A work ticket decision-making auxiliary method and system

Abstract

The present invention provides a method and system for assisting decision-making in work orders, relating to the technical field of electric power operations. The method comprises: collecting and fusing multi-source data of an electric power operation system to obtain multimodal data of the electric power operation system; obtaining deep semantic features of the multimodal data based on the multimodal data using a large-scale language model, a data feature extraction model, and an image feature extraction model; matching the deep semantic features using a preset knowledge graph to obtain knowledge graph nodes and relationships related to the multimodal data; then, combining with a preset deep reinforcement learning network, determining an initial assistive decision for the electric power operation work order; inputting the initial assistive decision into an electric power operation system model for simulation to obtain an execution effect; and optimizing the initial assistive decision based on the execution effect and the expected execution effect to obtain a final assistive decision for the work order. The present invention improves the accuracy and consistency of decision-making, thereby enhancing the intelligence level of assistive decision-making in work orders.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power operations, and in particular to a work ticket auxiliary decision-making method and system. Background Art

[0002] Work permits are key documents for ensuring work safety in the power industry. They cover information such as work content, time, location, personnel, and safety measures. They serve as the basis for work permits, monitoring, and acceptance checks. They standardize operational procedures, clarify responsibilities, and ensure safety. Therefore, through the rational decisions contained in work permits, risks can be mitigated, safety, and efficiency can be ensured, making them the core of the power operation safety system.

[0003] Traditional work ticketing often requires staff to spend considerable time searching for relevant rules within massive amounts of historical data to formulate decisions. This makes it difficult to deeply mine and accurately utilize historical data. Furthermore, due to their over-reliance on human experience, differences in understanding and judgment between individuals can lead to unstable decision-making results, making accuracy and consistency difficult to guarantee. Summary of the Invention

[0004] The problem solved by the present invention is how to improve the intelligence level of work ticket auxiliary decision-making.

[0005] In order to solve the above problems, the present invention provides a work ticket decision-making auxiliary method and system.

[0006] In a first aspect, the present invention provides a work ticket decision-making assistance method, which is applied to a power distribution network, wherein the power distribution network includes multiple power distribution facilities, and each of the power distribution facilities corresponds to a different power operation system. The method includes: when it is necessary to operate the power distribution facility, multi-source data of the power operation system of the distribution facility is collected and integrated to obtain multimodal data of the power operation system; deep semantic features of the multimodal data are obtained based on the multimodal data through a large-scale language model, a data feature extraction model, and an image feature extraction model; knowledge graph nodes and relationships related to the multimodal data are matched through a preset knowledge graph according to the deep semantic features; based on the knowledge graph nodes and relationships, combined with a preset deep reinforcement learning network, an initial auxiliary decision for the power operation work ticket is determined; the initial auxiliary decision is input into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision; the initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain the final auxiliary decision for the work ticket.

[0007] Optionally, when it is necessary to operate the power distribution facility, multi-source data of the power operation system of the power distribution facility is collected and fused to obtain multimodal data of the power operation system, including: when it is necessary to operate the power distribution facility, text data, sensor data, image data and equipment operation data in the power operation system of the distribution facility are collected, and the text data, the sensor data, the image data and the equipment operation data are used as the multi-source data; the multi-source data is converted into a preset format through data fusion technology, and the multi-source data in a unified format are fused to obtain the multimodal data of the power operation system.

[0008] Optionally, obtaining deep semantic features of the multimodal data based on the multimodal data by using a large-scale language model, a data feature extraction model, and an image feature extraction model includes:

[0009] The multimodal data is input into the large-scale language model, and semantic extraction is performed on the text data in the multimodal data through the large-scale language model to obtain semantic information of the multimodal data; the multimodal data is input into the data feature extraction model, and feature extraction is performed on the sensor data and the device operation data in the multimodal data through the data feature extraction model to obtain data features of the multimodal data; the multimodal data is input into the image feature extraction model, and feature extraction is performed on the image data in the multimodal data through the image feature extraction model to obtain image features of the multimodal data; the semantic information, the data features, and the image features are fused to obtain the deep semantic features.

[0010] Optionally, the matching is performed according to the deep semantic features through a preset knowledge graph to obtain knowledge graph nodes and relationships related to the multimodal data, including: vectorizing the deep semantic features to obtain feature vectors of the deep semantic features; embedding entities and relationships in the preset knowledge graph into a vector space through graph embedding technology to obtain a knowledge graph embedding vector; calculating the cosine similarity between the knowledge graph embedding vector and the feature vector; and determining the knowledge graph nodes and relationships related to the multimodal data based on the cosine similarity.

[0011] Optionally, the initial auxiliary decision for the work ticket of the power operation is determined based on the knowledge graph nodes and relationships in combination with a preset deep reinforcement learning network, including: taking the risk assessment level, safety measure selection and workflow arrangement of the power operation as the action of the deep reinforcement learning network, and taking the knowledge graph nodes and relationships as the state of the deep reinforcement learning network; splicing the embedding vectors of the knowledge graph nodes and relationships with the feature vectors of the work site environment parameters to form a state vector and input it into the deep reinforcement learning network; updating the action according to the state vector selection, and judging whether the update of the action is completed according to a preset reward function; and taking the action after the update as the initial auxiliary decision for the work ticket.

[0012] Optionally, the initial auxiliary decision is input into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision, including: inputting the initial auxiliary decision into the power operation system model, and performing simulation execution according to the risk assessment level, the safety measure selection and the workflow arrangement; when the simulation execution is completed, obtaining the safety indicators, efficiency indicators and standard indicators of the power operation system model after the simulation execution; and using the safety indicators, the efficiency indicators and the standard indicators as the execution effect.

[0013] Optionally, the initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain the final auxiliary decision of the work ticket, including: judging whether the initial auxiliary decision meets the job requirements by comparing the expected execution effect with the execution effect; when the initial auxiliary decision meets the job requirements, the initial auxiliary decision is used as the final auxiliary decision; when the initial auxiliary decision does not meet the job requirements, the initial auxiliary decision is optimized according to the comparison result of the expected execution effect and the execution effect, and the optimized initial auxiliary decision is used as the final auxiliary decision.

[0014] Optionally, the determining whether the initial auxiliary decision meets the operational requirements by comparing the expected execution effect with the execution effect includes: determining whether the safety indicators, the efficiency indicators and the standard indicators all meet the corresponding expected execution effects; if the safety indicators, the efficiency indicators and the standard indicators all meet the corresponding expected execution effects, then determining that the initial auxiliary decision meets the operational requirements; if any one of the safety indicators, the efficiency indicators and the standard indicators does not meet the corresponding expected execution effects, then determining that the initial auxiliary decision does not meet the operational requirements.

[0015] Optionally, the initial auxiliary decision is optimized based on the comparison result between the expected execution effect and the execution effect, including: when the initial auxiliary decision does not meet the job requirements, according to the indicators in the execution effect that do not meet the expected execution effect, determining the indicator gap between the indicator and the expected execution effect corresponding to the indicator, and using the indicator gap as the comparison result; based on the indicator gap, the initial auxiliary decision is modified in combination with preset adjustment rules to obtain the final auxiliary decision.

[0016] In a second aspect, the present invention provides a work ticket decision-making assistance system, which is applied to a power distribution network, wherein the power distribution network includes multiple power distribution facilities, and each of the power distribution facilities corresponds to a different power operation system. The system includes: a data processing unit, which is used to collect and fuse multi-source data of the power operation system of the distribution facility when it is necessary to operate the distribution facility, so as to obtain multimodal data of the power operation system; a feature extraction unit, which is used to obtain deep semantic features of the multimodal data based on the multimodal data through a large-scale language model, a data feature extraction model, and an image feature extraction model; a matching unit, which is used to match the deep semantic features through a preset knowledge graph to obtain knowledge graph nodes and relationships related to the multimodal data; an initial decision unit, which is used to determine the initial auxiliary decision of the power operation work ticket based on the knowledge graph nodes and relationships in combination with a preset deep reinforcement learning network; a simulation unit, which is used to input the initial auxiliary decision into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision; and an optimization unit, which is used to optimize the initial auxiliary decision based on the execution effect and the expected execution effect to obtain the final auxiliary decision of the work ticket.

[0017] The present work ticket decision-making support method and system, based on the fact that a power distribution network includes multiple distribution facilities, each with its own corresponding power operation system, collects and integrates multi-source data from these diverse systems to ensure that work tickets are formulated with comprehensive consideration of the operating status and operational requirements of all relevant facilities. For example, data on electrical parameters, equipment status, and operating environment is collected for each facility, including substations, switchgear, distribution lines, and distribution transformers, to avoid missing critical information.

[0018] By collecting and fusing multi-source data from power operation systems, multimodal data can be generated. This data provides a more complete picture of the on-site situation, providing comprehensive and rich foundational information for subsequent analysis and decision-making. This addresses the problem of traditional work order decision-making, which relies on manual experience and limited data, making it difficult to deeply mine and accurately utilize historical data. Furthermore, multimodal data (such as text, images, and time series data) provided by different power operation systems can complement and verify each other, thereby improving the accuracy of the assessment of the actual status of distribution facilities. For example, by combining image recognition technology to detect abnormalities in the appearance of equipment (such as broken insulators), combined with analysis of operating parameters (such as current and voltage), it is possible to more accurately determine whether equipment needs repair or replacement.

