence corresponding to the standard operation procedure document.

2. The power substation operation personnel safety authentication and operation management method based on iris recognition according to claim 1, characterized in that, The space-time association network contains iris feature codes of operation and maintenance personnel, post role dynamic permission parameters bound to the iris feature codes, and valid time windows and spatial location constraints of the post role dynamic permission parameters on the time axis of the power substation for each key device, and the pre-constructed space-time association network of operation and maintenance personnel iris features and dynamic operation permissions specifically comprises: Receiving an input instruction of operation and maintenance personnel basic identity information input by a substation safety manager through a security management terminal, creating an operation and maintenance personnel basic information record unit in a local database according to the input instruction, and the operation and maintenance personnel basic information record unit contains an operation and maintenance personnel name full name text field, an operation and maintenance personnel work number code number field, an operation and maintenance personnel department name text field, and an operation and maintenance personnel standard post role name text field in the power substation; Calling a special iris collection device in communication connection with the security management terminal to perform continuous multiple iris image shooting processing on the eyes of the operation and maintenance personnel, obtaining a sequence of original iris images covering all texture areas of the eyes, and the sequence of original iris images contains left eye iris image frames and right eye iris image frames; The image sharpness index calculation process is performed on each frame of the original iris image in the original iris image pair sequence, the image gradient amplitude distribution feature and the image frequency spectrum high-frequency component proportion feature are extracted, the image quality score parameter is generated, the blurred iris image frame with the image quality score parameter lower than the preset quality threshold is removed from the original iris image pair sequence, and the effective iris image frame sequence meeting the feature extraction requirement is reserved; The iris region positioning process is performed on each frame of the effective iris image in the effective iris image frame sequence, the pupil boundary circular parameter and the iris outer boundary circular parameter are fitted out through the iris inner edge detection algorithm and the iris outer edge detection algorithm respectively, and the radius range and the center coordinate of the iris annular region are calculated according to the pupil boundary circular parameter and the iris outer boundary circular parameter; The iris annular region image is intercepted from the effective iris image according to the radius range and the center coordinate, the image normalization process is performed on the iris annular region image, the iris annular region image is unfolded from the polar coordinate space to the fixed-size rectangular normalized iris image, and the iris texture unfolding graph with uniform geometric size is obtained; The iris texture feature point extraction algorithm is used to perform multi-scale two-dimensional Gabor filter convolution processing on the iris texture unfolding graph, the filter response amplitude and phase information of each pixel point in the iris texture unfolding graph in multiple directions and multiple frequencies are extracted, and the original iris texture feature point distribution graph is generated; The feature point screening process is performed on the original iris texture feature point distribution graph, the noise feature points with the response amplitude lower than the preset response threshold are removed, the screened core texture feature points are obtained, and the feature descriptor vector of the core texture feature point is generated according to the coordinate position and the corresponding phase information of the core texture feature point in the iris texture unfolding graph; The feature descriptor vectors of all the core texture feature points corresponding to the same operation and maintenance personnel are spliced and combined in a preset order to generate a unique binary format operation and maintenance personnel iris feature code of the operation and maintenance personnel; The operation and maintenance personnel iris feature code is encrypted by using a secure hash algorithm to generate an encrypted iris feature ciphertext string, and the encrypted iris feature ciphertext string is stored in association with a standard post role name text field in the operation and maintenance personnel basic information record unit in a dual-hardware hot backup system server; The function area division electronic map and the responsibility range description document corresponding to the standard post role name text field in the power substation are generated, the access permission state of the standard post role name at each function area entrance is analyzed, and the spatio-temporal attribute association binding of the permission parameter is performed in combination with the valid time window corresponding to the standard post role name. 3.The method of claim 2, wherein, The analysis of the access permission state of the standard post role name at each function area entrance and the spatio-temporal attribute association binding of the permission parameter in combination with the valid time window corresponding to the standard post role name, comprises: The access permission status that the standard job role name should have at the entrance of each functional area is parsed out. The access permission status includes two binary status identifiers: allowed entry and prohibited entry. Extract the functional area name text and area code number corresponding to the allowed binary status identifier to generate a list of area codes that the standard job role name can pass through. Sort each area in the area code list from high to low according to the area security level value to obtain the spatial access priority sorting sequence in the job role dynamic permission parameters. Obtain the unique identification code of each key equipment in the power substation, as well as the planned operation and maintenance time window and temporary operation and