Power grid dispatching automation intelligent operation and maintenance method and system based on multi-modal AI large model

Abstract

The application provides a power grid dispatching automation intelligent operation and maintenance method and system based on a multi-modal AI large model, relates to the technical field of power grid dispatching operation and maintenance, and first collects multi-modal information of power grid dispatching operation and maintenance and performs coupling processing to output a coupling feature set; then inputs the coupling feature set into a multi-modal AI large model to mine an operation and maintenance risk chain and pre-position deduction, and outputs a risk evolution chain and full-stage risk state description information; next, based on the risk evolution chain and the full-stage risk state description information, a plurality of operation and maintenance intervention strategies adapted to different risk nodes are deduced; the operation and maintenance intervention strategies are subjected to full-scene efficiency simulation verification to output an optimal strategy; finally, the optimal strategy is converted into a standardized instruction sequence and pushed to an execution end and tracked for execution, and an operation and maintenance intervention result is output. The application improves the accuracy, safety and intelligent level of power grid dispatching operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching and operation and maintenance technology, and more specifically, to a power grid dispatching automation and intelligent operation and maintenance method and system based on a multimodal AI large model. Background Technology

[0002] In the field of automated operation and maintenance of power grid dispatch, traditional methods face numerous challenges. Currently, the information sources relied upon for power grid dispatch and maintenance are relatively singular, mainly focusing on limited information types such as equipment operating status. There is insufficient integration of key multimodal information such as power grid dispatch command information, system interaction information, and environmental impact information. These different types of information are interconnected and mutually influential, jointly determining the operational status and potential risks of power grid dispatch. However, traditional methods have failed to comprehensively couple and analyze them, resulting in an inaccurate grasp of the overall operational status of the power grid.

[0003] In terms of risk assessment and response, traditional methods lack the ability to deeply explore and proactively predict the operational risk chain. They typically can only analyze and address risks after they have occurred, making it difficult to predict their development trends and evolution paths in advance, and hindering the implementation of effective preventative measures in the early stages of risk development. Furthermore, traditional methods struggle to quickly generate suitable operational intervention strategies for different risk evolution stages, often employing a one-size-fits-all approach that lacks specificity and flexibility.

[0004] Furthermore, traditional operation and maintenance intervention strategies lack a full-scenario performance simulation and verification process after they are formulated, making it impossible to accurately assess the actual effects of the strategies under different risk scenarios. This results in the adopted operation and maintenance intervention strategies not being the optimal solutions, increasing the risks and costs of power grid dispatch and operation and maintenance. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model, the method comprising: Collect multimodal information on power grid dispatching and maintenance, and perform coupling processing on the multimodal information on power grid dispatching and maintenance to output a coupling feature set. The multimodal information on power grid dispatching and maintenance includes power grid dispatching instruction information, equipment operating status information, system interaction information, and environmental impact information. The coupled feature set is input into a multimodal AI model to perform operation and maintenance risk chain mining and preliminary inference, and outputs risk evolution chain and full-stage risk status description information; Based on the risk evolution chain and the full-stage risk status description information, dynamic intervention strategy simulation is performed, and multiple operation and maintenance intervention strategies adapted to different risk evolution nodes are output. Perform full-scenario performance simulation verification on multiple operation and maintenance intervention strategies, and output the optimal operation and maintenance intervention strategy covering the entire risk evolution process; The optimal operation and maintenance intervention strategy is transformed into a standardized sequence of execution instructions, which is then pushed to the power grid dispatch automation operation and maintenance execution terminal to track the entire execution process and output the operation and maintenance intervention results.

[0006] Furthermore, embodiments of the present invention also provide a power grid dispatching automation and intelligent operation and maintenance system based on a multimodal AI large model, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned intelligent operation and maintenance method for power grid dispatching based on a multimodal AI large model by executing the machine-executable instructions.

[0007] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model.

[0008] Based on the above, by collecting and coupling multimodal information on power grid dispatch and maintenance, this method integrates information from various aspects such as power grid dispatch instructions, equipment operating status, system interactions, and environmental impacts. It outputs a coupled feature set, which is then input into a multimodal AI model to perform maintenance risk chain mining and pre-analysis. This enables early prediction of risk development trends and evolution paths, outputting risk evolution chains and full-stage risk status descriptions. This achieves a shift from passively responding to risks to proactively preventing them, significantly improving the safety and reliability of power grid dispatch and maintenance. Based on risk information, dynamic intervention strategy simulations are performed, outputting multiple maintenance intervention strategies adapted to different risk evolution nodes. The optimal maintenance intervention strategy is then verified through full-scenario performance simulation, making the intervention strategy highly targeted and flexible. It can select the most appropriate response measures according to different risk scenarios, effectively reducing maintenance risks and costs. The optimal maintenance intervention strategy is transformed into a standardized execution instruction sequence and pushed to the execution end for full-process tracking, achieving standardization and automation of maintenance intervention, improving maintenance efficiency and execution effectiveness. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the execution flow of the power grid dispatch automation and intelligent operation and maintenance method based on a multimodal AI large model provided in the embodiments of the present invention.

[0010] Figure 2 This is a schematic diagram of exemplary hardware and software components of the power grid dispatch automation and intelligent operation and maintenance system based on a multimodal AI large model provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model, according to an embodiment of the present invention. The following is a detailed description of this method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model.

[0012] Step S110: Collect multimodal information on power grid dispatching and maintenance, and perform coupling processing on the multimodal information on power grid dispatching and maintenance to output a coupling feature set. The multimodal information on power grid dispatching and maintenance includes power grid dispatching instruction information, equipment operating status information, system interaction information, and environmental impact information.

[0013] This embodiment uses a daily operation and maintenance scenario of a city power grid dispatch center as an example. In this scenario, the power grid dispatch center needs to monitor and dispatch the entire city's power grid in real time to ensure the safe and stable operation of the power grid.

[0014] Step S110-1: Deploy data collection nodes for collecting multi-dimensional information to cover the entire power grid dispatching and maintenance scenario. The data collection nodes are respectively set in the power grid dispatching command issuing terminal, equipment operation status monitoring port, system interaction data interface, and environmental parameter sensing area. The collection scope includes the issuance time, transmission path, and execution feedback of power grid dispatching commands; voltage parameters, current parameters, temperature parameters, and vibration parameters of equipment operation; data transmission rate, interaction frequency, and data integrity of system interaction; and environmental impacts such as temperature changes, humidity changes, wind force level, precipitation intensity, and surrounding electromagnetic interference intensity. After all data collection nodes are synchronously connected to a unified time base for data collection, multi-dimensional collected information is output.

[0015] In this urban power grid dispatch center scenario, a large number of data acquisition nodes need to be deployed first. Command acquisition software is installed on the power grid dispatch command issuing terminals, such as dispatcher workstations, to record the issuance time of each dispatch command, accurate to the millisecond. Regarding transmission paths, traffic monitoring tools are deployed on network devices such as network switches and routers within the dispatch center to collect information on the transmission path of commands in the network, including the IP addresses of the network nodes traversed. Execution feedback is achieved by communicating with execution terminals, such as the RTUs (Remote Terminal Units) in substations, to obtain the execution status of commands, such as successful execution, partial execution, or execution failure. For equipment operation status monitoring ports, corresponding sensors are installed on core equipment such as transformers, circuit breakers, and disconnectors in the substation. For example, voltage sensors, current sensors, temperature sensors, and vibration sensors are installed on transformers to collect voltage and current parameters on the high-voltage and low-voltage sides, winding and core temperature parameters, and vibration parameters of the transformer body, respectively. Regarding system interaction data interfaces, data acquisition devices are deployed at the communication interfaces between the main scheduling system and subsystems, such as SCADA (Supervisory Control and Data Acquisition) systems and EMS (Energy Management System). These devices collect data transmission rates (the amount of data transmitted per unit time), interaction frequency (the number of interactions per unit time), and data integrity, using methods such as checksums to determine data completeness. For environmental parameter sensing, environmental monitoring stations are set up in areas such as substations and along transmission lines to collect data on temperature changes (the magnitude of temperature rise and fall per unit time), humidity changes (the humidity change per unit time), wind speed (measured by wind speed sensors and converted to the corresponding wind speed level), precipitation intensity (calculated based on rainfall amount and duration), and surrounding electromagnetic interference intensity (measured using an electromagnetic interference tester). All acquisition nodes are synchronously connected to a unified time base via GPS or BeiDou time synchronization to ensure the consistency of the collected multi-dimensional information in time. This information is then aggregated and output to form multi-dimensional collected data.

[0016] Step S110-2: Perform full-process trajectory restoration processing on the power grid dispatch instruction information in the multi-dimensional collected information, track the complete transmission link of each power grid dispatch instruction from the issuing entity to the execution terminal, record the instruction reception time, response delay and data verification results of each node in the complete transmission link, extract the associated parameters in the instruction execution process, and output the full-process trajectory data of instruction execution.