[0019] Furthermore, different distribution facilities within the distribution network (such as substations, switchgear, overhead lines, and cable lines) have distinct functions and operational requirements. By collecting and analyzing data from each facility's power operation system, the personalized operational needs of different types of facilities can be met. For example, substation operations may involve complex relay protection and automation systems, while distribution line operations focus more on line inspection and fault location. Power operation systems for different distribution facilities can adapt to the complex and changing distribution network environment. For example, the configuration and operating conditions of distribution facilities vary significantly in urban cores, industrial areas, and rural areas. By configuring an independent power operation system for each facility, operational challenges in diverse environments can be better addressed, improving the applicability and effectiveness of work permits.

[0020] The system then integrates a large-scale language model, a data feature extraction model, and an image feature extraction model to derive deep semantic features from the multimodal data. This allows the system to deeply understand the underlying meaning of the data, rather than merely superficially analyzing it like traditional methods. This provides deeper feature support for precise decision-making. The system then matches these deep semantic features against a pre-defined knowledge graph to identify knowledge graph nodes and relationships relevant to the multimodal data. Matching these deep semantic features with the knowledge graph connects the current worksite situation with existing expertise, providing context for subsequent decision-making. The knowledge graph provides a unified and standardized knowledge system, resolving the issue of unstable decision-making caused by traditional methods, which rely on manual experience and experience, resulting in differences in understanding and judgment among different personnel. Based on the knowledge graph nodes and relationships, the system combines a pre-defined deep reinforcement learning network to determine the initial decision support for the power operation work order. This deep reinforcement learning network learns the optimal decision-making strategy based on the current state (i.e., the matched knowledge graph nodes and relationships) and a pre-defined reward mechanism. It can select a relatively reasonable initial decision support strategy from a wide range of possible solutions, generating preliminary solutions more quickly and intelligently than traditional manual decision-making methods. The initial auxiliary decision is further input into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision. Through simulation operations, the operation of the decision in a virtual environment can be observed before the actual operation, potential problems can be discovered in advance, and the feasibility of the decision can be evaluated; this helps to avoid trial and error directly in actual operations and reduce risks. Finally, the initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain the final auxiliary decision of the work ticket. This process of the present invention utilizes a feedback mechanism to adjust the initial decision to make it more in line with actual needs and expected goals, thereby improving the accuracy and consistency of the decision, solving the problem of unstable decision results in traditional methods, and realizing an improvement in the intelligence level of auxiliary decision-making for work tickets. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of a work ticket decision-making assistance method according to an embodiment of the present invention;

[0022] Figure 2 Schematic diagram of the structure of a power distribution network according to an embodiment of the present invention;

[0023] Figure 3 This is a structural block diagram of the work ticket decision-making auxiliary system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0025] Combine Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a work ticket auxiliary decision-making method, which is applied to a power distribution network. The power distribution network includes multiple power distribution facilities, and each of the power distribution facilities corresponds to a different power operation system.

[0026] Specifically, in a power distribution network, distribution infrastructure primarily includes various substations, switchgear, distribution lines (such as overhead lines and cable lines), distribution transformers, circuit breakers, disconnectors, and other equipment. These facilities form the complete path for power transmission from the power generation point to the user end. For example, a city's distribution network may have multiple 35kV substations. These substations receive power from higher-voltage transmission lines (such as 110kV or 220kV). Transformers then reduce the voltage to approximately 10kV, which is then transmitted to various user areas via 10kV distribution lines. Within the user areas, switchgear further distributes the power, and distribution transformers convert the 10kV voltage to 380V or 220V for general consumption.

[0027] In a preferred embodiment of the present invention, Figure 2 As shown in the figure, the power operation systems corresponding to different distribution facilities, such as the substation power operation system, are complex and integrated systems. These include an electrical equipment monitoring system, which monitors the operating parameters of substation equipment such as transformers, circuit breakers, and busbars in real time, such as voltage, current, active power, reactive power, and temperature. For example, through various sensors installed on the transformer, the monitoring system can obtain data such as transformer winding temperature and oil temperature, which are crucial for determining whether the transformer is operating properly. The relay protection system, when a fault (such as a short circuit) occurs within the substation, quickly activates relay protection devices to disconnect the faulty equipment and prevent further escalation. Its configuration varies depending on factors such as the substation wiring method and equipment type. For example, in a substation with a dual-busbar connection, the relay protection system must consider busbar segmentation and configure appropriate bus differential protection and line protection. The automation system provides remote control and automated operation of substation equipment, such as remote control of circuit breaker opening and closing operations.

[0028] Another example is the power operation system of a switchgear station. The switchgear station is mainly used for the distribution of electric energy. The power operation system mainly includes a switchgear control system, which is used to control the opening and closing operations of each incoming and outgoing line switch to achieve the distribution of electric energy and line switching. For example, when a 10kV distribution line needs to be repaired, the switchgear station disconnects the maintenance line by controlling the corresponding switch, and adjusts the power supply method through other switches to ensure the normal power supply to other users. At the same time, it also includes a corresponding monitoring system for monitoring the status of the switchgear in the switchgear station, such as the opening and closing position of the switch, the energy storage status, etc., as well as basic electrical parameters such as busbar voltage, to ensure the normal operation of the switchgear station.

[0029] Or, for distribution line power operation systems, for example, overhead lines. These systems include line inspection systems, which inspect conductors, insulators, towers, and other equipment through manual or drone inspections. For example, they check for damage to conductors, breakage or contamination discharge on insulators, and tilted towers. They also include line fault location systems, which quickly locate the fault point when a line fault (such as a ground fault or short circuit) occurs. This is achieved by installing a fault indicator on the line. The fault indicator emits a signal when it detects a fault current, allowing personnel to locate the fault point and repair it.

[0030] This also includes, for example, the power operation system for distribution transformers. This system includes a transformer monitoring system, which primarily monitors operating parameters such as oil temperature, winding temperature, and load current. For example, for oil-immersed distribution transformers, oil temperature is a critical parameter; excessively high oil temperatures can lead to problems such as insulation degradation. It also includes protection systems for overload and temperature protection. When the transformer load exceeds a certain percentage of the rated capacity or the temperature is excessively high, the protection device will issue an alarm or automatically cut off the power supply to protect the transformer.

[0031] The method comprises:

[0032] When it is necessary to operate the power distribution facility, multi-source data of the power operation system of the power distribution facility are collected and fused to obtain multi-modal data of the power operation system.

[0033] Specifically, in power operation systems, diverse multi-source data exists, including but not limited to equipment operating parameters (such as voltage, current, and power), operator behavior data (such as operating movements and movement trajectories), on-site meteorological data (such as temperature, humidity, and wind speed), and equipment image or video data. This multi-source data is collected through sensors, monitoring equipment, and personnel records. The collected data, in various formats and types, is then fused to form a unified, coherent multimodal dataset. For example, equipment operating data can be combined with corresponding image data to more comprehensively reflect the actual status of the equipment on-site. Multimodal data is generated by collecting and fusing multi-source data from multiple channels in power operation systems. Fusion technology then integrates this data into a single, consistent multimodal data format, providing comprehensive foundational information for subsequent analysis and decision-making. This overcomes the problem of traditional work order decision-making relying on manual experience and limited data, providing comprehensive and rich foundational information for subsequent analysis.

[0034] Through a large-scale language model, a data feature extraction model and an image feature extraction model, deep semantic features of the multimodal data are obtained according to the multimodal data.

[0035] Specifically, first, large-scale language models are used to analyze text-related data (such as equipment fault descriptions and work requirements). By learning from large amounts of text corpora, large-scale language models can understand the semantic information in text. For example, they can identify semantic features such as the type of equipment fault described in the text and the urgency of the work requirements. Second, for structured data (such as equipment operating parameters and meteorological data), data feature extraction models are used. These models can extract key data features, such as the mean and variance of equipment operating parameters through statistical analysis, or identify abnormal changes in meteorological data. Third, for image or video data, image feature extraction models are used. These models can identify visual features such as objects (such as equipment components and tools), scenes (such as indoor substations and outdoor transmission lines), and object status (such as whether the equipment is damaged). By combining the features extracted from various components of multimodal data using these three models, deep semantic features of the multimodal data are ultimately derived. These deep semantic features can more deeply and accurately describe the actual situation of the power operation system, providing an information-rich basis for subsequent decision-making.

[0036] Multimodal data is processed using large-scale language models, data feature extraction models, and image feature extraction models. The large-scale language model understands the semantic information in text descriptions, the data feature extraction model extracts key features from structured data, and the image feature extraction model analyzes visual features in live images or videos. By combining these three models, the deep semantic features of multimodal data are extracted, enabling the system to deeply understand the meaning behind the data, rather than just superficially, providing deeper feature support for precise decision-making.