maintenance time window of each key equipment. After merging and deduplicating the planned operation and maintenance time window and the temporary operation and maintenance time window, generate a list of valid time windows for the key equipment. The standard job title text field, the spatial access priority sorting sequence, the list of effective time windows for key equipment, and the encrypted iris feature ciphertext string are subjected to a four-dimensional association binding operation to form a complete node record unit in the spatiotemporal association network. The data storage structure is organized by the ascending order of the employee ID number field of the operation and maintenance personnel, and the complete graph data structure of the spatiotemporal correlation network is constructed. The complete graph data structure is then synchronously stored in the dual-machine hot standby system server and the remote cloud database. Extract the functional area name text and area code number corresponding to the allowed binary status identifier to generate a list of area codes that the standard job role name can pass through. Sort each area in the area code list from high to low according to the area security level value to obtain the spatial access priority sorting sequence in the job role dynamic permission parameters. Obtain the unique identification code of each key equipment in the power substation, as well as the planned operation and maintenance time window and temporary operation and maintenance time window of each key equipment. After merging and deduplicating the planned operation and maintenance time window and the temporary operation and maintenance time window, generate a list of valid time windows for the key equipment. The standard job title text field, the spatial access priority sorting sequence, the list of effective time windows for key equipment, and the encrypted iris feature ciphertext string are subjected to a four-dimensional association binding operation to form a complete node record unit in the spatiotemporal association network. The data storage structure is organized by the ascending order of the employee ID number field of the maintenance personnel, and the complete graph data structure of the spatiotemporal correlation network is constructed. The complete graph data structure is then synchronously stored in the dual-machine hot standby system server and the remote cloud database. 4.The method of claim 1, wherein, The process involves collecting the first iris image of maintenance personnel at the main entrance / exit of the power substation using a main iris recognition terminal, matching the first iris image with the iris feature code in the spatiotemporal correlation network to generate a successfully matched first identity identifier, and extracting the corresponding first job role dynamic permission parameters and first effective time window from the spatiotemporal correlation network based on the first identity identifier. Specifically, this includes: The system monitors the sensing level signal output by the infrared proximity sensor built into the main iris recognition terminal. When the sensing level signal changes from a low level to a high level, it determines that maintenance personnel are approaching the main iris recognition terminal, triggering the main iris recognition terminal to switch from standby power-saving mode to full-function operation mode and generating a first iris recognition trigger signal. In the full-function operation mode, the near-infrared light-emitting diode array inside the main iris recognition terminal is controlled to emit near-infrared light towards the eye area of ​​the maintenance personnel with a preset pulse width modulation waveform, illuminating the iris tissue; The high-resolution complementary metal-oxide-semiconductor image sensor inside the main iris recognition terminal is driven to continuously capture multiple frames of raw images containing the eye regions of the maintenance personnel, thereby obtaining a continuous time series of raw image frame sequences. For each frame of the original image in the original image frame sequence, human eye detection processing based on gray-scale integral projection is performed. The vertical coordinate interval of the human eye region is determined by the horizontal gray-scale integral projection curve, and the horizontal coordinate interval of the human eye region is determined by the vertical gray-scale integral projection curve. Based on the vertical coordinate interval and the horizontal coordinate interval, a local image block containing only the human eye region is cropped from the original image to obtain a human eye local image sequence. For each frame of the human eye local image sequence, image sharpness evaluation processing based on the Laplacian operator is performed. The Laplacian operator response value of each pixel in the human eye local image is calculated. The Laplacian operator response values ​​of all pixels are summed to generate an image sharpness score. Blurry human eye local images with image sharpness scores lower than a preset sharpness threshold are removed from the human eye local image sequence, and valid human eye local image sequences with sharpness requirements are retained. Motion feature analysis based on optical flow is performed on each frame of the effective human eye local image sequence. The optical flow vector of the pixels between two adjacent frames of effective human eye local images is calculated. Based on the direction consistency parameter and amplitude stability parameter of the optical flow vector, the micro-motion feature curve of the eye is generated. The eye micro-movement feature curve is input into a pre-trained liveness detection classifier for real / false detection processing. The liveness detection classifier outputs the probability value of the effective human eye local image sequence belonging to a real live iris. When the probability value is greater than a preset liveness probability threshold, a liveness detection pass result is generated. Based on the liveness detection result, the effective human eye local image with the highest image quality score is selected from the effective human eye local image sequence as the first iris image. The first iris image is subjected to iris region localization, normalization processing and feature extraction to generate a first iris feature code, and the first iris feature code is matched to determine the first identity identifier and the corresponding permission parameters.