[0017] After obtaining multi-dimensional collected information, the power grid dispatching instructions are processed. Each power grid dispatching instruction has a unique identifier, such as an instruction number. Taking an instruction to "adjust transformer tap changers" issued from the dispatch center to a substation as an example, its complete transmission link is traced. This power grid dispatching instruction is issued from the dispatcher's workstation (the issuing entity), passes through the local area network switch within the dispatch center, is transmitted to a router, then through the dedicated power communication network to the substation's communication room, and finally reaches the substation's RTU (Execution Unit). During this process, the instruction reception time of each node is recorded, such as the time when the switch receives the instruction, the time when the router receives the instruction, and the time when the RTU receives the instruction. Response latency refers to the time spent by each node from receiving the instruction to starting to process it. Data verification results are obtained by the nodes performing verification and calculation on the instruction data and comparing it with the sender's verification, recording whether the verification passed. Associated parameters include the dispatching task number corresponding to the instruction, the equipment number involved, and the preconditions required for instruction execution. After organizing the above information, the entire instruction execution trajectory data is output.

[0018] Steps S110-21: Extract the unique identifier code of each power grid dispatch instruction from the power grid dispatch instruction information collected from multiple dimensions. Using the unique identifier code as an index, associate the information of each link of instruction issuance, transmission, reception and execution, establish an instruction full-process information association table, and output the instruction full-process information association table.

[0019] For the aforementioned "adjust transformer tap changer" instruction, its unique identifier code is extracted from the multi-dimensional collected information, assumed to be "ZD-20231026-001". Using this identifier code as an index, the information related to the instruction's issuance process is searched in the multi-dimensional collected information, such as the issuance time being 10:00:00 on October 26, 2023, and the issuing device number being "GP-001"; transmission process information, such as the sequence of network node IP addresses; receiving process information, such as the RTU device number being "BT-005" and the receiving time being 10:00:02 on October 26, 2023; and execution process information, such as the execution start time being 10:00:03 on October 26, 2023, and the execution completion time being 10:00:10 on October 26, 2023. Link the above information together to create a full-process instruction information association table. The table contains fields such as identification code, issued information, transmitted information, received information, and executed information. Output the association table.

[0020] Steps S110-22: Trace the information of the instruction issuance process, determine the equipment number, issuance timestamp, and power grid operation load status of the instruction issuing entity, and enter the issuance process field of the instruction full process information association table, and output the first instruction full process information association table.

[0021] Continuing with the example of instruction “ZD-20231026-001”, we trace its issuance process. By querying the equipment ledger and operation records of the dispatch center, we determine that the equipment number of the issuing entity is “GP-001,” which is the dispatcher's workstation number. The issuance timestamp is accurate to milliseconds, specifically October 26, 2023, at 10:00:00.123. The power grid load status at the time of issuance is obtained by querying real-time data from the SCADA system at that point in time, such as the total active and reactive loads of the entire city's power grid. This information is then entered into the issuance process field of the instruction's full-process information association table; this association table is now the first instruction's full-process information association table.

[0022] Steps S110-23: Track the instruction transmission information, track all transmission nodes that the power grid dispatch instruction passes through from the issuing body to the execution terminal and record the node sequence, record the device number of each transmission node, the timestamp of the power grid dispatch instruction entering the transmission node, the timestamp of leaving the transmission node, the data verification result of the transmission node for the power grid dispatch instruction, calculate the instruction response delay of each transmission node, and enter the transmission link field of the first instruction full process information association table, and output the second instruction full process information association table.

[0023] The transmission process of the instruction “ZD-20231026-001” is traced, assuming the sequence of transmission nodes it passes through is as follows: dispatch center LAN switch (device number “JHJ-001”), core router (device number “LYQ-001”), power communication network access equipment (device number “TXW-001”), and substation communication room switch (device number “JHJ-002”). For each transmission node, its device number and the timestamp of the instruction entering the node are recorded. For example, the timestamp of entering “JHJ-001” is 10:00:00.150 on October 26, 2023, and the timestamp of leaving the node is 10:00:00.180 on October 26, 2023. The data verification result is obtained by each node verifying the instruction data. For example, the verification result of “JHJ-001” is “passed”. The instruction response delay is calculated by subtracting the entry timestamp from the departure timestamp. For "JHJ-001", the response delay is 0.030 seconds. This information is then entered into the transmission stage field of the first instruction full-process information association table to obtain the second instruction full-process information association table.

[0024] Steps S110-24: Extract instruction receiving information, determine the device number of the instruction receiving terminal, the receiving timestamp, the integrity status of the instruction data at the time of reception, and the feedback information of the receiving terminal. If there is missing or incorrect instruction data, record the specific fields of the missing or incorrect data and the supplementary or corrective measures, and enter the receiving stage field of the second instruction full-process information association table, and output the third instruction full-process information association table.

[0025] The receiving terminal for the instruction “ZD-20231026-001” is the RTU of the substation, with device number “BT-005”. The receiving timestamp is 10:00:02.050 on October 26, 2023. By performing an integrity check on the received instruction data, such as checking data length and the existence of key fields, the instruction data integrity status is determined to be “complete”. The receiving terminal's feedback message is “Instruction received, ready to be executed.” Since the data is complete and there are no missing or incorrect entries, the above information is entered into the receiving stage field of the second instruction full-process information association table, forming the third instruction full-process information association table.

[0026] Steps S110-25: Collect information on the instruction execution process, track the start execution timestamp after the execution terminal receives the instruction, the timestamps of key operation nodes during the execution process, and the execution completion timestamp, determine the scheduling task type corresponding to the instruction, record the device number of the instruction, the changes in the device's operating parameters during the execution process, and the result feedback information after the execution is completed, calculate the instruction execution completion time, and enter the execution process field of the third instruction full process information association table, and output the fourth instruction full process information association table.

[0027] After receiving the instruction from terminal "BT-005", the execution start timestamp was 10:00:03.000 on October 26, 2023. Key operational nodes during execution, such as starting tap adjustment and adjusting the tap to the target position, had timestamps of 10:00:03.500 and 10:00:08.000 on October 26, 2023, respectively. The execution completion timestamp was 10:00:10.000 on October 26, 2023. The scheduling task type corresponding to this instruction was "Transformer Tap Adjustment". The device executing the instruction was a transformer in the substation, with device number "BYQ-001". During execution, sensors collected changes in the transformer's operating parameters, such as voltage and current changes before and after tap adjustment. The feedback information after execution completion was "Tap adjustment successful, current position is the target position". The execution completion time of the instruction is calculated by subtracting the start execution timestamp from the completion timestamp, and is 7 seconds. This information is then entered into the execution stage field of the third instruction full-process information association table to obtain the fourth instruction full-process information association table.

[0028] Steps S110-26: Extract core information from the fourth instruction full-process information association table, organize the extracted core information according to the time sequence of instruction execution, and output the organized instruction full-process core information.

[0029] Core information is extracted from the fourth instruction's full-process information association table, including the instruction's unique identifier code, the issuing device number, the issuing timestamp, the transmission node sequence and related information for each node, the receiving terminal device number, the receiving timestamp, the execution start timestamp, the timestamps of key operation nodes, the execution completion timestamp, the scheduling task type, the executing device number, and the result feedback information. Then, this core information is organized according to the chronological order of instruction execution—that is, the order of issuing, transmitting, receiving, and executing—to form the organized core information for the entire instruction process.

[0030] Steps S110-27: Based on the timeline, the core information of the entire instruction process is linked together in the order of sending, transmitting, receiving, and executing to form an instruction execution trajectory diagram. At the same time, a structured data file containing all core information is generated, and the entire instruction execution trajectory data is output.

[0031] Using the timeline as the horizontal axis, the events and information corresponding to each time point in the core information of the entire instruction execution process are marked on the timeline to form an instruction execution trajectory diagram, which intuitively shows the entire process from instruction issuance to execution. At the same time, all core information is generated into a structured data file according to a certain format, such as XML or JSON, to facilitate subsequent processing and analysis, and finally output the entire instruction execution trajectory data.

[0032] Step S110-3: Perform correlation analysis on the equipment operation status information in the multi-dimensional collected information, select the core equipment in the power grid dispatching system as the analysis subject, track the status parameter change curve of a single device in different operating periods, compare the parameter fluctuation correlation of different devices under the same dispatching task, extract the collaborative change characteristics and abnormal fluctuation correlation characteristics of the equipment operation status, and output the equipment operation correlation status data.

[0033] In the scenario of an urban power grid dispatch center, transformers, circuit breakers, and disconnectors in substations are selected as core equipment for analysis. Taking transformer "BYQ-001" as an example, its status parameter change curves are tracked during different operating periods throughout the day, such as 8:00-10:00 AM, 12:00-2:00 PM, and 6:00-8:00 PM. For example, voltage parameters fluctuate with changes in grid load at different times, and a voltage-time curve is plotted. Simultaneously, the correlation of parameter fluctuations between transformer "BYQ-001" and its connected circuit breaker "DLQ-001" is compared under the same dispatch task, such as "increasing the load on a certain line." The correlation is observed to see if their current parameters rise or fall simultaneously, and whether there is a proportional relationship between the fluctuation amplitudes, thus extracting the coordinated change characteristics of equipment operating status. If, at a certain moment, the transformer's temperature parameter suddenly rises abnormally while the current parameter of its related cooling system equipment drops abnormally, this may be an abnormal fluctuation correlation characteristic. After organizing the above coordinated change characteristics and abnormal fluctuation correlation characteristics, the equipment operating status correlation data is output.