[0037] By matching the preset knowledge graph according to the deep semantic features, knowledge graph nodes and relationships related to the multimodal data are obtained.

[0038] Specifically, the pre-built knowledge graph contains a wealth of knowledge in the field of power operations. Nodes represent various entities (such as equipment type, failure mode, and safety measures), while relationships represent the connections between these entities (for example, a certain equipment type is prone to a certain failure mode, and a certain failure mode requires specific safety measures). The extracted deep semantic features are then matched against the knowledge graph.

[0039] For example, if the deep semantic features contain a specific fault description of a certain device, then the corresponding device fault node and related fault handling methods, associated equipment and other nodes can be found in the knowledge graph. At the same time, the relationship between these nodes can also be obtained (such as the fault handling process, the electrical connection relationship between devices, etc.).

[0040] By matching the extracted deep semantic features with the pre-set knowledge graph, the knowledge graph nodes and relationships related to the multimodal data are found. The knowledge graph integrates a wealth of knowledge and experience related to power operations, such as the normal operating parameter ranges of equipment and the safety measures corresponding to different operation types. Matching deep semantic features with the knowledge graph connects the current work site situation with existing professional knowledge, providing the knowledge background for subsequent decision-making. This overcomes the problem of unstable decision-making caused by traditional methods due to reliance on human experience and differences in understanding and judgment among different personnel, and instead utilizes a unified and standardized knowledge system to assist decision-making.

[0041] Based on the knowledge graph nodes and relationships, combined with a preset deep reinforcement learning network, the initial auxiliary decision for the work ticket of the power operation is determined.

[0042] Specifically, the deep reinforcement learning network selects the optimal action strategy based on the current state (i.e., the on-site situation reflected by the nodes and relationships obtained from the knowledge graph). In this scenario, the deep reinforcement learning network generates a reasonable initial work order based on the knowledge provided by the knowledge graph and the actual on-site situation to assist in decision-making.

[0043] For example, the nodes and relationships in the knowledge graph indicate that there's equipment failure on-site and that repair work is required. Based on a pre-defined reward mechanism (e.g., rewards for ensuring safety, rewards for improving efficiency), the deep reinforcement learning network considers various possible work options (e.g., varying power outage scopes, different safety measures deployment methods), and selects a relatively optimal initial decision-making support plan. This plan includes suggested work content, work schedules, personnel division of labor, and safety measures.

[0044] Based on the knowledge graph nodes and relationships, combined with a pre-set deep reinforcement learning network, the initial auxiliary decision for the power operation work order is determined. The deep reinforcement learning network learns the optimal decision-making strategy based on the current state (i.e., the matched knowledge graph nodes and relationships) and the pre-set reward mechanism. It then selects a relatively reasonable initial auxiliary decision from a wide range of possible decision options. Compared to traditional manual decision-making methods, this method generates preliminary solutions more quickly and intelligently.

[0045] The initial auxiliary decision is input into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision.

[0046] Specifically, a power operation system model was constructed that simulated the entire power operation process, including changes in equipment operating status, operator behavior, and the implementation of safety measures. Initial support decisions were input into this model, and the model simulated the operation content, process, and safety measures specified in the decision. Through simulation, the effectiveness of the initial support decision in a virtual environment was determined.

[0047] For example, we can observe whether the equipment failure was successfully eliminated in the simulation, whether the operation was completed within the specified time, whether the safety measures effectively prevented the occurrence of accidents, and other execution effect information. This information can help evaluate the feasibility and effectiveness of the initial auxiliary decision.

[0048] The initial support decision is input into the power operation system model and simulated to determine the corresponding execution results. This simulation allows for observing the execution of the decision in a virtual environment before actual operation, identifying potential issues in advance and assessing the feasibility of the decision. This helps avoid trial and error in actual operations and reduces risk.

[0049] The initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain a final auxiliary decision for the work ticket.

[0050] Specifically, the simulated execution results are compared with the pre-set expected execution results. The expected execution results include aspects such as the completion time of the operation meeting the planned requirements, the operation process being safe and accident-free, and the equipment returning to normal operation.

[0051] If the initial decision-making support deviates from the expected outcome, such as when the simulation reveals an operation taking too long or presenting safety hazards, the initial decision-making support needs to be optimized and adjusted. By incorporating a feedback mechanism, such as adjusting relevant decision-making parameters based on the degree of deviation (for example, increasing the number of personnel or changing the sequence of work steps), the simulation is re-evaluated until the execution effect reaches or approaches the expected outcome. The resulting work ticket-based decision-making support will be more reasonable and reliable, helping to ensure the safety and efficiency of power operations.

[0052] The present work ticket decision-making support method and system, based on the fact that a power distribution network includes multiple distribution facilities, each with its own corresponding power operation system, collects and integrates multi-source data from these diverse systems to ensure that work tickets are formulated with comprehensive consideration of the operating status and operational requirements of all relevant facilities. For example, data on electrical parameters, equipment status, and operating environment is collected for each facility, including substations, switchgear, distribution lines, and distribution transformers, to avoid missing critical information.

[0053] By collecting and fusing multi-source data from power operation systems, multimodal data can be generated. This data provides a more complete picture of the on-site situation, providing comprehensive and rich foundational information for subsequent analysis and decision-making. This addresses the problem of traditional work order decision-making, which relies on manual experience and limited data, making it difficult to deeply mine and accurately utilize historical data. Furthermore, multimodal data (such as text, images, and time series data) provided by different power operation systems can complement and verify each other, thereby improving the accuracy of the assessment of the actual status of distribution facilities. For example, by combining image recognition technology to detect abnormalities in the appearance of equipment (such as broken insulators), combined with analysis of operating parameters (such as current and voltage), it is possible to more accurately determine whether equipment needs repair or replacement.

[0054] Furthermore, different distribution facilities within the distribution network (such as substations, switchgear, overhead lines, and cable lines) have distinct functions and operational requirements. By collecting and analyzing data from each facility's power operation system, the personalized operational needs of different types of facilities can be met. For example, substation operations may involve complex relay protection and automation systems, while distribution line operations focus more on line inspection and fault location. Power operation systems for different distribution facilities can adapt to the complex and changing distribution network environment. For example, the configuration and operating conditions of distribution facilities vary significantly in urban cores, industrial areas, and rural areas. By configuring an independent power operation system for each facility, operational challenges in diverse environments can be better addressed, improving the applicability and effectiveness of work permits.

[0055] The system then integrates a large-scale language model, a data feature extraction model, and an image feature extraction model to derive deep semantic features from the multimodal data. This allows the system to deeply understand the underlying meaning of the data, rather than merely superficially analyzing it like traditional methods. This provides deeper feature support for precise decision-making. The system then matches these deep semantic features against a pre-defined knowledge graph to identify knowledge graph nodes and relationships relevant to the multimodal data. Matching these deep semantic features with the knowledge graph connects the current worksite situation with existing expertise, providing context for subsequent decision-making. The knowledge graph provides a unified and standardized knowledge system, resolving the issue of unstable decision-making caused by traditional methods, which rely on manual experience and experience, resulting in differences in understanding and judgment among different personnel. Based on the knowledge graph nodes and relationships, the system combines a pre-defined deep reinforcement learning network to determine the initial decision support for the power operation work order. This deep reinforcement learning network learns the optimal decision-making strategy based on the current state (i.e., the matched knowledge graph nodes and relationships) and a pre-defined reward mechanism. It can select a relatively reasonable initial decision support strategy from a wide range of possible solutions, generating preliminary solutions more quickly and intelligently than traditional manual decision-making methods. The initial auxiliary decision is further input into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision. Through simulation operations, the operation of the decision in a virtual environment can be observed before the actual operation, potential problems can be discovered in advance, and the feasibility of the decision can be evaluated; this helps to avoid trial and error directly in actual operations and reduce risks. Finally, the initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain the final auxiliary decision of the work ticket. This process of the present invention utilizes a feedback mechanism to adjust the initial decision to make it more in line with actual needs and expected goals, thereby improving the accuracy and consistency of the decision, solving the problem of unstable decision results in traditional methods, and realizing an improvement in the intelligence level of auxiliary decision-making for work tickets.

[0056] Optionally, when it is necessary to operate the power distribution facility, multi-source data of the power operation system of the power distribution facility is collected and fused to obtain multimodal data of the power operation system, including: when it is necessary to operate the power distribution facility, text data, sensor data, image data and equipment operation data in the power operation system of the distribution facility are collected, and the text data, the sensor data, the image data and the equipment operation data are used as the multi-source data; the multi-source data is converted into a preset format through data fusion technology, and the multi-source data in a unified format are fused to obtain the multimodal data of the power operation system.

