Transformer substation safety measure ticket execution management and control terminal equipment and method supporting local AI identification

Abstract

The invention relates to a transformer station safety measure ticket execution management and control terminal device and method supporting local AI identification, the terminal device comprises a safety measure ticket execution management and control terminal and a transformer station auxiliary terminal, the safety measure ticket execution management and control terminal comprises an acquisition network, an extraction network, a unified video platform, a power equipment monitoring management application system and an artificial intelligence platform, the acquisition network is connected with the unified video platform, and the unified video platform and the extraction network are connected with the power equipment monitoring management application system; the power equipment monitoring and management application system is in closed-loop bidirectional connection with the artificial intelligence platform through the agent calling interface; the substation auxiliary terminal comprises a fire-fighting information transmission control unit, a security monitoring unit and a power and environment monitoring unit; the method has the advantages that multi-source heterogeneous data integration management is achieved, high-precision identification of the equipment state is achieved, and collection-analysis-application-optimization closed-loop management is formed.

Description

Technical Field

[0001] This invention belongs to the field of substation management and control technology, specifically relating to a substation safety ticket execution management and control terminal device and method that supports local AI recognition. Background Technology

[0002] In traditional substation work permit management, workers need to obtain paper work permits and pass an examination before entering the substation. This process is cumbersome due to the application and review process, the fragility of paper work permits (e.g., easy damage and loss), and the vulnerability of their storage to environmental factors. These issues severely impact worker efficiency and data security. Furthermore, issuing work permits requires manual review and communication, incurring high communication costs. The actual work process within the substation is not effectively monitored, and the work behavior cannot be effectively controlled based on the work permit's content. Therefore, it is essential to provide a substation safety permit execution control terminal device and method that integrates and manages multi-source heterogeneous data, achieves high-precision equipment status identification, and forms a closed-loop management system encompassing data acquisition, analysis, application, and optimization, while supporting local AI recognition. Summary of the Invention

[0003] (I) Technical Solution

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a substation safety ticket execution control terminal device and method that supports local AI recognition, enabling multi-source heterogeneous data integration and management, high-precision identification of equipment status, and the formation of a closed-loop management system of acquisition-analysis-application-optimization.

[0005] The objective of this invention is achieved as follows: Firstly, a substation safety ticket execution and control terminal device supporting local AI recognition, comprising a safety ticket execution and control terminal and a substation auxiliary terminal. The substation auxiliary terminal establishes a connection with the safety ticket execution and control terminal using a local AI recognition unit. The safety ticket execution and control terminal includes a data acquisition network, an extraction network, a unified video platform, a power equipment monitoring and management application system, and an artificial intelligence platform. The data acquisition network is connected to the unified video platform, and the unified video platform and the extraction network are connected to the power equipment monitoring and management application system. The power equipment monitoring and management application system achieves a closed-loop bidirectional connection with the artificial intelligence platform through an intelligent agent call interface. The local AI recognition unit includes an edge computing unit, an IoT tag reading and writing module, and an alarm module. The substation auxiliary terminal includes a fire information transmission and control unit, a security monitoring unit, and an environmental monitoring unit.

[0006] Furthermore, the acquisition network consists of mobile terminals, access points (APs), and aggregation switches, used to acquire video and / or image data from field devices and connect to a unified video platform.

[0007] Furthermore, the extraction network consists of a PMS and a safety ticket, and is used to extract the safety ticket business data from the PMS.

[0008] Furthermore, the artificial intelligence platform includes an ORC recognition module, a knowledge base, a state recognition model module, and a state analysis intelligent agent; wherein the ORC recognition module is used to perform OCR recognition on safety tickets and store the recognition results in the knowledge base after structured processing; the state recognition model module uses image samples of the equipment as the training set, and is trained and optimized based on the 0.45b Guangming Power visual large model to obtain a proprietary model for accurate identification of equipment state, thereby achieving high-precision visual recognition of equipment state.

[0009] Furthermore, the power equipment monitoring and management application system also includes a to-do items module, a completed items module, a safety measure ticket issuance module, a safety measure ticket editing module, a safety measure ticket execution module, a safety measure ticket archiving module, an archive query module, a compilation ticket query module, and a work task module; wherein the to-do items module and the completed items module each include a pending review sub-module and a reviewed sub-module; the safety measure ticket execution module includes an execution sub-module and a recovery sub-module; the safety measure ticket archiving module includes a safety measure pre-archiving check module and a safety measure archiving module.

[0010] Furthermore, the IoT tag reading and writing module based on the local AI recognition unit is connected to the artificial intelligence platform, and the alarm module based on the local AI recognition unit is connected to the security monitoring unit of the substation auxiliary terminal.