[0034] Step S110-4: Perform transmission path mining processing on the system interaction information collected from the multi-dimensional information, extract the interaction links between the main power grid dispatching system and subsystems, and between subsystems, record the interaction data type, data transmission volume, interaction frequency, interaction response time and interaction failure records of each interaction link, analyze the transmission impact of any system interaction anomaly on other related systems, extract the transmission correlation characteristics of system interaction, and output the system interaction transmission path data.

[0035] The power grid dispatching system comprises multiple subsystems, such as the SCADA system, EMS system, and DTS (Dispatcher Training Simulator) system. The interaction links between the main dispatching system and these subsystems, as well as between the subsystems themselves, are extracted. For example, the transmission of real-time telemetry and teleindication data from the SCADA system to the EMS system constitutes an interaction link. This link records the data types of the interaction (telemetry and teleindication data), the data transmission volume (size of each transmission), the interaction frequency (e.g., 10 transmissions per minute), the interaction response time (time from sending a request to receiving a response), and interaction failure records (e.g., transmission failure at a specific point in time and the reason for the failure). When an anomaly occurs in one system interaction, such as a data transmission interruption between the SCADA and EMS systems, the impact of this anomaly on other related systems is analyzed. For instance, the EMS system's inability to obtain real-time data may lead to deviations in its load forecasting function, thus affecting dispatching plan formulation. The DTS system, which relies on EMS system data for training simulations, will also be affected. By analyzing these impacts, the transmission correlation characteristics of system interactions are extracted, such as data transmission interruption leading to abnormal functioning of systems that depend on that data, and system interaction transmission path data is output.

[0036] Step S110-5 involves performing effect analysis and processing on the environmental impact information collected from multiple dimensions, distinguishing the impact of natural environmental factors and human environmental factors on power grid dispatching and operation, tracking the fluctuation of power grid equipment operating parameters when a single environmental factor changes, analyzing the comprehensive impact effect under the superposition of multiple environmental factors, extracting the correlation characteristics between environmental factors and power grid operating status, and outputting environmental impact data.

[0037] The natural environmental factors in the environmental impact information include temperature changes, humidity changes, wind speed, and precipitation intensity, while the anthropogenic environmental factors mainly consist of the intensity of surrounding electromagnetic interference. Taking temperature change as a single natural environmental factor as an example, the fluctuations in the temperature parameters and cooling fan operation of transformer "BYQ-001" are tracked when the ambient temperature rises. Typically, an increase in temperature leads to a rise in transformer temperature, and the cooling fan may start or increase its speed. For the combined effects of multiple environmental factors, such as the simultaneous occurrence of high temperature, high humidity, and heavy precipitation, the comprehensive impact on the operating status of power grid equipment is analyzed. High temperature and high humidity may lead to a decrease in the insulation performance of equipment, while heavy precipitation may affect the operation of outdoor equipment, such as causing short circuits and other faults. By analyzing these situations, the correlation characteristics between environmental factors and the operating status of the power grid are extracted, such as the increased probability of equipment failure under high temperature and high humidity conditions, and the environmental impact data are output.

[0038] Step S110-6: Standardize the entire process trajectory data of instruction execution, equipment operation status data, system interaction transmission path data, and environmental impact data. Then, use a multimodal coupling algorithm to mine the interaction relationships between the standardized data and extract the coupling features formed by different data combinations. Because different types of data have different dimensions and formats, standardization is necessary. For example, numerical data such as instruction execution time and equipment operating parameters are normalized to ensure they fall within the same numerical range; textual data, such as instruction result feedback information, is encoded and converted into a computer-readable format. After standardization, multimodal coupling algorithms, such as deep learning-based multimodal feature fusion algorithms, are used to process the data. These algorithms can automatically learn the interaction relationships between different modalities, such as the relationship between instruction execution and equipment operating status, and the relationship between environmental factors and system interactions. By mining these relationships, coupling features formed by different data combinations are extracted, such as the coupling feature of "transformer temperature rises and scheduling instruction execution delay occurs under high-temperature conditions."

[0039] Step S110-7: Perform time-series filtering on the extracted coupling features, retain dynamic coupling features that change with time trends, remove static coupling features that have no risk, and output a coupling feature set containing various dynamic correlation information.

[0040] The extracted coupling features are subjected to time-series analysis to observe their changes over time. Dynamic coupling features that exhibit a certain trend over time, such as "as the electrical load increases, the transformer current gradually rises and the temperature also rises accordingly," are retained. Static coupling features that do not change over time or change irregularly and do not indicate any risk, such as "the fixed correspondence between equipment model and equipment location," are discarded. The final output is a set of coupling features containing various dynamic correlation information, which can reflect the dynamic changes and potential risks in the power grid dispatching and maintenance process.

[0041] Step S120: Input the coupled feature set into the multimodal AI large model to perform operation and maintenance risk chain mining and preliminary inference, and output risk evolution chain and full-stage risk status description information.

[0042] In the scenario of urban power grid dispatch center, the coupled feature set obtained above is input into the multimodal AI big model. This multimodal AI big model is trained with a large amount of power grid dispatch and maintenance data, which can identify potential risks and explore the risk evolution chain, and at the same time perform advance simulation of the risk development process.

[0043] This multimodal AI model is a deep neural network model designed for automated operation and maintenance scenarios in power grid dispatching. It consists of four modules connected in series: a multimodal feature fusion input layer, a feature recognition unit, a correlation mining unit, and a risk evolution inference unit. These modules are sequentially connected, with the output of one module serving as the input of the next, forming an end-to-end inference chain. This multimodal AI model corresponds to all sub-steps in step S120, namely, receiving the coupled feature set output from steps S110-7, processing it internally, and outputting the risk evolution chain and full-stage risk state description information.

[0044] The multimodal feature fusion input layer serves as the model's input point, receiving the coupled feature set output from step S110-7. This input layer contains four parallel modal encoding branches: a recurrent neural network branch (using a bidirectional long short-term memory network) for processing the entire instruction execution trajectory data; a one-dimensional convolutional neural network branch for processing equipment operation-related state data; a graph convolutional network branch for processing system interaction and transmission path data; and a two-dimensional convolutional neural network branch for processing environmental impact data. The feature vectors output from the four modal encoding branches undergo intermodal information interaction and weighted fusion through a cross-attention mechanism, ultimately outputting a multimodal fusion feature tensor.

[0045] The feature recognition unit executes step S120-1, which contains a learnable risk feature comparison rule matrix with dimensions D×K, where D is the embedding dimension of the multimodal fusion features and K is the preset number of risk feature patterns. Each risk feature pattern corresponds to a specific anomaly type parameterization in the power grid dispatching and maintenance field, including transformer winding temperature anomaly pattern, transformer core temperature anomaly pattern, transformer vibration anomaly pattern, circuit breaker opening and closing coil current anomaly pattern, circuit breaker contact temperature anomaly pattern, disconnector operating torque anomaly pattern, data acquisition and monitoring system data transmission interruption pattern, energy management system load prediction deviation pattern, dispatching master station and substation communication timeout pattern, and electromagnetic interference exceeding standard pattern. This unit calculates the dot product similarity between each feature vector in the input multimodal fusion feature tensor and the risk feature comparison rule matrix. When the similarity exceeds the activation threshold of 0.75, the feature vector is marked as a risk trigger source feature, and its output is a risk source label vector containing risk feature type code, trigger timestamp, initial parameter value, and scene label.

[0046] The association mining unit executes steps S120-2 and S120-3. Its internal structure is a feature propagation analysis network based on a multi-head attention mechanism, containing eight attention heads, each with a dimension of 64. This unit calculates the attention weights between any two feature vectors in the multimodal fusion feature tensor as the influence strength using the multi-head attention mechanism, and calculates the propagation time using positional encoding differences, outputting a feature association strength matrix and a propagation time matrix. This unit also includes a historical case matching submodule, used to supplement the initial risk propagation path with intermediate features, feature change nodes, and propagation path turning information from historical cases.