[0057] Specifically, in this optional embodiment, text data typically comes from worksite logs, equipment maintenance records, accident reports, and other documents. These documents can be manually entered or extracted from paper documents using optical character recognition (OCR) technology. For example, a work log might record textual descriptions of the equipment's daily operating status, minor faults, and corresponding handling measures. Sensor data is collected in real time by various sensors installed on power equipment. For example, current transformers and voltage transformers monitor current and voltage changes in circuits, respectively; temperature sensors monitor the temperature of key equipment components; and pressure sensors monitor oil or air pressure. These sensors convert physical quantities into electrical signals, which are then converted to digital data via analog-to-digital converters. Image data primarily comes from camera monitoring at the worksite and visual inspection images of the equipment. Cameras can be fixed devices used for real-time monitoring of the work area or portable cameras carried by workers to record the equipment's appearance and operator behavior. Equipment operating data includes equipment start / stop times, operating hours, load conditions, and downtime. This data is automatically recorded by the equipment's control system or monitoring system and stored as structured data, such as tables in a database.

[0058] Converting collected multi-source data into a unified, pre-defined format involves standardizing different types of data. Specifically, text data is converted to a unified text encoding format (such as UTF-8) and organized into structured text files or text fields in a database. For example, information such as dates, event descriptions, and action taken in work logs can be stored in separate database fields for later processing. Sensor data, typically time series data, is converted to a unified timestamp format and its dimensions and units are normalized. For example, all temperature data can be converted to Celsius and timestamps can be standardized in seconds. Image data is preprocessed, such as image resizing and color space conversion, to conform to a unified image format. For example, all images can be resized to the RGB color space and have a standardized resolution. Equipment operational data is converted from its original recording format (which may be a device-specific binary file or proprietary data format) to a common structured data format, such as CSV (Comma Separated Values) or JSON (JavaScript Object Notation). After converting the multi-source data into a unified format, data fusion techniques are used to integrate the data to generate multimodal data.

[0059] In one embodiment, data fusion can be aligned based on the time dimension. For example, based on time, the events described by text data, the physical quantity values ​​monitored by sensor data, the on-site scenes captured by image data, and the operating status recorded by equipment operation data within the same time point or time period are associated. In this way, a multimodal data set that comprehensively reflects the work site situation can be obtained, in which the data at each time point contains complementary information from different data sources. In another embodiment, data fusion can be based on the alignment of the spatial dimension. For example, sensor data from different locations are combined with the image data of the location to determine whether the operating status and appearance of a specific part of the equipment are consistent. If the temperature sensor of a certain part of the equipment shows an abnormally high temperature, and the image data of the part shows smoke or discoloration, then the fusion of the spatial dimension can more accurately determine the possible failure of the equipment.

[0060] In this optional embodiment, by collecting multi-source data such as text data, sensor data, image data and equipment operation data of the power operation system, and converting them into a unified format using data fusion technology and then fusing them, multimodal data is generated, which significantly improves the integrity and accuracy of the data. This comprehensive data collection method ensures that subsequent decisions can be made based on rich and multi-dimensional information, avoiding the one-sidedness of decisions caused by missing data. At the same time, the unified format conversion eliminates the format differences between different data sources, facilitates subsequent data processing and analysis, and improves the efficiency and accuracy of data processing. More importantly, multimodal data fusion enhances the correlation between different data, so that events described in text, changes in physical quantities monitored by sensor data, on-site images captured by image data, and operating status recorded by equipment operation data can verify each other, more accurately reflecting the actual situation of the power operation system, and providing a solid data foundation for auxiliary decision-making of work tickets.

[0061] The multi-source data collection and fusion process of this embodiment provides strong support for data-driven work ticket decision-making, significantly improving the efficiency and quality of work ticket decision-making. In traditional methods, staff need to spend a lot of time searching and organizing information in different formats and scattered data sources, and it is easy for incomplete or inaccurate data to lead to low decision-making efficiency and uneven quality. The multimodal data after collection and fusion can be directly used by the decision-making system, reducing the workload of manual search and organization of data. At the same time, due to the improvement of data integrity and accuracy, the decision-making system can more accurately identify potential risks and patterns, thereby generating more reasonable and reliable work ticket auxiliary decision-making solutions, effectively ensuring the safety and efficiency of power operations.

[0062] Optionally, obtaining deep semantic features of the multimodal data based on the multimodal data by using a large-scale language model, a data feature extraction model, and an image feature extraction model includes:

[0063] The multimodal data is input into the large-scale language model, and semantic extraction is performed on the text data in the multimodal data through the large-scale language model to obtain semantic information of the multimodal data; the multimodal data is input into the data feature extraction model, and feature extraction is performed on the sensor data and the device operation data in the multimodal data through the data feature extraction model to obtain data features of the multimodal data; the multimodal data is input into the image feature extraction model, and feature extraction is performed on the image data in the multimodal data through the image feature extraction model to obtain image features of the multimodal data; the semantic information, the data features, and the image features are fused to obtain the deep semantic features.

[0064] Specifically, first, the collected text data is preprocessed, including removing noise (such as meaningless symbols, blank characters, etc.), word segmentation (dividing continuous text into a sequence of words or phrases), and part-of-speech tagging (marking each word with its part of speech, such as noun, verb, etc.).

[0065] For example, for the text description in the work log "The main transformer had an overload alarm at 08:30 and lasted for 15 minutes", it is divided into words such as "main transformer", "at", "08:30", "appeared", "overload alarm", "duration", "for", "15 minutes" after preprocessing, and the parts of speech are marked. The preprocessed text data is input into the large-scale language model. Among them, the large-scale language model of this embodiment is based on the Transformer architecture, which includes a multi-layer self-attention mechanism and a feedforward neural network. The self-attention mechanism can capture the correlation between different words in the text, no matter how far apart they are in the text. For example, in the above text, there is an important semantic association between "main transformer" and "overload alarm", and the self-attention mechanism can calculate the correlation weight between them, so that the model can understand that "main transformer" is the object of "overload alarm".

[0066] By performing a pooling operation at the output layer of a large-scale language model (e.g., taking the output vector of the first token in a text sequence or average pooling the output vectors of all tokens), a fixed-dimensional vector representation is obtained. This vector represents the semantic information of the text data. This vector captures key semantic features in the text data, such as the type of event described (equipment failure, maintenance operation, etc.), the equipment involved, and the severity of the event. Sensor data and equipment operation data are first time-aligned to ensure a unified timestamp format. The data is then normalized to scale the values ​​to a specific range (e.g., 0 to 1) to eliminate the impact of different dimensions and ranges on feature extraction. For example, temperature sensor data (measured in degrees Celsius, ranging from -50°C to 100°C) and current sensor data (measured in amperes, ranging from 0A to 1000A) are converted to the range of 0 to 1 using normalization formulas (e.g., min-max normalization).

[0067] The data feature extraction model can use recurrent neural network architectures such as long short-term memory (LSTM) or gated recurrent unit (GRU) to process time series data. In this optional embodiment, taking LSTM as an example, LSTM comprises three structures: an input gate, a forget gate, and an output gate. The input gate controls how much of the current input information is written into the cell state; the forget gate determines how much of the cell state from the previous moment is retained; and the output gate determines how much of the current cell state is output. In this way, LSTM can learn long-term dependencies in time series data. For sensor data and device operation data, LSTM can extract features such as data trends, periodic patterns, and anomalies. For example, for device current monitoring data, LSTM can capture the rising current trend during device startup, the stable current value during normal operation, and possible abnormal current surges.

[0068] The collected image data undergoes preprocessing, including image resizing (e.g., scaling to a uniform 224×224 pixel size), color space conversion (e.g., from RGB to grayscale), or other color enhancement to highlight specific features, as well as data augmentation (e.g., random rotation and flipping) to improve model generalization. The image feature extraction model utilizes a convolutional neural network (CNN) architecture. First, the image is convolved using a convolutional layer, where multiple convolution kernels slide across the image to extract local features (e.g., edges and textures). For example, a 3×3 convolution kernel sliding across the image can detect edge features. An activation function (e.g., ReLU) is then used to introduce nonlinearity, enabling the model to learn more complex feature representations. The feature map is then downsampled using a pooling layer (e.g., max pooling) to reduce the data size and extract key features. After multiple layers of convolution and pooling, the resulting feature map represents high-level semantic features of the image, such as the device's shape, signs of damage (e.g., cracks, deformation), and the relative positions of its components.

[0069] The semantic information vector extracted by the large-scale language model, the data feature vector obtained by the data feature extraction model, and the image feature vector obtained by the image feature extraction model are fused. A concatenation fusion method can be used to concatenate the three vectors end-to-end to form a longer fused vector. For example, assuming the dimension of the semantic information vector is 512, the dimension of the data feature vector is 256, and the dimension of the image feature vector is 1024, the dimension of the concatenated fused vector is 512 + 256 + 1024 = 1792. Furthermore, to better utilize the fused features for subsequent tasks such as knowledge graph matching and decision-making, the fused vector can be input into a fully connected neural network layer for further feature integration and dimensionality reduction. Furthermore, normalization (such as L2 normalization) can be applied to ensure that the fused feature vector has unit length, facilitating subsequent operations such as similarity calculation. This process yields a deep semantic feature vector for multimodal data. This vector comprehensively represents various information about the power operation system, including the semantics of textual descriptions, the characteristics of equipment operation and environmental monitoring data, and the visual features of images, providing powerful feature support for subsequent work ticket decision-making.