[0011] Furthermore, the fire information transmission control unit is connected to an automatic fire alarm system via an RS485 / RJ45 / CAN interface, to a fixed fire extinguishing system and other controlled fire-fighting equipment via an I / O interface, and to a fire water tank level transmitter, a fire pipeline pressure transmitter, and a fire power supply voltage transmitter via an analog interface.

[0012] Furthermore, the security monitoring unit is connected to an audible and visual alarm, an infrared dual-technology detector, an infrared beam detector, an emergency alarm button, and an access control controller via an I / O interface, and is connected to an access control controller, an electronic fence, and a burglar alarm controller via RJ45 / RS485.

[0013] Furthermore, the environmental monitoring unit is connected to a water immersion detector, a water leakage detector, a fan control box, a water pump control box, and a dehumidifier controller via an I / O interface, and to a micro-weather sensor, a temperature and humidity sensor, a water level sensor, an SF6 fluoride content sensor, an air conditioning controller, and a lighting controller via an RS485 interface.

[0014] Secondly, a substation safety ticket execution control method supporting local AI recognition is provided. Based on the aforementioned substation safety ticket execution control terminal equipment supporting local AI recognition, intelligent monitoring of the substation and closed-loop management of safety tickets are achieved. The control method includes the following steps:

[0015] Step 1: The acquisition network acquires video and / or image data from field devices via mobile terminals, access points (APs), and computing sticks, and connects them to a unified video platform through an aggregation switch;

[0016] Step 2: Extract network data and simultaneously extract security ticket business data from PMS;

[0017] Step 3: The two types of data obtained in Step 1-2 provide basic data sources for the artificial intelligence platform and the power equipment monitoring and management application system, realizing the integration and preliminary management of multi-source heterogeneous data;

[0018] Step 4: Model training and agent construction on the artificial intelligence platform;

[0019] Step 5: The power equipment monitoring and management application system calls the interface through the intelligent agent, relies on the artificial intelligence platform to carry out production business, and feeds back the results of the business execution to the artificial intelligence platform for model iteration, intelligent agent optimization and scenario encapsulation upgrade, forming a closed-loop management of "data collection → intelligent analysis → business application → feedback optimization".

[0020] (II) Beneficial Effects

[0021] 1. This invention acquires video / image data from field equipment via a data acquisition network and connects it to a unified video platform; simultaneously, it extracts safety ticket business data from the PMS via the network. These two types of data provide basic data sources for the artificial intelligence platform and the power equipment monitoring and management application system, respectively, thereby realizing the integration and preliminary management of multi-source heterogeneous data.

[0022] 2. This invention performs OCR recognition on safety tickets, stores the recognition results in a knowledge base after structured processing, and conducts model training and optimization based on the knowledge base to provide data and knowledge support for the intelligent analysis of safety tickets and construct IoT tags for screen cabinets;

[0023] 3. This invention uses image samples of devices such as "pressure plate, power circuit breaker, and operating handle" as a training set, and is trained and optimized based on the 0.45b Guangming Power vision large model to obtain a proprietary model for accurate identification of pressure plate status, thereby achieving high-precision visual identification of device status.

[0024] 4. This invention integrates the capabilities of "intelligent identification of safety tickets" and "precise identification of pressure plate status" to construct an intelligent agent for intelligent comparison and analysis of safety tickets and pressure plate status; at the same time, it encapsulates various intelligent application scenarios of the artificial intelligence platform, enabling it to be called by the upper-level power equipment monitoring and management application system.

[0025] 5. The power equipment monitoring and management application system of the present invention uses intelligent agents to call interfaces, relies on the capabilities of artificial intelligence platforms to carry out production operations, and feeds back the results of the operations to the artificial intelligence platform, forming a closed-loop technical process of "data acquisition → intelligent analysis → business application → feedback optimization". Attached Figure Description

[0026] Figure 1 This is a diagram of the overall architecture of the present invention.

[0027] Figure 2 This is a diagram showing the architecture of the power equipment monitoring and management application system of the present invention.

[0028] Figure 3 This is a structural diagram of the fire information transmission control unit of the present invention.

[0029] Figure 4 This is a structural diagram of the security monitoring terminal of the present invention.

[0030] Figure 5 This is a structural diagram of the environmental monitoring terminal of the present invention.

[0031] Figure 6 This is a flowchart of the present invention. Detailed Implementation

[0032] The present invention will be further described below with reference to the embodiments and / or accompanying drawings.

[0033] Example 1

[0034] like Figure 1-5As shown, a substation safety ticket execution and control terminal device supporting local AI recognition includes a safety ticket execution and control terminal and a substation auxiliary terminal. The substation auxiliary terminal establishes a connection with the safety ticket execution and control terminal using a local AI recognition unit. The safety ticket execution and control terminal includes a data acquisition network, an extraction network, a unified video platform, a power equipment monitoring and management application system, and an artificial intelligence platform. The data acquisition network is connected to the unified video platform, and the unified video platform and the extraction network are connected to the power equipment monitoring and management application system. The power equipment monitoring and management application system achieves a closed-loop bidirectional connection with the artificial intelligence platform through an intelligent agent call interface. The local AI recognition unit includes an edge computing unit, an IoT tag reading and writing module, and an alarm module. The substation auxiliary terminal includes a fire information transmission and control unit, a security monitoring unit, and an environmental monitoring unit.