[0047] The risk evolution simulation unit executes steps S120-4 to S120-8. Its internal structure is a spatiotemporal graph neural network based on the power grid physical topology and operational logic constraints. This network uses the feature correlation strength matrix output by the correlation mining unit and the real-time power grid topology graph as joint inputs. The nodes in the real-time power grid topology graph are equipment entities and system functional entities within the power grid. Equipment entities include transformers, circuit breakers, disconnectors, transmission lines, and busbars. System functional entities include the data acquisition and monitoring system master station, the energy management system master station, remote terminal units, and phasor measurement units. Edge types include physical connection edges and logical interaction edges. Physical connection edges represent electrical connections between devices, while logical interaction edges represent data communication relationships between systems. The network uses gated recurrent units for state prediction in the time dimension and graph convolution operations for risk state propagation along the power grid topology in the spatial dimension. Finally, it outputs the risk evolution chain and the full-stage risk state description information of each node. The risk evolution chain starts with the risk trigger source characteristics, ends with the final risk impact characteristics, and uses the intermediate transmission characteristics as nodes. The risk state description information of each node includes the feature type, feature strength, occurrence time, impact range, and the risk evolution stage to which it belongs.

[0048] The training of this multimodal AI model is divided into two stages. The first stage is an unsupervised pre-training stage, which uses massive amounts of historical power grid operation data from the past five years to train masked feature prediction on the four modal coding branches of the multimodal feature fusion input layer. The training task is to randomly mask a portion of the input features and predict the masked portion based on context. Simultaneously, the feature transmission analysis network of the association mining unit is trained to reconstruct the risk transmission trajectory. The training task is to predict the complete transmission path given a portion of the risk transmission trajectory. The second stage is a supervised fine-tuning stage, which performs joint fine-tuning on a historical risk case dataset containing 5,000 historical risk cases labeled by power grid dispatch and maintenance experts. The joint loss function during fine-tuning is a weighted sum of two sub-loss functions: the first term is the risk evolution chain prediction error, using the cross-entropy loss function with a weight of 0.6; the second term is the risk state prediction error, using the mean squared error loss function with a weight of 0.4. By minimizing this joint loss function, the model learns the mapping relationship from the multimodal coupled feature input to the risk evolution chain and risk state description information output.

[0049] The following section elaborates on each step of this method, taking into account the specific structure and function of the multimodal AI large model.

[0050] Step S120-1: Input the coupled feature set into the feature recognition unit of the multimodal AI large model, filter out the coupled features with risk triggering tendency through the preset risk feature comparison rules, determine the risk triggering source features, mark the power grid dispatching and maintenance scenario corresponding to the risk triggering source features, record the initial parameters, occurrence time and related data of the risk triggering source features, and output the risk triggering source features and related information.

[0051] The feature recognition unit of the multimodal AI large model first processes the input coupled feature set. Pre-defined risk feature comparison rules are based on historical risk cases and expert experience; for example, "transformer temperature continuously rising and cooling fan not starting" might be considered a coupled feature with a risk-triggering tendency. By comparing each feature in the coupled feature set with the pre-defined rules, features that meet the rules are selected and identified as risk-triggering source features. The corresponding power grid dispatching and maintenance scenario for this risk-triggering source feature is labeled, such as "substation transformer operation scenario." Its initial parameters, such as the current transformer temperature and cooling fan status, the occurrence time (i.e., the time when the feature was detected), and related data, such as the ambient temperature and load conditions at the time, are recorded. The risk-triggering source feature and related information are then output.

[0052] Step S120-2: Based on the risk trigger source features and related information, the association mining unit of the multimodal AI model is used to mine the transmission relationship between the risk trigger source and other features in the coupled feature set, analyze the influence intensity and transmission time between different features, determine the initial transmission direction after the risk is triggered, and output the initial risk transmission path.

[0053] The association mining unit of the multimodal AI large model utilizes machine learning algorithms, such as association rule mining algorithms, to analyze the relationship between the risk triggering source feature and other features in the coupled feature set. For example, the risk triggering source feature "transformer temperature continues to rise and cooling fan does not start" may have a transmission correlation with features such as "transformer insulation performance deteriorates" and "transformer protection device operates". The unit analyzes the influence strength between these features, i.e., the degree to which a change in one feature affects a change in another, and the transmission time, i.e., the time required for a change in one feature to lead to a corresponding change in another feature. Based on these analysis results, the initial transmission direction after risk triggering is determined, such as from abnormal transformer temperature to insulation performance deterioration, and then to protection device operation, outputting the initial risk transmission path.

[0054] Step S120-3: Import historical risk data of power grid dispatch automation operation and maintenance, screen historical risk cases that are consistent with the characteristics of risk triggering sources, extract risk transmission process data from the historical risk cases, supplement them into the initial risk transmission path, and output the improved risk transmission path.

[0055] Steps S120-31: Collect historical risk data of power grid dispatch automation operation and maintenance. The historical risk data of power grid dispatch automation operation and maintenance covers various operation and maintenance risk cases that have occurred in the power grid dispatch system, including the time of occurrence of the risk, the scenario of occurrence, the risk triggering characteristics, the risk transmission process record, the scope of risk impact, intervention measures and effects.

[0056] Data on various operation and maintenance risk cases that occurred in the city's power grid dispatch center over the past few years were collected. These data recorded in detail the time of the risk occurrence, such as "peak electricity consumption period in the summer of 2022", the scenario of occurrence, such as "substation equipment failure scenario", the risk triggering characteristics, such as "abnormal increase in transformer temperature", the risk transmission process, that is, the characteristics and events of each stage of the risk from triggering to development, the scope of risk impact, such as the number of affected equipment and the geographical area, the intervention measures, such as the cooling measures taken and equipment shutdown and maintenance, and the intervention effect, such as whether the risk was controlled and the recovery time.

[0057] Steps S120-32 involve classifying and organizing historical risk data for power grid dispatch automation operation and maintenance, dividing it into multiple categories based on the type of risk triggering characteristics, and further subdividing each category according to the type of equipment or system affected by the risk, thus forming a historical risk case classification library.

[0058] Historical risk data is categorized according to the type of risk triggering characteristics, such as temperature anomalies, current anomalies, and voltage anomalies. Within each category, it is further subdivided according to the type of equipment or system affected by the risk; for example, the temperature anomaly category is divided into transformer temperature anomalies, circuit breaker temperature anomalies, etc. This resulting historical risk case classification database facilitates rapid retrieval and filtering of relevant cases.

[0059] Steps S120-33: Extract the core parameters from the risk trigger source features and related information, use the extracted core parameters as search conditions to search the historical risk case classification database, and output the retrieved historical risk cases.

[0060] Core parameters are extracted from the characteristics and related information of the risk triggering source, such as "abnormal increase in transformer temperature," "cooling fan not started," and "substation scenario." Using these core parameters as search criteria, a search is conducted in the historical risk case classification database to find historical risk cases with similar triggering source characteristics to the current risk.

[0061] Steps S120-34: Calculate the similarity of the core parameters of the risk trigger features and the risk trigger source features in the retrieved historical risk cases, output the similarity calculation results, filter out historical risk cases with similarity higher than the feature matching threshold in the similarity calculation results, form a set of common source risk cases, sort the cases in the set of common source risk cases, arrange them from high to low similarity, and output the sorted set of common source risk cases.

[0062] For each retrieved historical risk case, the similarity of its risk triggering features with the core parameters of the current risk triggering source features is calculated. Similarity calculation can be achieved by comparing the matching degree of each core parameter, such as parameter type, numerical range, etc. A feature matching threshold, such as 80%, is set, and historical risk cases with similarity higher than this threshold are selected to form a set of common-source risk cases. Then, the set of common-source risk cases is sorted in descending order of similarity to prioritize the cases with the highest similarity.

[0063] Steps S120-35: Analyze the top N cases in the sorted set of cases with the same source of risk, extract the risk transmission process record for each case, extract the intermediate features, feature change nodes, and transmission path turning information in the risk transmission process, and output the extracted risk transmission process related information.

[0064] Select the top N cases (N can be set according to the actual situation, such as 5 cases) from the sorted set of cases with the same source of risk for analysis. Extract the risk transmission process record for each case, and obtain intermediate features, i.e., other abnormal features that appear during the risk transmission process, feature change nodes, i.e., the time points when intermediate features appear, and transmission path reversal information, i.e., the situation where the direction of risk transmission changes, such as from affecting device A to affecting device B, etc. Output these extracted risk transmission process related information.

[0065] Steps S120-36 involve deduplicating the extracted risk transmission process information and analyzing the compatibility between the core information of the deduplicated risk transmission process and the initial risk transmission path. This includes determining whether intermediate features can be integrated into the initial risk transmission path, whether feature change nodes are consistent with the temporal logic of the initial risk transmission path, and whether the transmission path turning information conforms to the current risk evolution trend. Information that conforms to the temporal logic and evolution trend of the initial risk transmission path is retained, and risk transmission process information that meets the requirements is output.

[0066] The extracted risk transmission process information is deduplicated, removing redundant intermediate features, feature change nodes, and transmission path reversal information. Then, the compatibility of the deduplicated core information with the initial risk transmission path is analyzed. For example, it is determined whether the intermediate feature "transformer insulation performance degradation" can be reasonably integrated into the initial risk transmission path, located after "temperature increase" and before "protective device operation." The temporal sequence of feature change nodes is consistent with the temporal logic of the initial risk transmission path, such as temperature increase preceding insulation performance degradation. The transmission path reversal information is considered to conform to the possible evolution trend of the current risk; for example, if the current risk mainly affects the transformer, but the reversal information indicates the risk may affect adjacent circuit breakers, this is not realistic. Only the information that meets the requirements is retained, and the compliant risk transmission process information is output.