[0070] In this optional embodiment, deep semantic features of multimodal data are extracted through large-scale language models, data feature extraction models, and image feature extraction models, thereby achieving targeted processing of different types of data. The large-scale language model can deeply explore the semantic information in text data and accurately capture key semantic points such as events, objects, and their associations described in the text; the data feature extraction model focuses on extracting data features from sensor data and equipment operation data, and grasps the laws of data changes, periodicity, and abnormal conditions; the image feature extraction model can accurately extract visual features from image data and identify important information such as objects, scenes, and states in the image. On this basis, the three features are fused to integrate the semantics of the text, the laws of the data, and the visual content of the image, forming a comprehensive, in-depth, and unified feature description of the power operation system, namely the deep semantic features, which provide a high-quality, information-rich feature foundation for subsequent knowledge graph matching and decision support.

[0071] This optional embodiment significantly improves the efficiency of understanding and utilizing multimodal data. On the one hand, professional models are used for feature extraction for text, data, and images respectively, ensuring the accuracy and professionalism of feature extraction for each type of data, and giving full play to the information value contained in each type of data; on the other hand, the feature fusion process breaks down the barriers between data types, integrating scattered features into unified deep semantic features, so that subsequent processing can take into account the comprehensive information conveyed by text, data, and images at the same time, avoiding the information fragmentation and incompleteness caused by processing each type of data separately, thereby providing a more accurate and comprehensive basis for work ticket auxiliary decision-making, which helps to improve the scientificity and reliability of decision-making.

[0072] Optionally, the matching is performed according to the deep semantic features through a preset knowledge graph to obtain knowledge graph nodes and relationships related to the multimodal data, including: vectorizing the deep semantic features to obtain feature vectors of the deep semantic features; embedding entities and relationships in the preset knowledge graph into a vector space through graph embedding technology to obtain a knowledge graph embedding vector; calculating the cosine similarity between the knowledge graph embedding vector and the feature vector; and determining the knowledge graph nodes and relationships related to the multimodal data based on the cosine similarity.

[0073] Specifically, the fused deep semantic feature vector is normalized to a length of 1 to facilitate subsequent similarity calculations. For example, L2 normalization is used to calculate the L2 norm of the vector (i.e., the square root of the sum of the squares of the vector elements), and then each element is divided by this norm to obtain the normalized feature vector.

[0074] Assume that the deep semantic feature vector is , whose L2 norm is , the normalized vector is ;

[0075] A normalized deep semantic feature vector is obtained, which contains the key information of the multimodal data of the power operation system, and each dimension corresponds to a specific semantic or feature meaning. For example, a certain dimension may reflect the severity of the equipment failure, and another dimension may reflect the stability of the equipment's operating status. The preset knowledge graph includes various entities of the power operation system (such as equipment type, failure mode, safety measures, operation process, etc.) and the relationships between entities (such as "cause", "need", "belong to", etc.). For example, there is a "cause" relationship between the entities "circuit breaker" and "overload failure", and there is a "need" relationship between the entities "overload failure" and "increase cooling". The TransE algorithm is used for graph embedding. Initialize each entity and relationship as a random vector, and then optimize the objective function through training so that the embedding vector of the correct triple (head entity, relationship, tail entity) in the knowledge graph meets certain distance constraints. For example, for triples such as "circuit breaker", "cause", and "overload failure", after training, the head entity vector plus the relationship vector is approximately equal to the tail entity vector, that is, .

[0076] In this way, the entities and relationships in the knowledge graph are embedded into a low-dimensional vector space to obtain a set of knowledge graph embedding vectors. For each entity and relationship embedding vector in the knowledge graph, the cosine similarity is calculated with the deep semantic feature vector.

[0077] Among them, the calculation formula of cosine similarity is: , where a is the knowledge graph embedding vector and b is the deep semantic feature vector.

[0078] Since both are normalized to a denominator of 1, the calculation can be simplified to a vector dot product. For example, if the dot product between the embedding vector of the "overload fault" entity in the knowledge graph and the deep semantic feature vector is 0.85, it indicates a high degree of similarity between the two. A similarity matrix is ​​constructed, with rows representing entities and relationships in the knowledge graph and columns representing the deep semantic features to be matched. The matrix elements are the corresponding cosine similarity values, providing a quantitative basis for subsequently identifying relevant knowledge graph nodes and relationships. A cosine similarity threshold, such as 0.7, is then set based on practical needs and experience. Entities and relationships in the similarity matrix that exceed this threshold are identified as knowledge graph nodes and relationships relevant to the multimodal data. Nodes and relationships with similarities above the threshold are sorted from high to low by similarity, and the top K (e.g., top 5) are selected as the final matching results. These results represent the most relevant knowledge graph information for the current multimodal data of the power operation system, including specific entities and their relationships. This provides expert knowledge support for subsequent work ticket decision-making.

[0079] In this optional embodiment, the preset knowledge graph specifically includes various types of equipment used in power operations and their parameters, operating status, and location information, such as the rated voltage, current, power, and operating temperature range of equipment such as transformers and circuit breakers; includes possible fault types, phenomena, causes, and impacts, such as short circuits, grounding, overload faults, and the resulting equipment damage and power supply impacts; covers operator safety measures, such as safety tools, operating specifications, personal protective equipment, and safe distances; clarifies specific operating procedures, such as equipment inspection, fault location, maintenance operations, and test verification steps, sequence, and related systems; and also involves environmental parameters, such as meteorological conditions and workplace conditions.

[0080] For example, a transformer may experience an overload failure due to overload, while a circuit breaker may experience a ground fault due to insulation damage. A short circuit requires an emergency power outage and the installation of multiple grounding wires. A ground fault requires the display of warning signs and inspections performed with insulating boots. Overload prevention measures include adding cooling equipment and regularly checking equipment loads. Furthermore, equipment inspections include wearing a hard hat and arc-flash protection clothing, while maintenance procedures include the installation of safety fences and the display of warning signs. Grounding wire installation occurs after power outages, while removal occurs after the work is completed and inspected. Equipment inspections target faulty equipment such as transformers and circuit breakers, while testing and verification operations target repaired faulty equipment. Maintenance operations can change the equipment status from "faulty" to "normal," while power outages change the equipment status from "operating" to "outage." Environmental impacts on equipment include: high temperatures increase the probability of overheating, while humid environments increase the risk of leakage.

[0081] In this optional embodiment, through such a knowledge graph, it is possible to achieve accurate matching of the deep semantic features of multimodal data, provide rich background knowledge and reasoning basis for the work ticket auxiliary decision-making system, thereby effectively solving the problem of intelligent work ticket formulation in power operations and improving the scientificity and safety of decision-making. Specifically, by matching the preset knowledge graph according to the deep semantic features, the knowledge graph nodes and relationships related to the multimodal data are obtained, and the effective docking of the deep semantic features and the knowledge graph is achieved. The deep semantic features are vectorized and can be converted into a mathematical representation that is isomorphic to the knowledge graph embedding vector, making the two comparable. At the same time, the entities and relationships in the knowledge graph are embedded in the vector space using graph embedding technology, which fully retains the topological structure and semantic information of the knowledge graph, laying the foundation for subsequent similarity calculations. The method of calculating cosine similarity can accurately measure the semantic similarity between feature vectors and knowledge graph embedding vectors. The results are intuitive and easy to interpret. By setting a reasonable similarity threshold and sorting and filtering mechanism, the knowledge graph nodes and relationships related to multimodal data can be accurately determined, effectively filtering out irrelevant information, and improving the accuracy and pertinence of matching.

[0082] Optionally, the initial auxiliary decision for the work ticket of the power operation is determined based on the knowledge graph nodes and relationships in combination with a preset deep reinforcement learning network, including: taking the risk assessment level, safety measure selection and workflow arrangement of the power operation as the action of the deep reinforcement learning network, and taking the knowledge graph nodes and relationships as the state of the deep reinforcement learning network; splicing the embedding vectors of the knowledge graph nodes and relationships with the feature vectors of the work site environment parameters to form a state vector and input it into the deep reinforcement learning network; updating the action according to the state vector selection, and judging whether the update of the action is completed according to a preset reward function; and taking the action after the update as the initial auxiliary decision for the work ticket.