[0035] The acquisition network consists of mobile terminals, access points (APs), and aggregation switches, and is used to acquire video and / or image data from field devices and connect them to a unified video platform.

[0036] The extraction network consists of a PMS and a security ticket, and is used to extract the security ticket business data from the PMS.

[0037] The artificial intelligence platform includes an ORC recognition module, a knowledge base, a state recognition model module, and a state analysis intelligent agent. The ORC recognition module is used to perform OCR recognition on safety tickets and stores the recognition results in the knowledge base after structured processing. The state recognition model module uses image samples of the equipment as the training set and is trained and optimized based on the 0.45b Guangming Power visual large model to obtain a proprietary model for accurate equipment state recognition, thereby achieving high-precision visual recognition of equipment state.

[0038] The power equipment monitoring and management application system also includes a to-do items module, a completed items module, a safety measure ticket issuance module, a safety measure ticket editing module, a safety measure ticket execution module, a safety measure ticket archiving module, an archive query module, a compilation ticket query module, and a work task module; wherein the to-do items module and the completed items module each contain a pending review sub-module and a reviewed sub-module; the safety measure ticket execution module contains an execution sub-module and a recovery sub-module; the safety measure ticket archiving module contains a safety measure pre-archiving check module and a safety measure archiving module.

[0039] As a specific implementation method, ① the to-do items module includes two sub-modules: the pending review sub-module and the reviewed sub-module.

[0040] 1) Pending Review Submodule: Enables viewing and reviewing basic information such as applicant information, repair location, repair content, work content, and repair time; allows downloading, viewing, and reviewing attachments of safety measures pending review; after review, the module is signed and transferred, and the item is added to the completed items list.

[0041] 2) Reviewed Submodule: This module allows users to view the review status of secondary safety measures that have already been reviewed. If the review status is "approved," users can proceed to the "completed items" section and begin implementing the safety measures. If the review status is "rejected," users need to modify the safety measures according to the rejection comments and resubmit them for review.

[0042] ② Completed Items Module: Contains two sub-modules: Pending Review Sub-module and Reviewed Sub-module.

[0043] 1) Sub-module pending review: Allows users to view basic information such as maintenance location, maintenance content, work content, and maintenance time. After the safety measures editing module completes the submission for review, it will automatically enter the completed items and mark them as pending review.

[0044] 2) Reviewed Submodule: Allows users to view the review status of secondary safety measures that have already been reviewed. Items are automatically transferred to this module after being processed from the pending items.

[0045] ③ Safety Measures Ticket Issuance Module: This module provides editing capabilities for work ticket number, work time, substation, and maintenance content. The power equipment monitoring and management application system will send a safety measures generation request to the server. The server will generate a safety measures ticket based on the above information and send it to the power equipment monitoring and management application system. Upon confirmation, the system will redirect to the safety measures editing module. This module also supports the issuance of safety measures tickets for intelligent stations, regular stations, and the retrieval of historical tickets. For practical considerations, intelligent stations use the typical ticket method, while regular stations use the compiled safety measures example ticketing method.

[0046] ④ Ancuo Ticket Editing Module: In this module, you can select all the ancuo tickets automatically generated for this ancuo, view and modify them one by one. After all the editing is completed, click to apply for review and sign, and then enter the circulation process.

[0047] ⑤ Safety Measures Execution Module: Includes an execution submodule and a recovery submodule. It supports selecting all approved safety measures tickets for this maintenance. Before execution or recovery, both the executor and the monitoring personnel must sign, and the execution status checkbox will be enabled after signing. It supports sequential execution of safety measures tickets. Before execution, the operation and maintenance personnel must sign in the safety measures editing status record interface. After execution, it returns to the main interface. It supports reverse execution of safety measures tickets. After recovery, it returns to the safety measures editing status record interface, where the operation and maintenance personnel must sign. It supports selecting executed items during execution; it supports missing item prompts; and it supports re-editing the workflow during execution.

[0048] ⑥ Safety Measures Ticket Archiving Module: This module includes two sub-functions: pre-archiving inspection and archiving. The inspection module converts the safety measures content of this maintenance work into Word format for display. The Word format allows for a direct and intuitive inspection of all safety measures tickets for this maintenance. After the inspection is completed, the archiving module uploads the safety measures tickets for this maintenance to the server for archiving. Once archiving is complete, the archiving ticket query module can be used to check whether the archiving is finished.