[0067] Steps S120-37: Insert intermediate features from the qualified risk transmission process information into the corresponding time nodes of the initial risk transmission path, supplement the identifiers of feature change nodes and the parameter change ranges, update the turning information of the transmission path, improve the link settings of the transmission path, and output the improved risk transmission path.

[0068] Based on the compliant risk transmission process information, intermediate features are inserted into the corresponding time nodes of the initial risk transmission path in chronological order of their appearance. An identifier is added to each feature change node, such as "Feature A Occurrence Node," and the range of parameter changes at that node is supplemented, such as temperature increasing from value X to value Y. Based on the transmission path reversal information, the direction of the risk transmission path and the equipment or systems involved are updated, and each link in the transmission path is improved to make the risk transmission path more complete and accurate. The improved risk transmission path is then output.

[0069] Step S120-4: Sort all features in the improved risk transmission path according to the time sequence of risk transmission, and construct a risk transmission sequence with the risk trigger source feature as the starting point, the final risk impact feature as the ending point, and the intermediate transmission features as nodes. Label each node with feature type, feature intensity, and occurrence time, and label the transmission rate and impact degree of the transmission link between nodes.

[0070] Based on the refined risk transmission path, all features are sorted according to the chronological order of risk transmission. The risk triggering source feature is used as the starting point of the sequence, and the final risk impact feature, such as "transformer tripping," is used as the ending point. Intermediate transmission features are designated as nodes to construct the risk transmission sequence. Each node is labeled with its feature type (e.g., temperature, current), feature intensity (i.e., the severity of the feature, such as mild, moderate, or severe), and the time of occurrence. The transmission links between nodes are labeled with their transmission rate (how quickly the risk propagates from one node to the next) and their degree of impact (i.e., the magnitude of the influence of the previous node's feature on the next node's feature).

[0071] Step S120-5: Construct an operation and maintenance risk evolution chain based on the risk transmission sequence, build the association logic and transmission rules of each node in the operation and maintenance risk evolution chain, and output a visualized operation and maintenance risk evolution chain structure.

[0072] Based on the risk transmission sequence, an operational risk evolution chain is constructed. The logical connections between each node are clearly defined, such as "the occurrence of feature A will lead to the occurrence of feature B," as well as the transmission rules, such as that feature A will only be transmitted to feature B when its intensity reaches a certain threshold. Through visualization techniques, such as using a directed graph, the nodes and transmission links in the operational risk evolution chain are displayed, forming a visualized operational risk evolution chain structure, allowing dispatchers to intuitively understand the risk evolution process.

[0073] Step S120-6: Simulate the entire evolution process of the operation and maintenance risk evolution chain using a multimodal AI large model, set different power grid dispatch operation scenario parameters, deduce the characteristic changes of each node in the operation and maintenance risk evolution chain under different scenarios, and output operation and maintenance risk evolution simulation data under different scenarios.

[0074] The multimodal AI model simulates the entire evolution process of an operational risk based on a constructed operational risk evolution chain. Different power grid dispatching operation scenario parameters are set, such as normal load scenarios, peak load scenarios, and equipment aging scenarios. Under each scenario, the model extrapolates the characteristic changes of each node in the operational risk evolution chain, such as changes in characteristic intensity and the earlier or later occurrence of events, outputting simulated operational risk evolution data for different scenarios.

[0075] Step S120-7: Record time node data, feature intensity change data, and impact range expansion data in the simulation data of operation and maintenance risk evolution under different scenarios; extract key nodes in the risk evolution process of the operation and maintenance risk evolution chain; and output key node information of operation and maintenance risk evolution.

[0076] In the simulation data of operational risk evolution in different scenarios, time node data is recorded, i.e., the time when each feature appears and changes; feature intensity change data, i.e., how the intensity of each feature changes over time; and impact scope expansion data, i.e., the expansion of the scope of equipment or systems affected by the risk. By analyzing this data, key nodes in the risk evolution process are extracted, such as nodes where the risk begins to spread rapidly and nodes that may lead to serious consequences, and key node information of operational risk evolution is output.

[0077] Step S120-8: Integrate simulation data of operation and maintenance risk evolution under different scenarios with key node information of operation and maintenance risk evolution to form risk status description information covering the entire stage of risk triggering, transmission, diffusion and attenuation. Link and bind the risk status description information with the operation and maintenance risk evolution chain so that each node in the operation and maintenance risk evolution chain corresponds to risk status data. Output the linked operation and maintenance risk evolution chain and risk status description information for the entire stage.

[0078] This approach integrates simulation data of operational risk evolution across different scenarios with key node information on operational risk evolution to form detailed descriptions of each stage of risk triggering, propagation, diffusion, and attenuation—essentially, risk status description information. For example, in the risk triggering stage, it describes the characteristics of the risk triggering source; in the propagation stage, it describes the propagation process between characteristics. This risk status description information is then linked and bound to each node in the operational risk evolution chain, ensuring that each node has corresponding risk status data, such as its risk status description and characteristic parameters. Finally, the linked operational risk evolution chain and full-stage risk status description information are output.

[0079] Step S130: Based on the risk evolution chain and the full-stage risk status description information, perform dynamic intervention strategy simulation and output multiple operation and maintenance intervention strategies adapted to different risk evolution nodes.

[0080] Based on the obtained risk evolution chain and risk status description information at all stages, corresponding dynamic intervention strategies are deduced for different nodes in the risk evolution process. These strategies aim to control the development of risks and reduce the losses caused by risks.

[0081] Step S130-1: Analyze the operation and maintenance risk evolution chain and the risk status description information of the whole stage. Divide the different stages of risk evolution of the operation and maintenance risk evolution chain with the key nodes in the operation and maintenance risk evolution chain as the boundary. Extract the risk characteristic parameters, risk transmission rate, risk impact range and related equipment and systems corresponding to each stage, and output the characteristic information of the risk evolution stage of each operation and maintenance risk evolution chain.

[0082] A detailed analysis of the operational risk evolution chain and the description of risk status at each stage is conducted. Using key nodes, such as the point where risk begins to spread or the point where the impact expands, the risk evolution chain is divided into different stages, such as the risk triggering stage, initial propagation stage, rapid spread stage, and stable decay stage. For each stage, corresponding risk characteristic parameters are extracted, such as the main risk characteristic type and intensity; risk propagation rate (how fast the risk propagates at this stage); risk impact range (the equipment and systems affected by the risk at this stage); and associated equipment and systems (the equipment and systems related to the risk at this stage). The characteristic information of each risk evolution stage is then output.

[0083] Step S130-2: Based on the risk triggering stage feature information in the risk evolution stage feature information of each operation and maintenance risk evolution chain, analyze the formation cause and initial diffusion trend of the risk triggering source feature, determine the core direction of intervention for the risk triggering stage, select intervention methods, extract the operation object, operation steps and implementation time window of each intervention method, and output the intervention elements of the risk triggering stage.

[0084] Steps S130-21 combine the associated data in the coupling feature set to trace the direct cause of the risk trigger source feature formation, analyze the specific manifestations and causes of parameter anomalies, the parameter indicators of equipment anomalies, the time point of anomaly occurrence and its associated impact with other equipment, and output the analysis results of the cause of risk trigger source feature formation.

[0085] By combining the associated data related to the risk triggering source characteristics in the coupling feature set, such as the equipment operating parameters, environmental factors, and system interactions at the time, the direct cause of the risk triggering source characteristics is traced. For example, if the risk triggering source characteristic is "the transformer temperature continues to rise and the cooling fan does not start," the analysis suggests that the temperature rise may be due to excessively high ambient temperature or excessive load, and the cooling fan not starting may be due to fan failure or a control loop problem. The specific manifestations of abnormal parameters are analyzed, such as how much the temperature exceeds the normal range, abnormal equipment parameters such as zero fan current, the time point of the anomaly, and the impact of the anomaly on other related equipment, such as whether it affects the temperature of adjacent equipment. The analysis results of the risk triggering source characteristic formation cause analysis are then output.

[0086] Step S130-22: Based on the analysis results of the cause of risk trigger source characteristics, predict its initial diffusion trend, including the main direction of diffusion, the possible associated characteristics, the change law of diffusion rate, and the possible characteristic variations during the diffusion process. Combined with the real-time status of power grid dispatch operation, determine the potential impact of initial diffusion on core equipment, key systems and overall dispatch stability of the power grid, and output the initial diffusion trend prediction results.