[0083] Specifically, in a deep reinforcement learning network, the risk assessment level of the power operation is considered as part of the action. Risk assessment levels can be categorized into multiple levels, such as low risk, medium risk, and high risk. Each risk level corresponds to a different decision-making strategy. In implementation, the risk assessment level can be converted into a discrete action space, for example, using integer values ​​to represent different risk levels (e.g., 0 for low risk, 1 for medium risk, and 2 for high risk). Safety measures are selected as another part of the action. Safety measures can include various specific safety operations, such as wearing safety gear, setting up warning signs, and shutting off power. These safety measures can constitute a discrete set of actions. For example, an action space can be defined that includes multiple safety measures, with each safety measure associated with a unique identifier (e.g., 0 for wearing a hard hat, 1 for setting up a fence, etc.). Furthermore, workflow scheduling is also considered as part of the action. A workflow can include multiple steps, such as equipment inspection, troubleshooting, and equipment repair. These steps can form a sequenced action space. In implementation, the workflow can be broken down into multiple sub-steps, with an action defined for each sub-step. For example, integer sequences are used to represent different stages of a workflow (e.g., 0 for start, 1 for equipment inspection, 2 for troubleshooting, etc.). Entities and relationships in the knowledge graph are embedded into a vector space using graph embedding techniques (such as the TransE algorithm). The TransE algorithm is trained to ensure that the embedding vectors of correct triples (head entity, relationship, tail entity) in the knowledge graph satisfy certain distance constraints. Worksite environmental parameters can include sensor data (such as temperature, humidity, and voltage) and equipment operation data (such as equipment status and operating time). These parameters can be converted into feature vectors after preprocessing (such as normalization). For example, values ​​such as temperature, humidity, and voltage can be normalized to a range of 0 to 1 and concatenated into a single vector. The embedding vectors of the knowledge graph nodes and relationships are concatenated with the feature vectors of the worksite environmental parameters to form a unified state vector.

[0084] For example, in a specific implementation, the two vectors can be connected end to end. For example, the knowledge graph embedding vector is , the characteristic vector of the working site environment parameters is .

[0085] In a preferred embodiment, the reinforcement learning network can employ a deep Q-network (DQN) architecture. The network's input layer receives a state vector s, and the hidden layers consist of multiple fully connected layers, each using the Reluctant Unit (ReLU) activation function. For example, the first hidden layer has 128 neurons, and the second hidden layer has 64 neurons. The number of neurons in the output layer corresponds to the size of the action space. Each neuron outputs a Q-value corresponding to an action, representing the expected reward of performing that action in the current state. At each decision step, the state vector is input into the DQN to obtain the Q-values ​​of all actions. The action with the highest Q-value is selected as the current decision. The completion of the action update is determined based on a preset reward function. The reward function can be defined as follows: a positive reward (e.g., +1) is given when the selected action effectively reduces risk and complies with safety regulations; a negative reward (e.g., -1) is given when the action increases risk or violates safety regulations; and a zero reward is given otherwise. During training, an experience replay buffer is used to store the most recent four-tuples (state, action, reward, next state). During training iterations, mini-batches of samples are randomly drawn from the experience replay buffer to update the DQN parameters. The target network is regularly updated to generate stable target Q values, guiding network training. After training, the current state vector is input into the trained DQN to obtain the optimal action, which serves as the initial auxiliary decision for the power operation work order. This initial auxiliary decision includes risk assessment level, safety measure selection, and workflow arrangement, guiding operators to execute tasks according to the decision.

[0086] In this optional embodiment, the initial auxiliary decision of the power operation work ticket is determined by a deep reinforcement learning network, which realizes the precise control and optimization of the key elements of the power operation; the risk assessment level, safety measure selection and workflow arrangement are defined as the actions of the deep reinforcement learning network, which ensures that the decision-making plan can fully cover the core links of the power operation, and each action is closely related to the safety and efficiency of the operation. At the same time, the knowledge graph nodes and relationships are used as state inputs, which fully integrates the professional knowledge and actual on-site conditions in the field of power operation, and provides a rich basis for decision-making for the network. By splicing the knowledge graph embedding vector and the work site environment parameter feature vector to form a state vector, the organic fusion of multi-source information is realized, so that the network can make decisions based on comprehensive state information. Through continuous learning and optimization, the deep reinforcement learning network can automatically generate the optimal initial auxiliary decision in different on-site environments, avoiding the decision-making errors caused by experience differences or incomplete information when making traditional manual decisions, and improving the consistency and reliability of decisions.

[0087] Optionally, the initial auxiliary decision is input into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision, including: inputting the initial auxiliary decision into the power operation system model, and performing simulation execution according to the risk assessment level, the safety measure selection and the workflow arrangement; when the simulation execution is completed, obtaining the safety indicators, efficiency indicators and standard indicators of the power operation system model after the simulation execution; and using the safety indicators, the efficiency indicators and the standard indicators as the execution effect.

[0088] Specifically, the power operation system model is a virtual simulation environment based on digital twin technology. It includes virtual models of power equipment, virtual agents for operators, a virtual execution module for workflows, and a virtual simulation module for environmental conditions. For example, for a substation operation scenario, the system model includes virtual models of equipment such as transformers, circuit breakers, and switchgear, as well as virtual agents for operators to simulate operational behaviors in actual operations. The virtual models of the equipment are constructed based on their actual parameters and operating characteristics, such as the transformer's power, current, and voltage, as well as its operating status under different load conditions.

[0089] The operator's virtual agent is capable of simulating human behavior, such as movement, equipment operation, and safety measures. These models are implemented using programming languages ​​(such as Python or C++) and simulation engines (such as Unity or Unreal Engine). The workflow virtual execution module defines the steps and processes of work tasks, such as equipment inspection, troubleshooting, and maintenance operations. The environmental conditions virtual simulation module simulates the impact of environmental factors such as temperature, humidity, and wind speed on work. These modules are configured through scripts and configuration files to ensure simulation accuracy and realism. Initial decision support (including risk assessment level, safety measure selection, and workflow schedule) is passed as input parameters to the power operation system model. For example, if the risk assessment level is "medium risk," the safety measure selection is "wear a safety helmet and install a grounding wire," and the workflow is "equipment inspection → fault location → maintenance operation → test verification."

[0090] The system model simulates execution based on the input decision parameters. The operator's virtual agent performs the corresponding actions according to the workflow, and the equipment's virtual model responds based on operational and environmental conditions. For example, in the simulation, the virtual agent first checks the equipment status, then wears a safety helmet and sets a grounding wire as required by safety measures, and then performs fault location and repair operations.

[0091] During the simulation, the system model monitors and records various indicators in real time. Safety indicators include whether equipment operation complies with safety regulations and whether workers avoid hazardous areas; efficiency indicators include work completion time and equipment downtime; and compliance indicators include whether operating procedures conform to standard procedures. This data is collected and stored through sensor simulation and logging functions. Safety indicators are evaluated from two perspectives: equipment safety and personnel safety. For example, equipment safety indicators can be assessed by monitoring whether equipment is overloaded or malfunctioning; personnel safety indicators can be assessed by checking whether the virtual agent enters hazardous areas and whether safety equipment is properly worn. Furthermore, in actual application operations, a safety violation counter can be set. Each time a safety violation occurs, the counter increments by 1. After the simulation, the number of violations is counted as a quantitative value for the safety indicator.

[0092] Efficiency metrics include work completion time and resource utilization. For example, work completion time can be calculated by recording the time from the start to the end of a simulation; resource utilization can be assessed by monitoring equipment usage time and personnel workload. In actual applications, timestamps and resource usage logs can be used to record relevant data for analysis and calculation. Standardization metrics primarily check whether the workflow conforms to pre-set standards and specifications. For example, this can be assessed by checking whether each step in the workflow is executed in the correct order and whether key steps are omitted. In specific implementations, the actual workflow can be compared with the pre-set standard process, and the number of steps that conform to the standard process and the number of steps that are incorrectly sequenced can be counted to quantify the standardization metrics. Safety, efficiency, and standardization metrics obtained from the simulation can be compiled and analyzed. For example, a report containing the numerical values ​​of each metric can be generated, such as zero safety violations, 45 minutes for work completion time, and 90% of steps that conform to the standard process.

[0093] In this optional embodiment, by inputting the initial auxiliary decision into the power operation system model for simulation, a virtual execution and effectiveness evaluation of the power operation plan is achieved. During the simulation, the system model operates according to the risk assessment level, safety measure selection, and workflow arrangement in the initial auxiliary decision, comprehensively verifying the feasibility and effectiveness of the decision in actual operation. After the simulation is completed, the system's safety indicators, efficiency indicators, and standard indicators are obtained. These indicators can accurately and quantitatively reflect the effectiveness of the decision after execution. For example, safety indicators can reveal potential risks and security vulnerabilities in the operation process, efficiency indicators can reflect the smoothness of the workflow and resource utilization efficiency, and standard indicators can verify whether the operation complies with established safety and operating standards. This provides data support and intuitive feedback for optimizing power operation decisions. Through simulation, potential problems and risks in the initial auxiliary decision can be discovered in a timely manner before actual operation, avoiding direct trial and error in actual operation, thereby reducing safety accidents and economic losses that may occur due to decision-making errors. In addition, the quantitative assessment of safety, efficiency and standardization indicators provides a clear direction and basis for the adjustment and optimization of decision-making, helps to formulate more scientific and reasonable work ticket decisions, improves the overall safety and efficiency of power operations, and ensures that power operations run on a safe, efficient and standardized track.

[0094] Optionally, the initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain the final auxiliary decision of the work ticket, including: judging whether the initial auxiliary decision meets the job requirements by comparing the expected execution effect with the execution effect; when the initial auxiliary decision meets the job requirements, the initial auxiliary decision is used as the final auxiliary decision; when the initial auxiliary decision does not meet the job requirements, the initial auxiliary decision is optimized according to the comparison result of the expected execution effect and the execution effect, and the optimized initial auxiliary decision is used as the final auxiliary decision.