[0049] ⑦ Archived Ticket Inquiry Module: This module allows users to download and view archived secondary safety measures tickets for each substation (Word file).

[0050] ⑧ Compilation Ticket Query Module: This module allows users to download and view (Word files) compilation tickets for regular and intelligent stations compiled in this region.

[0051] ⑨ Work Task Module: This module supports importing and editing work task Excel spreadsheets, distributing monthly work tasks, and allowing users to view their individual work tasks.

[0052] The IoT tag reading and writing module based on the local AI recognition unit is connected to the artificial intelligence platform, and the alarm module based on the local AI recognition unit is connected to the security monitoring unit of the substation auxiliary terminal; wherein the edge computing unit of the local AI recognition unit adopts an edge computing node and cloud collaborative architecture.

[0053] The fire information transmission control unit is connected to an automatic fire alarm system via RS485 / RJ45 / CAN interface, to a fixed fire extinguishing system and other controlled fire equipment via I / O interface, and to a fire water tank level transmitter, a fire pipeline pressure transmitter, and a fire power supply voltage transmitter via analog interface.

[0054] As a specific and feasible implementation method, the fire information transmission control unit (which can be the BW-4UXF-FI-TCU-ZK type control unit) is deeply integrated with fire alarm controllers, smoke detectors and other equipment to collect key signals such as smoke concentration and temperature anomalies in real time, and quickly transmits the information to the security monitoring unit through standard protocols such as DLT860 and IEC61850.

[0055] Intelligent linkage: Triggers coordinated responses from multiple systems, including video capture, audible and visual alarms, and non-fire-fighting power cut-off, significantly improving emergency response efficiency.

[0056] The security monitoring unit is connected to an audible and visual alarm, an infrared dual-technology detector, an infrared beam detector, an emergency alarm button, and an access control controller via an I / O interface, and is connected to an access control controller, an electronic fence, and a burglar alarm controller via an RJ45 / RS485 interface.

[0057] As a specific implementation method, the security monitoring unit (which can be a BW-4UAF-SD-SD-ZK type terminal device) integrates signals from devices such as infrared beam detectors, electronic fences, and access control systems to monitor abnormal behaviors such as personnel intrusion and forced dismantling of equipment in real time, and uploads the data to the main station through a standardized protocol; it can also be connected to an alarm module based on a local AI recognition unit to achieve localized AI recognition and real-time alarms.

[0058] Panoramic Perception: Supports the access of multiple types of security devices, enabling "one-screen visualization" of the security status within the station.

[0059] The environmental monitoring unit is connected to a water immersion detector, a water leakage detector, a fan control box, a water pump control box, and a dehumidifier controller via an I / O interface, and to a micro-meteorological sensor, a temperature and humidity sensor, a water level sensor, an SF6 fluoride content sensor, an air conditioning controller, and a lighting controller via an RS485 interface.

[0060] As a feasible implementation method, the environmental monitoring unit (which can adopt BW-4UDH-PE-PED-ZK) accurately collects data such as micro-meteorological conditions, temperature and humidity, SF6 concentration, and water immersion, and controls equipment such as fans, air conditioners, and dehumidifiers in conjunction with it to achieve: precise control: automatically adjust the temperature and humidity in the station to prevent risks such as equipment condensation and SF6 leakage; reliable interaction: communicate efficiently with the main station or security monitoring unit through the DL / T860 protocol, support edge computing to achieve rapid local decision-making, and realize the integration of data acquisition, protocol parsing, edge storage, and alarm linkage. It supports local touch screen operation and remote WEB management to achieve "terminal-level closed-loop control".

[0061] This invention relates to a terminal device and method for controlling the execution of safety tickets in substations, supporting local AI recognition. In use, this invention acquires video / image data from field equipment via a network and connects it to a unified video platform. Simultaneously, it extracts safety ticket business data from the PMS (Power Management System). These two types of data provide basic data sources for the artificial intelligence platform and the power equipment monitoring and management application system, respectively, achieving the integration and preliminary management of multi-source heterogeneous data. This invention performs OCR recognition on safety tickets, structures the recognition results, and stores them in a knowledge base. Based on the knowledge base, it conducts model training and optimization, providing data and knowledge support for the intelligent analysis of safety tickets and constructing IoT tags for power supply cabinets. This invention uses image samples of equipment such as pressure plates, circuit breakers, and operating handles as a training set, and trains and optimizes based on the 0.45b Guangming Power Vision Large Model. This invention obtains a proprietary model for accurate identification of pressure plate status, achieving high-precision visual recognition of equipment status. It integrates the capabilities of "intelligent identification of safety tickets" and "accurate identification of pressure plate status," constructing an intelligent agent for intelligent comparison and analysis of safety tickets and pressure plate status. Simultaneously, it encapsulates various intelligent application scenarios of the artificial intelligence platform, enabling it to be invoked by upper-level power equipment monitoring and management application systems. The power equipment monitoring and management application system of this invention, through the intelligent agent's call interface, relies on the capabilities of the artificial intelligence platform to conduct production operations and feed back the results of these operations to the artificial intelligence platform, forming a closed-loop technical process of "data acquisition → intelligent analysis → business application → feedback optimization." This invention has the advantages of achieving integrated management of multi-source heterogeneous data, high-precision identification of equipment status, and forming a closed-loop management system of acquisition-analysis-application-optimization.