5. The method of claim 4, wherein the method further comprises: determining whether the user is an authorized user based on the iris recognition; and if the user is an authorized user, allowing the user to access the power substation. The step of performing iris region localization, normalization processing, and feature extraction on the first iris image to generate a first iris feature code, and matching the first iris feature code to determine the first identity identifier and corresponding permission parameters, includes: The first iris image is subjected to iris region localization, normalization processing and feature extraction, and the first iris feature code is generated using the same iris texture feature point extraction algorithm as when constructing the spatiotemporal correlation network; The first iris feature code is compared with the iris feature code of each maintenance personnel in the spatiotemporal association network to calculate the Hamming distance, and a first similarity score set is obtained. The iris feature code of the maintenance personnel with the smallest Hamming distance is selected from the first similarity score set as the benchmark feature code for successful matching. Based on the successfully matched baseline feature code, a reverse index query is performed in the spatiotemporal correlation network to obtain a first identity identifier uniquely bound to the baseline feature code; Based on the first identity identifier, the corresponding first job role dynamic permission parameters and the first valid time window are extracted from the spatiotemporal association network. The first valid time window is compared with the current system time. If the current system time falls within the first valid time window, the first job role dynamic permission parameters are retained; otherwise, the first job role dynamic permission parameters are marked as invalid and an expiration alarm is triggered. 6.The method of claim 1, wherein, The step of extracting a set of tasks to be executed that overlap with the first valid time window from the operation and maintenance task database based on the first identity identifier and the first valid time window, filtering the set of tasks to be executed based on the dynamic permission parameters of the first job role, generating a first authorized task list, and pushing the first authorized task list to the main iris recognition terminal for display specifically includes: Access the operation and maintenance task database deployed on the system server with dual-machine hot standby, and read all operation and maintenance task records stored in the operation and maintenance task database for the day. Each operation and maintenance task record includes task number, task type, task start timestamp, task end timestamp, task associated device identifier, and a list of roles that the task is allowed to execute. The first effective time window is parsed to obtain the start and end times of the first effective time window, and the task plan start timestamp and task plan end timestamp are compared with the start and end times of the first effective time window to determine the overlap of time intervals. For each maintenance task record, if the task plan start timestamp is earlier than or equal to the end time of the first effective time window and the task plan end timestamp is later than or equal to the start time of the first effective time window, then it is determined that the time interval of the maintenance task record overlaps with the first effective time window, and the maintenance task record is included in the candidate task set. Extract the list of roles that the task can execute for each operation and maintenance task record in the candidate task set, and perform an intersection operation between the list of roles that the task can execute and the standard job role names contained in the dynamic permission parameters of the first job role. If the result of the intersection operation is not empty, it is determined that the role and permission matching of the operation and maintenance task record is successful, and the operation and maintenance task record is retained; otherwise, the operation and maintenance task record is removed from the candidate task set, and a set of tasks to be executed after dual filtering by time window and role and permission is obtained. For each maintenance task record in the set of tasks to be executed, the task priority is parsed, the value of the task priority field is extracted, and the set of tasks to be executed is sorted in descending order of task priority value to generate the first authorized task list. The first authorized task list is encapsulated into a data format that conforms to the main iris recognition terminal display protocol, and the encapsulated data is sent to the main iris recognition terminal via industrial Ethernet; After receiving the first authorized task list, the main iris recognition terminal displays the summary information of each task in a list format on the touch screen. The summary information includes the task number, task type, task start timestamp, and task end timestamp. A touch selection button is set for each task on the touch screen of the main iris recognition terminal. When the maintenance personnel click the touch selection button, a task selection signal containing the corresponding task number is generated and the task selection signal is sent back to the system server. 7.The method of claim 1, wherein the method further comprises: receiving a request for a power substation operation and maintenance personnel safety authentication and operation management from the power substation operation and maintenance personnel; and transmitting the request to the power substation operation and maintenance personnel safety authentication and operation management server. The process of receiving the target task identifier selected by the maintenance personnel in the first authorized task list, retrieving the corresponding target device identifier and the standard operation procedure document of the target device from the device ledger database based on the target task identifier, and sending the target device identifier and the standard operation procedure document to the device-level iris recognition terminal deployed next to the target device, specifically includes: The main iris recognition terminal receives a task selection signal generated after the maintenance personnel click the touch selection button through its communication module, and parses the task number of the selected target task from the task selection signal as the target task identifier. The target task identifier is obtained by indexing and querying the operation and maintenance task database according to the target task identifier, extracting the task-associated device identifier corresponding to the target task identifier; The target device identifier is queried in the equipment ledger database to retrieve the equipment type, equipment model, equipment installation location coordinates, and standard operation procedure document stored in the equipment ledger database that correspond to the target device identifier. The standard operating procedure document includes a list of pre-operation safety checks for the target device, a sequence of operating steps, the specified range of operating parameters for each step, and a list of post-operation status confirmation items. Based on the target device identifier, retrieve the list of valid time windows for the target device in the spatiotemporal