[0087] Based on the analysis of the causes of risk triggering, the initial diffusion trend is predicted. The main direction of diffusion may be from the triggering device to adjacent devices, or from a part of the device to the whole system. Potentially related characteristics may be affected, such as the possibility of decreased insulation performance due to increased temperature. The diffusion rate may vary, initially slow but gradually accelerating over time. Potential characteristic variations during diffusion may occur, such as a shift from temperature anomalies to pressure anomalies. Combining this with the real-time status of power grid dispatching and operation, such as current load levels and system stability, the potential impact of the initial diffusion on core power grid equipment (e.g., transformers, circuit breakers), critical systems (e.g., SCADA systems, EMS systems), and overall dispatching stability is assessed, and the initial diffusion trend prediction result is output.

[0088] Steps S130-23: Based on the initial diffusion trend prediction results, determine the core direction of intervention for this risk evolution stage, select intervention measures for adjusting command execution parameters based on the core direction of intervention, extract the operation objects, operation steps, and implementation time windows of the intervention measures for adjusting command execution parameters, and output the elements of the intervention measures for adjusting command execution parameters.

[0089] Based on the initial diffusion trend prediction results, the core direction of intervention is determined, such as reducing transformer temperature and restoring the function of cooling fans. Based on this core direction, intervention measures with adjustment command execution parameters are selected, such as adjusting the transformer's load distribution to reduce its current load. The target transformer is "BYQ-001". The operation steps include querying the current transformer load status, formulating a load adjustment plan, and issuing load adjustment commands to relevant dispatch terminals. The implementation time window refers to the period during which intervention must be completed before the risk spreads to a certain extent, such as within 30 minutes before the transformer temperature reaches the tripping threshold, outputting the elements of the adjustment command execution parameters intervention measures.

[0090] Steps S130-24: Select intervention measures to suspend the operation of related equipment based on the core direction of the intervention, extract the operation objects, operation steps, and implementation time windows of the intervention measures to suspend the operation of related equipment, and output the elements of the intervention measures to suspend the operation of related equipment.

[0091] If the risk triggering source involves the abnormal operation of a related piece of equipment, and the operation of that equipment would exacerbate the risk's spread, then suspending the operation of the related equipment can be selected as an intervention measure. For example, if a pump in a transformer's cooling system malfunctions, causing a decrease in heat dissipation, it may be necessary to suspend the operation of that pump and switch to a standby pump. The target of the operation is the faulty pump. The operation steps include checking the status of the standby pump, issuing a command to suspend the operation of the faulty pump, and starting the standby pump. The implementation time window also needs to be determined based on the risk's spread, outputting the elements of the intervention measure to suspend the operation of the related equipment.

[0092] Steps S130-25: Select the intervention method for shielding the interference source based on the core direction of the intervention, extract the operation object, operation steps, and implementation time window of the intervention method for shielding the interference source, and output the elements of the intervention method for shielding the interference source.

[0093] If the risk trigger is caused by environmental interference, such as abnormal equipment control signals due to surrounding electromagnetic interference, then the intervention method of shielding the environmental interference source should be selected. The target of the operation is the electromagnetic interference source, such as a nearby electromagnetic device. The operation steps include locating the electromagnetic interference source and taking shielding measures, such as installing a shielding cover or adjusting the equipment position. The implementation time window is to output the elements of the environmental interference source shielding intervention method before the interference causes more serious impact on the equipment.

[0094] Step S130-26: Prioritize the intervention methods of adjusting command execution parameters, suspending the operation of associated equipment, and shielding environmental interference sources; set the switching conditions between different intervention methods; and output the priority of intervention methods and switching conditions.

[0095] Based on factors such as the urgency, effectiveness, and impact on power grid operation, the selected intervention measures, including adjusting command execution parameters, suspending the operation of related equipment, and shielding environmental interference sources, are prioritized. For example, shielding environmental interference sources may have a higher priority if it can quickly eliminate the source of risk. Switching conditions between different intervention measures are set; for instance, if the risk is not controlled after implementing a certain intervention measure, the system switches to the next higher priority intervention measure. The priority of the intervention measure and the switching conditions are then output.

[0096] Step S130-27: Construct a coordinated plan for intervention measures such as adjusting command execution parameters, suspending the operation of associated equipment, and shielding environmental interference sources. This plan includes time connection rules, operation connection procedures, and data sharing specifications for the implementation of different measures, and outputs intervention elements for the risk triggering stage.

[0097] To improve intervention effectiveness, a coordinated approach for various intervention methods needs to be developed. Timing rules specify the order and time intervals for implementing different intervention methods, such as first shielding environmental interference sources before adjusting command execution parameters. Operational workflows ensure a smooth transition to the next intervention method after the previous one is completed; for example, after pausing related equipment, it's necessary to confirm that the equipment has safely stopped before proceeding with subsequent operations. Data sharing specifications ensure timely sharing of relevant data during the implementation of each intervention method; for example, after adjusting command execution parameters, the adjusted parameter data is shared with personnel implementing other relevant intervention methods, ultimately outputting the intervention elements for the risk-triggered stage.

[0098] Step S130-3: Based on the initial transmission stage feature information in the feature information of each operation and maintenance risk evolution stage, analyze the key nodes and transmission intensity of the risk transmission path of the operation and maintenance risk evolution chain, determine the core direction of intervention in the initial transmission stage, select intervention methods, extract the implementation sequence and mutual cooperation logic of the intervention methods, and output the intervention elements of the initial transmission stage.

[0099] In the initial transmission phase, key nodes in the risk transmission path are analyzed; these nodes are crucial hubs for risk transmission. The transmission strength of each node is assessed, i.e., the ability of the risk to be transmitted through that node. Based on these analyses, the core direction of intervention is determined, such as blocking transmission at key nodes or weakening its intensity. Appropriate intervention methods are selected, such as isolating key nodes or reducing the risk characteristic intensity at those nodes. The implementation sequence of intervention methods is extracted, such as intervening in key nodes with high transmission strength first, followed by addressing nodes with low transmission strength. The coordination logic refers to how different intervention methods work together, such as reducing the load on surrounding equipment after isolating nodes, outputting the intervention elements for the initial transmission phase.

[0100] Step S130-4: Based on the characteristic information of the rapid diffusion stage in the characteristic information of each operation and maintenance risk evolution stage, analyze the expansion law of the risk impact scope and the core impact objects, determine the core direction of intervention in the rapid diffusion stage, select intervention methods, construct the emergency implementation process and resource allocation plan for intervention methods, and output the intervention elements of the rapid diffusion stage.

[0101] During the rapid diffusion phase, the scope of risk impact expands rapidly, necessitating analysis of its expansion patterns, such as expansion along specific geographical areas or equipment types. Identifying the core affected objects—the equipment or systems most affected during risk diffusion—is crucial. The core intervention direction may be controlling the expansion of the risk's impact scope and protecting the core affected objects. Intervention measures should be selected, such as activating backup equipment or switching system operating modes. An emergency implementation process for intervention measures should be constructed, clearly defining the sequence of steps and responsible parties to ensure rapid and orderly intervention. Resource allocation plans involve the allocation of personnel, materials, and equipment to meet the needs of intervention implementation, thus outputting the intervention elements for the rapid diffusion phase.

[0102] Step S130-5: Based on the characteristic information of the stable decay stage in the characteristic information of each operation and maintenance risk evolution stage, analyze the decay law of risk intensity and residual risk characteristics, determine the core direction of intervention for the stable decay stage, select intervention methods, extract the implementation cycle and effect evaluation criteria of the intervention methods, and output the intervention elements of the stable decay stage.

[0103] During the stable decay phase, the risk intensity gradually decreases. Analyze the decay pattern of risk intensity, such as exponential or linear decay over time. Identify residual risk characteristics, i.e., risk characteristics that still exist after risk decay. The core direction of intervention is to accelerate risk decay and eliminate residual risk. Select intervention methods, such as continuous monitoring of risk characteristics and treatment of residual risk. Extract the implementation cycle of the intervention methods, i.e., the length of time that the intervention needs to be continuously implemented, and the effectiveness evaluation criteria, such as whether the intensity of residual risk characteristics has decreased below a safe threshold, and output the intervention elements for the stable decay phase.

[0104] Step S130-6: Based on the intervention elements of the risk triggering stage, the initial transmission stage, the rapid diffusion stage, and the stable decay stage, and combined with the technical specifications of power grid dispatch automation operation and maintenance, generate initial intervention strategies that are adapted to different risk evolution stages of the operation and maintenance risk evolution chain. Each draft contains the core content of intervention objectives, intervention measures, implementation steps, resource requirements, and time requirements, and outputs multiple initial intervention strategies.

[0105] By integrating intervention elements at each stage and combining them with technical specifications for automated operation and maintenance of power grid dispatch, such as operating procedures and safety standards, initial intervention strategies are generated for each risk evolution stage. Each draft initial intervention strategy includes the intervention objective, i.e., the desired effect of the strategy; the intervention measures, i.e., the specific intervention methods to be adopted; the implementation steps, i.e., the execution process of the intervention measures; the resource requirements, i.e., the personnel, equipment, materials, etc. required to implement the strategy; and the time requirements, i.e. the time constraints and time nodes in the implementation process, resulting in the output of multiple initial intervention strategies.