[0062] Example 2

[0063] like Figure 6 As shown, a substation safety measure ticket execution control method supporting local AI recognition is described above. Based on the substation safety measure ticket execution control terminal equipment supporting local AI recognition, intelligent monitoring of the substation and closed-loop management of safety measure tickets are achieved. The control method includes the following steps:

[0064] Step 1: The acquisition network acquires video and / or image data from field devices via mobile terminals, access points (APs), and computing sticks, and connects them to a unified video platform through an aggregation switch;

[0065] Step 2: Extract network data and simultaneously extract security ticket business data from PMS;

[0066] Step 3: The two types of data obtained in Step 1-2 provide basic data sources for the artificial intelligence platform and the power equipment monitoring and management application system, realizing the integration and preliminary management of multi-source heterogeneous data;

[0067] As a feasible implementation method, a data acquisition network consisting of mobile terminals, APs (wireless access points), and computing sticks is used to acquire video / image data of field equipment (pressure plates, power circuit breakers, operating handles, etc.) and connect them to a unified video platform; at the same time, safety ticket business data is extracted from PMS (Power Production Management System); the two types of data provide basic data sources for the artificial intelligence platform and the power equipment monitoring and management application system, respectively, realizing the integration and preliminary management of multi-source heterogeneous data.

[0068] Specifically, 1. Video / image data feature extraction algorithm: Edge detection algorithm (Sobel operator) formula: Among them, the right / left / bottom / top pixel values ​​refer to the pixel grayscale values ​​in the four adjacent directions of the target pixel; edge strength is a sharpness index used to identify the edge contour of the device, with a value range of 0-255.

[0069] The formula for the SIFT feature matching algorithm is: Feature point descriptor = [Gradient direction histogram (0°-360°) × gradient magnitude normalization coefficient], where the gradient direction histogram is the statistical distribution of gradient directions by dividing 360 degrees into 8 intervals; the gradient magnitude normalization coefficient is a correction factor to eliminate the influence of illumination changes, and its value is 1 / total number of pixels in the neighborhood of the feature point.

[0070] 2. Structured processing of safety ticket business data: Operation step coding algorithm formula: Operation code = Equipment type code × 1000 + Operation type code × 100 + Voltage level code, where the equipment type code is pressure plate = 1, circuit breaker = 2, handle = 3 (three-bit binary code); the operation type code is closing = 1, opening = 2, adjustment = 3; the voltage level code is 220kV = 1, 110kV = 2, 35kV = 3.

[0071] Temporal correlation algorithm formula: Temporal weight = e (-|当前时间戳-操作时间戳 / τ) / (1+e (-Δt / τ) ), where τ is the time decay constant, which is 1 / 2 of the average time of the operation interval; Δt is the time difference between the current time and the operation time (unit: seconds); the time sequence weight is a weight coefficient that reflects the timeliness of the operation, ranging from [0,1].

[0072] 3. Multi-source data fusion algorithm: Feature-level fusion formula: Fusion feature vector = α × image feature vector + β × text feature vector, where α is the image feature weight coefficient, determined by cross-validation, with a value range of [0.6, 0.8]; β is the text feature weight coefficient, β = 1 - α; the image feature vector is a 128-dimensional feature vector after PCA dimensionality reduction; the text feature vector is a 300-dimensional semantic vector after Word2Vec encoding.

[0073] Decision-level fusion formula: Final decision result = max(image recognition confidence × γ, text parsing confidence × (1-γ)), where γ is the decision weight coefficient, which is dynamically adjusted according to the system misjudgment cost, and the value range is [0.4, 0.7]; the confidence is the probability value normalized by the Softmax function.

[0074] 4. Data quality assessment formula: Data integrity index: Integrity = (Valid data volume / Total data volume) × 100%, where the valid data volume is the amount of qualified data after screening by the outlier detection algorithm; the total data volume is the total amount of original collected data.

[0075] Data timeliness index: Timeliness = 1 - (Current time - Data generation time) / Maximum allowable delay, where the maximum allowable delay is the maximum data delay threshold set by the system (unit: seconds).

[0076] 5. Heterogeneous data standardization: Numerical data standardization: Standardized value = (Original value - Minimum value) / (Maximum value - Minimum value).

[0077] Categorical data encoding: One-hot encoding = [0,0,1,0] (taking 4 types of equipment as an example), which converts discrete categorical features into binary vector representations.