[0095] Specifically, safety, efficiency, and compliance indicators within the execution performance are compared with expected values. For safety indicators, such as zero safety violations, the closer the actual value is to zero, the better the decision. For efficiency indicators, such as work completion time ≤ 30 minutes, the lower the actual value, the better. For compliance indicators, such as process compliance rate ≥ 95%, the higher the actual value, the better. Furthermore, thresholds are set for each indicator, such as safety violations ≤ 1, work completion time ≤ 45 minutes, and process compliance rate ≥ 90%. If the actual value is within the threshold range, the initial decision is considered to meet the operational requirements; otherwise, it is considered to be unsatisfactory.

[0096] If the initial decision-making support fails to meet operational requirements, it is optimized based on the comparison results. If the risk assessment is too high, additional safety measures may be implemented or workflow adjustments may be made. These include: adding safety measures, such as requiring double grounding and adding on-site safety supervisors for operations with a "high risk" risk assessment level; and adjusting workflows, such as moving the "equipment inspection" step forward and adding a "safety review" step. The optimized decision-making support is then re-entered into the power operation system model for simulation until the simulation results meet operational requirements, forming a closed-loop optimization process to ensure the scientific and feasible nature of the final decision. The optimized decision-making support is then subjected to final verification and re-entered into the power operation system model for simulation execution to confirm its compliance with operational requirements. If the optimized decision-making support is verified to meet operational requirements through simulation, it is confirmed as the final decision-making support for the work order. This complete work order decision-making plan is formed, including the risk assessment level, safety measure selection, and workflow arrangement.

[0097] In this optional embodiment, by quantifying the comparison indicators and setting thresholds, it is possible to objectively and accurately judge whether the initial auxiliary decision meets the operational requirements, thus avoiding the uncertainty caused by subjective judgment. Secondly, the optimization process is highly targeted and operational. Adjusting the initial auxiliary decision according to the comparison results can quickly and effectively improve the scientificity and rationality of the decision. Furthermore, the closed-loop optimization process ensures the reliability of the final decision. Through multiple simulation verifications, continuous optimization and adjustment until the decision meets the operational requirements provides a solid guarantee for power operations. This process not only improves the quality of work ticket decisions, but also lays the foundation for the safe and efficient execution of power operations, and has important practical application value.

[0098] Optionally, the determining whether the initial auxiliary decision meets the operational requirements by comparing the expected execution effect with the execution effect includes: determining whether the safety indicators, the efficiency indicators and the standard indicators all meet the corresponding expected execution effects; if the safety indicators, the efficiency indicators and the standard indicators all meet the corresponding expected execution effects, then determining that the initial auxiliary decision meets the operational requirements; if any one of the safety indicators, the efficiency indicators and the standard indicators does not meet the corresponding expected execution effects, then determining that the initial auxiliary decision does not meet the operational requirements.

[0099] Specifically, safety indicators are assessed using preset thresholds. Safety indicator thresholds are set based on safety regulations and historical data for power operations. For example, the threshold for the number of safety violations is set to 0, the threshold for the equipment stable operation indicator is set to an equipment operation status code of 1 (indicating normal operation), and the threshold for the percentage of time workers spend in the safe zone is set to 100%. The actual number of safety violations, equipment operation status codes, and the percentage of time workers spend in the safe zone are obtained from the simulation results. For example, after a simulation, the number of safety violations is 0, the equipment operation status code is 1, and the percentage of time workers spend in the safe zone is 100%. If the actual number of safety violations does not exceed the threshold (i.e., 0), the equipment operation status code is 1, and the percentage of time workers spend in the safe zone is 100%, the safety indicators are considered to meet the requirements. Furthermore, efficiency indicator thresholds are set based on power operation efficiency requirements and historical data. For example, the work completion time is set to be within 30 minutes, and the average time between equipment operations is set to be within 5 minutes. The actual work completion time and the average time between equipment operations are obtained from the simulation results. For example, after the simulation is executed, the work completion time is 25 minutes, and the average equipment operation interval time is 4 minutes. If the actual work completion time is less than or equal to 30 minutes, and the average equipment operation interval time is less than or equal to 5 minutes, the efficiency index is determined to meet the requirements. Finally, based on the operating specifications and process requirements of power operations, the threshold of the standard indicator is set. For example, after the simulation is executed, the operation step sequence matching degree is 100%, and the proportion of operations that meet the standard requirements is 100%. If the actual operation step sequence matching degree is equal to 100%, and the proportion of operations that meet the standard requirements is equal to 100%, the standard indicator is determined to meet the requirements.

[0100] In general, when safety, efficiency, and compliance indicators all meet their respective expected performance targets, the initial decision support can be determined to meet operational requirements. This comprehensive and rigorous approach effectively ensures the safety, efficiency, and compliance of power operations. If any of these indicators fail to meet the requirements, this indicates a flaw in the initial decision support and requires further optimization and adjustment until all indicators achieve the expected results, thereby ensuring the reliability and effectiveness of the final decision support.

[0101] In this optional embodiment, by quantitatively comparing safety, efficiency, and standardization indicators with expected values, an accurate assessment is made of whether the initial auxiliary decision meets operational requirements. The assessment of safety indicators focuses on risk avoidance and ensuring the safety of operators and equipment; the assessment of efficiency indicators focuses on whether the operation can be completed efficiently within the specified time; and the assessment of standardization indicators verifies whether the operation follows established processes and specifications. If all indicators meet expectations, the initial auxiliary decision meets the operational requirements and can be directly used as the final decision. If any indicator does not meet expectations, it can be determined that the decision needs to be optimized. This optional embodiment provides clear evaluation standards and basis. By quantitatively comparing specific indicators, it eliminates the subjectivity and ambiguity of human judgment, making decision evaluation more objective and accurate. It achieves a comprehensive review of the initial auxiliary decision, covering key aspects such as safety, efficiency, and standardization, ensuring the scientific and reliable nature of the final decision. This embodiment can promptly identify potential problems and avoid applying risky decisions to actual operations, providing a strong guarantee for the safe and efficient execution of power operations and has important practical application value.

[0102] Optionally, the initial auxiliary decision is optimized based on the comparison result between the expected execution effect and the execution effect, including: when the initial auxiliary decision does not meet the job requirements, according to the indicators in the execution effect that do not meet the expected execution effect, determining the indicator gap between the indicator and the expected execution effect corresponding to the indicator, and using the indicator gap as the comparison result; based on the indicator gap, the initial auxiliary decision is modified in combination with preset adjustment rules to obtain the final auxiliary decision.

[0103] Specifically, when the initial support decision fails to meet operational requirements, the execution performance indicator that fails to meet the expected performance is first identified. Specifically, if the expected number of safety violations is 0, but the actual simulated execution results incurred 1 violation, the safety indicator gap is 1. If the expected completion time is 30 minutes, but the actual simulated execution took 35 minutes, the efficiency indicator gap is 5 minutes. If the expected compliance rate for the operational steps is 100%, but the compliance rate in the actual simulated execution is only 90%, the standard indicator gap is 10%. By comparing the simulated execution results with the preset expected performance, the gap is determined, and the specific value represents the degree of difference between the actual execution and the expected value. Preset adjustment rules are set based on the indicator type and the size of the gap. For example, for the safety indicator adjustment rule, if the safety indicator gap is 1 violation, additional safety measures can be added to the initial support decision, such as adding safety supervisors, increasing safety training, or adding protective equipment in specific areas. For the efficiency indicator adjustment rule, if the efficiency indicator gap is 5 minutes, workflow optimization can be implemented, such as adjusting the equipment operation sequence, increasing the number of operators, or improving equipment operating efficiency. Specifically, an additional operator can be added to share the workload, or the equipment operation sequence can be adjusted to reduce unnecessary waiting time. Regarding the standard indicator adjustment rules, if the standard indicator gap is 10%, the workflow can be reviewed and revised, operational standard training can be strengthened, or confirmation steps can be added to key operational steps. For example, a two-person confirmation step can be added after each key operational step to ensure that the operation complies with standard requirements. The modified initial auxiliary decision (i.e., the optimized decision) must be re-entered into the power operation system model for simulation verification: the optimized decision is entered into the power operation system model and re-simulated to obtain new safety, efficiency, and standard indicators. The new execution results are again compared with the expected results. If all indicators meet the expected values, the optimized decision is determined as the final auxiliary decision. If any indicators still do not meet the requirements, the above process is repeated. Based on the comparison of the new execution results with the expected results, the indicator gap is re-determined and modified according to the preset adjustment rules until the optimized decision meets the operational requirements.

[0104] In this optional embodiment, through quantitative comparison and targeted adjustment, continuous optimization of the initial auxiliary decision is achieved to ensure that the final decision meets the operational requirements. First, by clarifying the indicator gap, the shortcomings in the decision can be accurately located, providing a specific direction for subsequent adjustments. Secondly, the preset adjustment rules provide a clear operational guide for decision modifications, ensuring that the adjustment process has rules to follow and avoids blindness. Finally, through re-simulation verification and iterative optimization, a closed-loop optimization process is formed, which can continuously improve the decision-making plan until it meets all operational requirements. This data-driven and rule-guided optimization method not only improves the scientificity and rationality of decision-making, but also provides a strong guarantee for the safe and efficient execution of power operations.