[0106] Step S130-7: Input multiple initial intervention strategies into the multimodal AI large model for simulation and verification, build a simulation environment consistent with the actual power grid dispatch and operation and maintenance scenario, simulate the changes in the operation and maintenance risk evolution chain during the implementation of the draft, record the impact data of intervention measures on risk characteristics, disturbance data on normal power grid operation, and resource consumption data, and output simulation and verification data.

[0107] Multiple initial intervention strategies are input into a multimodal AI model. The model constructs a simulation environment consistent with actual urban power grid dispatching and operation scenarios, including power grid structure, equipment parameters, and operating status. Within this environment, the implementation process of each initial intervention strategy is simulated, observing changes in the operation and maintenance risk evolution chain, such as whether the risk is controlled, and whether the risk evolution stage is advanced or delayed. Data on the impact of intervention measures on risk characteristics, such as whether the risk characteristic intensity decreases; data on disturbances to normal power grid operation, such as whether intervention measures cause power grid load fluctuations, voltage and current changes; and resource consumption data, such as the human, material, and time resources consumed in implementing intervention measures, are recorded. The simulation verification data is then output.

[0108] Step S130-8: Optimize the initial intervention strategy based on the simulation and verification data, adjust measures where the intervention effect data does not meet the preset target, optimize steps where the disturbance data exceeds the preset target, adjust schemes where the resource consumption data exceeds the preset target, and output the optimized operation and maintenance intervention strategy.

[0109] The initial intervention strategy is optimized based on the simulation and verification data. If the data related to the intervention effect of a certain intervention measure does not meet the preset target, such as insufficient reduction in the intensity of risk characteristics, the measure is adjusted, such as by adding intervention methods or adjusting intervention parameters. If the disturbance data to the normal operation of the power grid exceeds the preset allowable range during implementation, such as excessive load fluctuations, the corresponding implementation steps are optimized, such as by adjusting the operation sequence or reducing the operation intensity. If the resource consumption data exceeds the preset limit, such as excessive manpower input, the resource allocation plan is adjusted, such as by optimizing the division of labor among personnel or using more efficient equipment, and the optimized operation and maintenance intervention strategy is output.

[0110] Step S130-9: Integrate the optimized operation and maintenance intervention strategies for each risk evolution stage to form multiple operation and maintenance intervention strategies covering different risk evolution nodes. Each operation and maintenance intervention strategy includes the appropriate risk evolution stage, core intervention objectives, specific implementation steps, and effect evaluation criteria.

[0111] The optimized operational intervention strategies for each stage of risk evolution are integrated. Based on different nodes in the risk evolution chain, intervention strategies adapted to the stage of that node are combined to form multiple operational intervention strategies covering different risk evolution nodes. Each strategy clearly defines the risk evolution stage it is applicable to, the core intervention objective (i.e., the main goal the strategy aims to achieve at that stage), the specific implementation steps (i.e., the detailed operational process), and the effectiveness evaluation criteria (i.e., how to determine the effectiveness of the intervention strategy). These operational intervention strategies are ultimately output.

[0112] Step S140: Perform full-scenario performance simulation verification on multiple operation and maintenance intervention strategies, and output the optimal operation and maintenance intervention strategy covering the entire risk evolution process.

[0113] To ensure that the selected intervention strategies work effectively under various conditions, it is necessary to conduct full-scenario performance simulation verification of multiple operation and maintenance intervention strategies.

[0114] Step S140-1: Build a full-scenario simulation environment for power grid dispatching and maintenance risks. The full-scenario simulation environment for power grid dispatching and maintenance risks includes various typical scenarios such as normal operation scenario, peak load scenario, extreme environment scenario, and equipment aging scenario. Import the equipment parameters, system configuration, and historical operation data of the power grid dispatching system to restore the power grid dispatching operation status under different scenarios and output the full-scenario simulation environment for power grid dispatching and maintenance risks.

[0115] For example, in step S140-11, all equipment parameters of the power grid dispatching system are collected to establish an equipment parameter database. The total equipment parameters include the model parameters, rated voltage, rated current, rated power, operating temperature threshold, and vibration threshold of the core equipment.

[0116] Collect parameters of all core equipment in the urban power grid dispatching system, such as transformer model, rated voltage (high-voltage side and low-voltage side), rated current, rated power, operating temperature thresholds (including winding temperature thresholds and core temperature thresholds), vibration thresholds, etc.; circuit breaker model, rated voltage, rated current, rated breaking current, etc. Establish an equipment parameter database after organizing the above parameters.

[0117] Step S140-12: Collect system configuration information of the power grid dispatching system and establish a system configuration library. The system configuration information includes the connection relationship between the main system and subsystems, the communication protocol between systems, the data interaction interface parameters, the software version information of the system, and the security configuration parameters of the system.

[0118] Collect information on the connection relationships between the main scheduling system and subsystems such as the SCADA system and EMS system, including the network topology used for connection; communication protocols between systems, such as TCP / IP and IEC61850; data interaction interface parameters, such as interface type, data transmission rate, and data format; software version information of the system, such as operating system version and application version; and system security configuration parameters, such as firewall rules and access control policies, and establish a system configuration library.

[0119] Step S140-13: Organize the historical operation data of the power grid dispatching system and output the historical operation database. The historical operation database covers the operation data of different seasons, different time periods and different load states, including the overall operation parameters of the power grid, the real-time operation parameters of each device, the data flow parameters of system interaction and environmental parameters.

[0120] We have compiled operational data from the power grid dispatching system over the past few years for different seasons (spring, summer, autumn, and winter), different time periods (peak, flat, and off-peak), and different load conditions (light load, normal load, and heavy load). This data includes overall power grid operating parameters such as total active power, total reactive power, and frequency; real-time operating parameters of various devices such as transformer voltage, current, and temperature, and circuit breaker opening and closing status; data flow parameters of system interactions such as data transmission volume, transmission rate, and interaction frequency; and environmental parameters such as temperature, humidity, and wind speed, forming a historical operational database.

[0121] Steps S140-14: Based on the equipment parameter library, system configuration library, and historical operation database, a basic simulation structure is built. The basic simulation structure includes an equipment simulation module, a system interaction simulation module, an environment simulation module, and a data acquisition and analysis module.

[0122] Based on an equipment parameter library, system configuration library, and historical operation database, a basic simulation structure is built. The equipment simulation module simulates the operating status of various devices in the power grid, calculating the output parameters of the devices based on equipment parameters and operating data. The system interaction simulation module simulates the data interaction process between the main system and subsystems, and between subsystems, following the communication protocols and interface parameters in the system configuration library. The environmental simulation module simulates different environmental conditions, such as changes in temperature and humidity. The data acquisition and analysis module collects various data during the simulation process and performs preliminary analysis.

[0123] Step S140-15: Configure the normal operation scenario in the basic simulation structure. Refer to the operation data in the historical operation database where the power grid load is in the standard range, the environmental parameters are stable, and the equipment is without abnormality. Set the equipment operation parameter range, system interaction parameters, and environmental parameters for the scenario to restore the normal dispatch operation state of the power grid and output the simulation data of the normal operation scenario.

[0124] Referring to the historical operation database, the power grid load is within the standard range, i.e., the load is between 30% and 70% of the rated load, environmental parameters are stable (e.g., temperature between 20 and 25 degrees Celsius, humidity between 40% and 60%), and there is no abnormal equipment operation data. In the basic simulation structure, the equipment operation parameter ranges for the scenario are set, such as transformer voltage fluctuating within ±5% of the rated voltage, and current between 30% and 70% of the rated current; system interaction parameters, such as data transmission rate stable within a certain range, and interaction frequency normal; environmental parameters stable within the above ranges, recreating the normal power grid dispatch operation state, and outputting normal operation scenario simulation data.

[0125] Step S140-16: Configure peak load scenario. Refer to the operation data of the power grid load in the high load range in the historical operation database, set the equipment operation parameters of the scenario to be in the high load range, increase the amount of system interaction data, and increase the interaction frequency, simulate the operation status of the power grid during the peak electricity consumption period, and output peak load scenario simulation data.

[0126] Referencing historical operational databases, the power grid load is set to be in a high-load range, meaning the load is above 70% of the rated load. The scenario assumes that equipment operating parameters are in a high-load range, such as transformer current approaching or reaching the rated current, and a corresponding increase in temperature; increased system interaction data, such as increased telemetry and telecontrol data transmitted from the SCADA system to the EMS system; and increased interaction frequency, such as an increase in the number of interactions per unit time. This simulates the power grid's operation during peak electricity consumption periods, such as a summer evening, and outputs simulated peak load scenario data.

[0127] Step S140-17: Configure extreme environment scenarios, select typical extreme environment types, set environmental parameters to exceed the normal range, refer to historical power grid operation data under extreme environments, set the fluctuation range of equipment operation parameters, simulate the impact of extreme environments on the operating status of power grid equipment and the stability of system interaction, and output extreme environment scenario simulation data.