[0078] This invention realizes the entire process from raw data acquisition to standardized fusion, ensuring the unified expression of multi-source heterogeneous data in the feature space and decision space, providing a structurally consistent and semantically aligned basic data source for artificial intelligence platforms, while meeting the requirements of power equipment monitoring and management application systems for data real-time performance and accuracy.

[0079] Step 4: Model training and agent construction on the artificial intelligence platform;

[0080] As one feasible implementation method, the construction process is as follows:

[0081] 4.1: The ORC recognition module performs OCR recognition on safety tickets, and stores the recognition results in a knowledge base after structured processing; based on the knowledge base, it conducts model training and optimization to provide data and knowledge support for the intelligent analysis of safety tickets and builds IoT tags for display cabinets; the constructed IoT tags for display cabinets can be read and written by the IoT tag reading and writing module.

[0082] 4.2: The status recognition model module uses image samples of equipment such as "pressure plate, power circuit breaker, and operating handle" as the training set. Based on the 0.45b Guangming Power visual large model, it is trained and optimized to obtain a proprietary model for accurate recognition of pressure plate status, realizing high-precision visual recognition of equipment status.

[0083] 4.3: The status analysis intelligent agent integrates "intelligent identification of safety tickets" and "precise identification of pressure plate status" to build an intelligent agent for intelligent comparison and analysis of safety tickets and pressure plate status; at the same time, it encapsulates various intelligent application scenarios of the artificial intelligence platform, enabling it to be called by the upper-level power equipment monitoring and management application system.

[0084] Specifically, 1. Ancuo Ticket OCR Recognition and Structured Processing Algorithm: OCR Text Parsing Formula: Structured Field = Regular Expression Matching (Original OCR Text, "Operating Equipment: ([Pressure Plate|Blower|Handle]) + Operation Type: ([Close|Open|Adjust]) + Voltage Level: ([220kV|110kV|35kV])"), where the original OCR text is the text string recognized by the Tesseract engine; the regular expression is a predefined matching pattern used to extract the three core fields of equipment type, operation type, and voltage level; the structured field is the generated JSON format data, such as {"Equipment Type": "Pressure Plate", "Operation Type": "Close", "Voltage Level": "220kV"}.

[0085] The knowledge base storage formula is: Knowledge Entry = Hash Code (Operation Step ID) + Structured Field + Timestamp. Where, Operation Step ID is a unique identifier generated based on the operation coding algorithm (e.g., "Power Plate_Close_220kV_001"); Hash Code is a 32-bit unique identifier generated using the SHA-256 algorithm; and Timestamp is the precise time (accurate to milliseconds) when the operation record was generated.

[0086] 2. Equipment Status Visual Recognition Model Training Algorithm: Visual Model Loss Function: Loss = -∑(True Label × log(Predicted Probability) + (1 - True Label) × log(1 - Predicted Probability)), where the true label is the actual labeled value of the equipment status (0 or 1, representing different states); the predicted probability is the probability value of the equipment status output by the model (processed by the Sigmoid activation function); the loss value is the objective function value used for backpropagation optimization, and the smaller the value, the more accurate the prediction.

[0087] Model optimization formula: New weights = Old weights - Learning rate × Gradient, where the learning rate is initially set to 0.001 and decays by 10% every 10 training rounds; the gradient is the partial derivative of the loss function with respect to the weights, calculated using the chain rule; and the new weights are the updated model parameter values ​​used in the next training iteration.

[0088] 3. Agent Comparison and Analysis Algorithm: State Matching Degree Calculation: Matching Degree = (Ancuo Ticket Feature Vector · Equipment State Feature Vector) / (Ancuo Ticket Feature Vector Modulus × Equipment State Feature Vector Modulus), where the Ancuo Ticket Feature Vector is a 128-dimensional semantic vector encoded by Word2Vec; the Equipment State Feature Vector is a 256-dimensional visual feature vector extracted by CNN; the matching degree is a cosine similarity value, ranging from [-1, 1], and the larger the value, the higher the state consistency.

[0089] Intelligent decision-making formula: Decision result = if (matching degree > threshold) then "state consistent" else "state abnormal", where the threshold is set to 0.85 based on experience and can be dynamically adjusted according to the actual scenario; the decision result is the final comparison conclusion, which is used to trigger alarms or normal processes.

[0090] 4. Intelligent Application Encapsulation Interface Specification: API Request Format: Request Body = {"Application Scenario": "Safety Ticket Comparison", "Input Data": {"Image Data": base64 encoded, "Safety Ticket Text": string}}, where the application scenario identifies the specific intelligent function module being called; the input data contains the image data required for visual recognition and the text data required for OCR recognition; the base64 encoding is the text format converted from the binary data of the image file.