correlation network, extract the planned maintenance time window and temporary maintenance time window for the target device, and compare the current system time with the list of valid time windows for the target device; If the current system time is not within any time window in the list of valid time windows of the target device, a device operation time violation alarm is generated and pushed to the main iris recognition terminal and the monitoring center. If the current system time is within the list of valid time windows of the target device, then navigation path planning data from the main entrance / exit to the target device installation location is generated based on the installation location coordinates of the target device and the location coordinates of the main iris recognition terminal. The navigation path planning data, the target device identifier, and the standard operation procedure document are packaged and processed to generate a joint data package containing navigation information and operation guidance; The combined data packet is sent via industrial Ethernet to a device-level iris recognition terminal deployed next to the target device. After receiving the packet, the device-level iris recognition terminal displays the navigation path and operation guidance information on its display screen. 8.The method of claim 1, wherein the method further comprises: determining whether the user is an authorized user based on the iris recognition; and if the user is the authorized user, allowing the user to access the power substation. The process involves acquiring a second iris image of the maintenance personnel via the device-level iris recognition terminal, matching the second iris image with the iris feature code in the spatiotemporal correlation network to generate a successfully matched second identity identifier, performing joint permission verification based on the second identity identifier and the target task identifier to generate an operation permission instruction, and sending the operation permission instruction to the device controller connected to the target device to drive the target device to execute the operation sequence corresponding to the standard operation procedure document. Specifically, this includes: The proximity sensor built into the device-level iris recognition terminal is monitored. When the terminal detects that maintenance personnel have arrived at the operation area next to the target device, it is woken up and put into working state. The terminal displays the navigation path planning data and the summary of the standard operation procedure document on its screen. The device-level iris recognition terminal acquires real-time iris images of the eyes of maintenance personnel through its iris acquisition module, and performs liveness detection and image quality screening on the real-time iris images to obtain a second iris image that meets the quality requirements. The second iris image is subjected to iris region localization, normalization processing and feature extraction, and the second iris feature code is generated using the same iris texture feature point extraction algorithm as when constructing the spatiotemporal correlation network; The second iris feature code is compared with the iris feature code of each maintenance personnel in the spatiotemporal association network to calculate the Hamming distance, and a second similarity score set is obtained. The iris feature code of the maintenance personnel with the smallest Hamming distance is selected from the second similarity score set as the second benchmark feature code for successful matching. Based on the second reference feature code, a reverse index query is performed in the spatiotemporal correlation network to obtain a second identity identifier that is uniquely bound to the second reference feature code; Based on the second identity identifier, extract the corresponding second job role dynamic permission parameters and the second effective time window from the spatiotemporal correlation network, and verify whether the current system time falls within the second effective time window; Based on the target task identifier, query the list of roles that the target task is allowed to execute in the operation and maintenance task database, and compare the standard job role name contained in the dynamic permission parameter of the second job role with the list of roles that the task is allowed to execute. If the standard job role name contained in the dynamic permission parameters of the second job role exists in the role list that the task is allowed to execute, a role matching pass signal is generated; otherwise, a role matching failure signal is generated and the operation process is terminated. Based on the target task identifier, query the standard operating procedure document of the target equipment in the equipment ledger database, and parse the sequence of operation steps and the range of operation parameter specifications for each step in the standard operating procedure document. The operation step sequence is presented in an interactive interface on the display screen of the device-level iris recognition terminal. The operation and maintenance personnel wait for the operation instructions and parameters to be entered one by one according to the operation step sequence, and verify the input operation instructions and parameters to generate an operation permission instruction and execute it. 9.The method of claim 8, wherein, The process of verifying the input operation instructions and parameters to generate and execute operation permission instructions includes: The system receives the first step operation instruction and the first step operation parameters input by the operation and maintenance personnel on the interactive interface. It performs string matching between the first step operation instruction and the expected operation instruction of the corresponding step in the standard operation process document, and performs numerical range matching between the first step operation parameters and the operation parameter specification range of the corresponding step. If the string and the numerical range are matched successfully, the operation permission instruction is generated and sent to the device controller connected to the target device. The device controller drives the operating mechanism of the target device to perform the action corresponding to the operation instruction in the first step according to the operation permission instruction. During the operation of the target device's operating mechanism, the status sensor data of the target device is collected in real time. The status sensor data is dynamically compared with the expected status change curve of the corresponding step in the standard operation procedure document to generate an operation execution process deviation index. If the deviation index of the operation execution process exceeds the preset deviation threshold, an emergency stop command is immediately sent to the device controller to suspend the operation of the target device and display an abnormal alarm message on the device-level iris recognition terminal.

10. An iris recognition-based power substation operator safety authentication and operation management system, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the iris recognition-based power substation maintenance personnel security authentication and operation management method according to any one of claims 1 to 9 by executing the machine-executable instructions.

Images