[0105] like Figure 3 As shown, the present invention also provides a work ticket decision-making auxiliary system, which is applied to a power distribution network, wherein the power distribution network includes a plurality of power distribution facilities, each of which corresponds to a different power operation system, and the system includes: a data processing unit for collecting and fusing multi-source data of the power operation system of the power distribution facility when it is necessary to operate the power distribution facility, so as to obtain multimodal data of the power operation system; a feature extraction unit for obtaining deep semantic features of the multimodal data according to the multimodal data through a large-scale language model, a data feature extraction model and an image feature extraction model; Features; a matching unit, used to match according to the deep semantic features through a preset knowledge graph, and obtain knowledge graph nodes and relationships related to the multimodal data; an initial decision unit, used to determine the initial auxiliary decision of the work ticket of the power operation based on the knowledge graph nodes and relationships, combined with a preset deep reinforcement learning network; a simulation unit, used to input the initial auxiliary decision into the power operation system model for simulation, and obtain the execution effect corresponding to the initial auxiliary decision; an optimization unit, used to optimize the initial auxiliary decision according to the execution effect and the expected execution effect, and obtain the final auxiliary decision of the work ticket.

[0106] The advantages of the work ticket decision-making assistance system of the present invention over the existing technology are the same as the advantages of the above-mentioned work ticket decision-making assistance method over the existing technology, and will not be repeated here.

[0107] Although the present invention is disclosed as above, the scope of protection disclosed by the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A work ticket decision-making assistance method, characterized in that: The method is applied to a power distribution network, the power distribution network including a plurality of power distribution facilities, each of the power distribution facilities corresponding to a different power operation system, and the method includes: When it is necessary to operate the power distribution facility, multi-source data of the power operation system of the power distribution facility is collected and integrated to obtain multi-modal data of the power operation system; Obtaining deep semantic features of the multimodal data based on the multimodal data through a large-scale language model, a data feature extraction model, and an image feature extraction model; Matching the deep semantic features with a preset knowledge graph to obtain knowledge graph nodes and relationships related to the multimodal data; Based on the knowledge graph nodes and relationships, combined with a preset deep reinforcement learning network, an initial auxiliary decision for the work ticket of the power operation is determined; specifically, the decision includes: using the risk assessment level, safety measure selection, and workflow arrangement of the power operation as the actions of the deep reinforcement learning network, and using the knowledge graph nodes and relationships as the states of the deep reinforcement learning network; Concatenating the embedding vectors of the knowledge graph nodes and relationships with the feature vectors of the worksite environment parameters to form a state vector and input it into the deep reinforcement learning network; Selecting to update the action according to the state vector, and determining whether the update of the action is completed according to a preset reward function; Using the updated action as the initial auxiliary decision of the work ticket; Inputting the initial auxiliary decision into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision; The initial auxiliary decision is optimized according to the execution effect and the expected execution effect to obtain a final auxiliary decision for the work ticket.

2. The work ticket decision-making auxiliary method according to claim 1, characterized in that: When the power distribution facility needs to be operated, multi-source data of the power operation system of the power distribution facility is collected and integrated to obtain multimodal data of the power operation system, including: When it is necessary to operate the power distribution facility, collecting text data, sensor data, image data, and equipment operation data in the power operation system of the power distribution facility, and using the text data, the sensor data, the image data, and the equipment operation data as the multi-source data; The multi-source data are converted into a preset format through data fusion technology, and the multi-source data in a unified format are fused to obtain the multimodal data of the power operation system.

3. The work ticket decision-making auxiliary method according to claim 2, characterized in that: The method of obtaining deep semantic features of the multimodal data based on the multimodal data by using a large-scale language model, a data feature extraction model, and an image feature extraction model includes: Inputting the multimodal data into the large-scale language model, and performing semantic extraction on the text data in the multimodal data using the large-scale language model to obtain semantic information of the multimodal data; Inputting the multimodal data into the data feature extraction model, and performing feature extraction on the sensor data and the device operation data in the multimodal data through the data feature extraction model to obtain data features of the multimodal data; Inputting the multimodal data into the image feature extraction model, and performing feature extraction on the image data in the multimodal data by the image feature extraction model to obtain image features of the multimodal data; The semantic information, the data features and the image features are fused to obtain the deep semantic features.

4. The work ticket decision-making auxiliary method according to claim 1, characterized in that: The matching of the preset knowledge graph according to the deep semantic features to obtain knowledge graph nodes and relationships related to the multimodal data includes: Performing vectorization processing on the deep semantic feature to obtain a feature vector of the deep semantic feature; Embed the entities and relationships in the preset knowledge graph into a vector space using graph embedding technology to obtain a knowledge graph embedding vector; Calculating the cosine similarity between the knowledge graph embedding vector and the feature vector; Based on the cosine similarity, the knowledge graph nodes and relationships related to the multimodal data are determined.

5. The work ticket decision-making auxiliary method according to claim 1, characterized in that: Inputting the initial auxiliary decision into the power operation system model for simulation to obtain the execution effect corresponding to the initial auxiliary decision includes: Inputting the initial auxiliary decision into the power operation system model, and performing simulation execution according to the risk assessment level, the safety measure selection, and the workflow arrangement; After the simulation is completed, the safety index, efficiency index and standard index of the power operation system model after the simulation are obtained; The safety index, the efficiency index and the standard index are used as the execution effect.

6. The work ticket decision-making auxiliary method according to claim 5, characterized in that: The optimizing the initial auxiliary decision according to the execution effect and the expected execution effect to obtain the final auxiliary decision of the work ticket includes: By comparing the expected execution effect with the execution effect, determining whether the initial auxiliary decision meets the operation requirements; When the initial auxiliary decision meets the operation requirements, the initial auxiliary decision is used as the final auxiliary decision; When the initial auxiliary decision does not meet the operation requirements, the initial auxiliary decision is optimized according to the comparison result between the expected execution effect and the execution effect, and the optimized initial auxiliary decision is used as the final auxiliary decision.

7. The work ticket decision-making auxiliary method according to claim 6, characterized in that: The step of comparing the expected execution effect with the execution effect to determine whether the initial auxiliary decision meets the operation requirements includes: Determining whether the safety indicator, the efficiency indicator, and the standard indicator all meet the corresponding expected performance effects; If the safety index, the efficiency index, and the standard index all meet the corresponding expected execution effects, then it is determined that the initial auxiliary decision meets the operation requirements; If any one of the safety index, the efficiency index and the standard index does not meet the corresponding expected execution effect, it is determined that the initial auxiliary decision does not meet the operation requirements.

8. The work ticket decision-making auxiliary method according to claim 6, characterized in that: The optimizing the initial auxiliary decision according to the comparison result between the expected execution effect and the execution effect includes: When the initial auxiliary decision does not meet the operation requirements, determining, based on the indicator of the execution effect that does not meet the expected execution effect, an indicator gap between the indicator and the expected execution effect corresponding to the indicator, and using the indicator gap as the comparison result; According to the indicator gap, the initial auxiliary decision is modified in combination with preset adjustment rules to obtain the final auxiliary decision.

9. A work ticket decision-making support system, characterized in that: The system is applied to a power distribution network, which includes a plurality of power distribution facilities, each of which corresponds to a different power operation system. The system includes: a data processing unit, configured to collect and fuse multi-source data of the power operation system of the power distribution facility when it is necessary to operate the power distribution facility, so as to obtain multi-modal data of the power operation system; A feature extraction unit, configured to obtain deep semantic features of the multimodal data based on the multimodal data by using a large-scale language model, a data feature extraction model, and an image feature extraction model; A matching unit, configured to perform matching based on the deep semantic features using a preset knowledge graph to obtain knowledge graph nodes and relationships related to the multimodal data; An initial decision-making unit is configured to determine an initial auxiliary decision for a work order for an electric power operation based on the knowledge graph nodes and relationships and in combination with a preset deep reinforcement learning network. Specifically, the initial decision-making unit comprises: using the risk assessment level, safety measure selection, and workflow arrangement of the electric power operation as actions of the deep reinforcement learning network, and using the knowledge graph nodes and relationships as states of the deep reinforcement learning network. Concatenating the embedding vectors of the knowledge graph nodes and relationships with the feature vectors of the worksite environment parameters to form a state vector and input it into the deep reinforcement learning network; Selecting to update the action according to the state vector, and determining whether the update of the action is completed according to a preset reward function; Using the updated action as the initial auxiliary decision of the work ticket; A simulation unit, configured to input the initial auxiliary decision into a power operation system model for simulation, and obtain an execution effect corresponding to the initial auxiliary decision; The optimization unit is used to optimize the initial auxiliary decision according to the execution effect and the expected execution effect to obtain a final auxiliary decision for the work ticket.

Images