[0091] API response format: Response body = {"Status code": 200, "Result data": {"Match degree": 0.92, "Decision conclusion": "Status consistent", "Timestamp": 1630412345678}}, where the status code is the standard HTTP status code, 200 indicates success, and 400 indicates an input error; the result data contains specific numerical values ​​and textual conclusions of the comparison analysis; the timestamp is the precise time when the response was generated, used for system log tracking.

[0092] This invention realizes the entire process from safety ticket OCR recognition to equipment status visual recognition, and completes the fusion decision of multi-source data through intelligent agent comparison analysis algorithm; all interfaces are encapsulated in standardized JSON format to ensure that the upper-level power equipment monitoring and management application system can call them seamlessly, and realize the deep integration of safety ticket intelligent recognition and equipment status accurate recognition.

[0093] Step 5: The power equipment monitoring and management application system calls the interface through the intelligent agent and relies on the artificial intelligence platform to carry out production business such as operation record monitoring, safety measure ticket operation record filing, operation video review, and intelligent comparison analysis. The results of the business execution are fed back to the artificial intelligence platform for model iteration, intelligent agent optimization and scenario encapsulation upgrade, forming a closed-loop management of "data collection → intelligent analysis → business application → feedback optimization".

[0094] As one possible specific implementation method, it is as follows:

[0095] 1. Data Acquisition Quality Verification Algorithm: Data Integrity Verification Formula: Integrity Score = (Number of Valid Frames / Total Number of Frames) × 100% + (Key Field Completeness Rate × Weighting Coefficient), where the number of valid frames is the number of image frames retained after noise reduction; the key field completeness rate is the percentage of complete fields such as device number and operation time in the safety ticket; and the weighting coefficient is 0.7 for image data and 0.3 for text data in this embodiment, reflecting the difference in importance between different data sources.

[0096] Data timeliness verification formula: Timeliness score = 1 - (Current time - Data generation time) / Maximum allowable delay, where the maximum allowable delay is a system preset value (e.g., operation records are required to be uploaded within 30 minutes); the timeliness score range is [0,1], and a score below 0.6 triggers an early warning mechanism.

[0097] 2. Intelligent Analysis Feedback Optimization Algorithm: Model Iteration Loss Function: New Loss Value = Original Loss Value + λ × (Business Feedback Error × Dynamic Weight), where λ is the regularization coefficient, with an initial value of 0.1, decaying by 10% in each iteration; the business feedback error is the converted value of the operation anomaly report from the business system (such as the number of misjudgments / total number of operations); the dynamic weight is adjusted according to the importance of the business scenario (the weight of high-risk operation scenarios is set to 0.8).

[0098] The intelligent agent optimization formula is: New decision threshold = original threshold + μ × (business feedback matching degree - target matching degree), where μ is the learning rate, which is set to 0.05 to ensure smooth threshold adjustment; the target matching degree is the system's preset benchmark value (e.g., 0.92); the adjustment direction is to increase the threshold if the feedback matching degree is higher than the target, and decrease it if the feedback matching degree is lower.

[0099] 3. Business Application Monitoring Algorithm: Operation Trajectory Tracking Formula: Trajectory Similarity = Σ(Actual Operation Vector · Standard Operation Vector) / (Actual Operation Vector Modulus × Standard Operation Vector Modulus), where the actual operation vector is a time-series vector composed of the sequence of equipment state changes; the standard operation vector is the standard operation sequence specified in the safety ticket; the similarity value range is [-1, 1], and a value below 0.85 triggers manual review.

[0100] Video review confidence calculation: Confidence = α × Image recognition confidence + β × Human review confidence, where α is the initial weight of 0.6, which gradually increases to 0.8 with the number of human reviews; β is the complementary weight (1-α); and the human review confidence is the 0-1 score given by the expert review panel.

[0101] 4. Closed-loop feedback mechanism: Feedback data standardization formula: Standardized feedback value = (Original feedback value - Historical mean) / Historical standard deviation, where the historical mean / standard deviation is calculated based on feedback data from the past 30 days; standardization eliminates differences in dimensions and facilitates cross-scenario comparison.

[0102] Iteration effect evaluation formula: Effect improvement rate = (New model accuracy - Old model accuracy) / Old model accuracy × 100%, where accuracy is the number of correct recognitions / total number of recognitions; The improvement rate threshold is set to 5%, and a second optimization process is triggered if it is lower than this value.

[0103] This invention achieves quantitative management of each link in the closed-loop process. In the data acquisition stage, the quality of source data is ensured through integrity / timeliness verification; in the intelligent analysis stage, model parameters are corrected through business feedback errors; in the business application stage, accurate monitoring is achieved through trajectory tracking and confidence calculation; and in the feedback optimization stage, model iteration is completed through standardized processing and effect evaluation. All parameters are set based on actual power production scenarios, forming a complete "acquisition-analysis-application-optimization" technical closed loop, supporting the efficient operation of the power equipment monitoring and management application system.

Claims

1. A substation safety ticket execution and control terminal device supporting local AI recognition, comprising a safety ticket execution and control terminal and a substation auxiliary terminal, wherein the substation auxiliary terminal establishes a connection with the safety ticket execution and control terminal based on a local AI recognition unit, characterized in that: The security ticket execution and control terminal includes a data acquisition network, an extraction network, a unified video platform, a power equipment monitoring and management application system, and an artificial intelligence platform. The data acquisition network is connected to the unified video platform, and the unified video platform and the extraction network are connected to the power equipment monitoring and management application system. The power equipment monitoring and management application system achieves a closed-loop bidirectional connection with the artificial intelligence platform through an intelligent agent call interface. The local AI recognition unit includes an edge computing unit, an IoT tag reading and writing module, and an alarm module. The substation auxiliary terminal includes a fire information transmission and control unit, a security monitoring unit, and an environmental monitoring unit.

2. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 1, characterized in that: The acquisition network consists of mobile terminals, access points (APs), and aggregation switches, and is used to acquire video and / or image data from field devices and connect them to a unified video platform.

3. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 2, characterized in that: The extraction network consists of a PMS and a security ticket, and is used to extract the security ticket business data from the PMS.

4. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 3, characterized in that: The artificial intelligence platform includes an ORC recognition module, a knowledge base, a state recognition model module, and a state analysis intelligent agent; wherein the ORC recognition module is used to perform OCR recognition on security tickets and store the recognition results in the knowledge base after structured processing. The status recognition model module uses image samples of the device as the training set and is trained and optimized based on the 0.45b Guangming Power visual large model to obtain a proprietary model for accurate device status recognition, thereby achieving high-precision visual recognition of device status.

5. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 1, characterized in that: The power equipment monitoring and management application system also includes a to-do items module, a completed items module, a safety measure ticket issuance module, a safety measure ticket editing module, a safety measure ticket execution module, a safety measure ticket archiving module, an archive query module, a compilation ticket query module, and a work task module; wherein the to-do items module and the completed items module each contain a pending review sub-module and a reviewed sub-module; the safety measure ticket execution module contains an execution sub-module and a recovery sub-module; the safety measure ticket archiving module contains a safety measure pre-archiving check module and a safety measure archiving module.

6. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 1, characterized in that: The IoT tag reading and writing module based on the local AI recognition unit is connected to the artificial intelligence platform, and the alarm module based on the local AI recognition unit is connected to the security monitoring unit of the substation auxiliary terminal.

7. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 6, characterized in that: The fire information transmission control unit is connected to an automatic fire alarm system via RS485 / RJ45 / CAN interface, to a fixed fire extinguishing system and other controlled fire equipment via I / O interface, and to a fire water tank level transmitter, a fire pipeline pressure transmitter, and a fire power supply voltage transmitter via analog interface.

8. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 6, characterized in that: The security monitoring unit is connected to an audible and visual alarm, an infrared dual-technology detector, an infrared beam detector, an emergency alarm button, and an access control controller via an I / O interface, and is connected to an access control controller, an electronic fence, and a burglar alarm controller via an RJ45 / RS485 interface.

9. The substation safety ticket execution and control terminal equipment supporting local AI recognition as described in claim 6, characterized in that: The environmental monitoring unit is connected to a water immersion detector, a water leakage detector, a fan control box, a water pump control box, and a dehumidifier controller via an I / O interface, and to a micro-meteorological sensor, a temperature and humidity sensor, a water level sensor, an SF6 fluoride content sensor, an air conditioning controller, and a lighting controller via an RS485 interface.

10. A method for controlling the execution of safety measures tickets in substations that supports local AI recognition, based on a substation safety measures ticket execution control terminal device that supports local AI recognition as described in any one of claims 1-9, to achieve intelligent monitoring of substations and closed-loop management of safety measures tickets, characterized in that: The control method includes the following steps: Step 1: The acquisition network acquires video and / or image data from field devices via mobile terminals, access points (APs), and computing sticks, and connects them to a unified video platform through an aggregation switch; Step 2: Extract network data and simultaneously extract security ticket business data from PMS; Step 3: The two types of data obtained in Step 1-2 provide basic data sources for the artificial intelligence platform and the power equipment monitoring and management application system, realizing the integration and preliminary management of multi-source heterogeneous data; Step 4: Model training and agent construction on the artificial intelligence platform; Step 5: The power equipment monitoring and management application system calls the interface through intelligent agents, relies on the artificial intelligence platform to carry out production business, and feeds back the results of business execution to the artificial intelligence platform for model iteration, intelligent agent optimization and scenario encapsulation upgrade, forming a closed-loop management of "data collection → intelligent analysis → business application → feedback optimization".

Images