[0128] Typical extreme environmental types are selected, such as high temperature, extreme cold, torrential rain, and strong winds. Environmental parameters are set to exceed normal ranges; for example, in high-temperature environments, the temperature is set above 40 degrees Celsius, and in strong-wind environments, the wind force is set to level 8 or higher. Referring to historical power grid operation data under extreme conditions, the fluctuation range of equipment operating parameters is set. For example, in high-temperature environments, the allowable temperature fluctuation range for transformers is larger, potentially approaching the tripping threshold; in strong-wind environments, the vibration parameters of transmission lines fluctuate more. The impact of extreme environments on the operating status of power grid equipment is simulated, such as increased equipment failure rates, and the impact on system interaction stability, such as increased probability of data transmission interruptions. Simulated data of extreme environment scenarios are output.

[0129] Step S140-18: Configure the equipment aging scenario, select equipment with an operating life close to the design life as the simulation object, refer to the parameter change pattern after equipment aging, adjust the operating parameter threshold of the equipment, simulate the operating state of increased parameter fluctuation and increased failure probability caused by equipment aging, and output the equipment aging scenario simulation data.

[0130] Equipment nearing the design life of its operational years in the urban power grid, such as transformers that have been in operation for 20 years (design life of 25 years), is selected as simulation objects. Referring to the parameter changes observed in this type of equipment after aging, such as decreased insulation resistance and increased losses, the operating parameter thresholds are adjusted. For example, the temperature tripping threshold for transformers is appropriately lowered because aging equipment has reduced temperature tolerance. The simulation depicts increased parameter fluctuations due to equipment aging, such as larger voltage and current fluctuations compared to normal equipment, leading to an increased probability of failure. The simulation data for the equipment aging scenario is then output.

[0131] Step S140-19: Import the operation and maintenance risk evolution data from the historical operation database, reproduce the evolution process of different types of operation and maintenance risks in normal operation scenarios, peak load scenarios, extreme environment scenarios, and equipment aging scenarios, adjust the parameter settings of the simulation environment, so that the simulated risk triggering, transmission, and diffusion process matches the historical actual risk process to a preset standard, and output the scenario parameter adjustment results.

[0132] The evolution data of operational risks from historical operation databases, such as past transformer failures and line trips, are imported into various scenarios. The evolution process of different types of operational risks is reproduced in normal operation scenarios, peak load scenarios, extreme environment scenarios, and equipment aging scenarios. By adjusting the parameters of the simulation environment, such as equipment failure probability and risk propagation rate, the simulated risk triggering, propagation, and diffusion processes are made to match the historical actual risk processes to a preset standard, such as above 85%, and the scenario parameter adjustment results are output.

[0133] Steps S140-110 involve deploying data monitoring nodes in normal operation scenarios, peak load scenarios, extreme environment scenarios, and equipment aging scenarios to collect equipment operating parameters, system interaction data, environmental parameters, and risk characteristic data in real time during the simulation process, and outputting scenario simulation monitoring data.

[0134] In each scenario, data monitoring nodes are deployed at key equipment, system interfaces, and environmental monitoring points. These nodes collect real-time data on equipment operating parameters during the simulation process, such as voltage, current, and temperature; system interaction data, such as data transmission rate, interaction frequency, and data integrity; environmental parameters, such as temperature, humidity, and wind speed; and risk characteristic data, such as the type, intensity, and occurrence time of risk characteristics. The output scenario simulation monitoring data is used to evaluate the accuracy of the scenario simulation and to verify its effectiveness in subsequent operations.

[0135] Steps S140-111 involve optimizing and adjusting the problems existing in the simulation environment, correcting equipment parameter settings, system interaction rules, and environmental impact models, until each typical scenario can restore the corresponding power grid dispatch operation status, and outputting a full-scenario simulation environment for power grid dispatch operation and maintenance risks.

[0136] Based on the scenario simulation monitoring data and scenario parameter adjustment results, problems existing in the simulation environment are analyzed, such as unreasonable equipment parameter settings leading to large deviations between simulation results and reality, system interaction rules not conforming to actual conditions, and inaccurate simulation of the impact of environmental influence models on equipment operating status. These problems are addressed through optimization and adjustments, including correcting equipment parameter settings (e.g., adjusting transformer loss parameters), improving system interaction rules (e.g., modifying the data transmission timeout retransmission mechanism), and optimizing environmental influence models (e.g., more accurately simulating the impact of temperature on equipment heat dissipation). After multiple adjustments and verifications, until each typical scenario can accurately reproduce the corresponding power grid dispatching operation status, a full-scenario simulation environment for power grid dispatching and maintenance risks is finally output.

[0137] Step S140-2: Reproduce the entire evolution process of the operation and maintenance risk evolution chain in the full-scenario simulation environment of power grid dispatch and operation and maintenance risk, and output the simulation results of the entire process of operation and maintenance risk evolution.

[0138] In the established full-scenario simulation environment of power grid dispatch and operation and maintenance risks, input the previously obtained operation and maintenance risk evolution chain and full-stage risk status description information, start the simulation environment, reproduce the entire process of the operation and maintenance risk evolution chain, record various data in the entire process from risk triggering, transmission, diffusion to decay, etc., such as risk characteristic changes, equipment status changes, system interaction changes, etc., and output the full-process simulation results of operation and maintenance risk evolution.

[0139] Step S140-3: Import multiple operation and maintenance intervention strategies covering different risk evolution nodes of the operation and maintenance risk evolution chain into the full-scenario simulation environment of power grid dispatch operation and maintenance risk. Implement corresponding intervention measures in sequence according to the risk evolution stage of the operation and maintenance risk evolution chain, record key data in the implementation process of each operation and maintenance intervention strategy, and output the implementation process data of each operation and maintenance intervention strategy.

[0140] Multiple operation and maintenance intervention strategies adapted to different risk evolution nodes are imported into a full-scenario simulation environment for power grid dispatch operation and maintenance risks. Within the simulation environment, intervention measures are implemented sequentially for each stage of the operation and maintenance risk evolution chain. For example, intervention measures for the risk triggering stage are implemented during the risk triggering stage, and intervention measures for the initial propagation stage are implemented during the initial propagation stage. Key data during the implementation of each operation and maintenance intervention strategy are recorded, such as the implementation time, implementation effect, resource consumption, and impact on power grid operation, and the implementation process data for each strategy is output.

[0141] Step S140-4: Obtain a set of performance evaluation indicators, which includes intervention effect, operational disturbance, resource consumption, and implementation efficiency.

[0142] The set of performance evaluation indicators is used to measure the effectiveness of operation and maintenance intervention strategies. Intervention effect refers to the degree to which the intervention strategy controls risks, such as whether risks are eliminated or whether the risk intensity is reduced to a safe level; operational disturbance refers to the degree of impact on the normal operation of the power grid during the implementation of the intervention strategy, such as whether it causes power grid load fluctuations or voltage and current anomalies; resource consumption refers to the various resources consumed in implementing the intervention strategy, such as manpower, material resources, and time; implementation efficiency refers to the length of time from the formulation of the intervention strategy to its completion and the smoothness of the implementation process.

[0143] Step S140-5: Based on the set of performance evaluation indicators, perform a comprehensive performance score on the implementation process data of each set of operation and maintenance intervention strategies, calculate the weighted sum of the scores of each performance evaluation indicator, and output the comprehensive performance score of each set of operation and maintenance intervention strategies.

[0144] Each indicator in the effectiveness evaluation indicator set is assigned a specific weight, determined based on its importance; for example, the weight for intervention effectiveness may be the highest. Then, based on the implementation data of each operation and maintenance intervention strategy, each indicator is scored. A higher score is awarded for better intervention effectiveness and less operational disruption. The weighted sum of the scores for each indicator is calculated to obtain the comprehensive effectiveness score for each operation and maintenance intervention strategy. A higher comprehensive effectiveness score indicates better overall effectiveness of the strategy.

[0145] Step S140-6: Select the operation and maintenance intervention strategy with the highest comprehensive performance score, verify the adaptability of the operation and maintenance intervention strategy with the highest comprehensive performance score in different scenarios, output the adaptability verification results, and based on the adaptability verification results, fine-tune the intervention measures of the corresponding risk evolution stage of the operation and maintenance risk evolution chain in the operation and maintenance intervention strategy with the highest comprehensive performance score, optimize the intervention parameters, adjust the implementation steps or supplement auxiliary intervention methods, and output the fine-tuned operation and maintenance intervention strategy.

[0146] The strategy with the highest overall effectiveness score is selected from all operation and maintenance intervention strategies. Then, its adaptability is verified in different scenarios within a full-scenario simulation environment of power grid dispatch and operation and maintenance risks, such as normal operation, peak load, extreme environment, and equipment aging. The strategy's effectiveness in controlling risks under different scenarios is observed, and indicators such as operational disturbances, resource consumption, and implementation efficiency are assessed, outputting the adaptability verification results. If adaptability is poor in certain scenarios, such as a decrease in intervention effectiveness in extreme environment scenarios, the intervention measures corresponding to the risk evolution stage in the strategy are fine-tuned. This could involve optimizing intervention parameters, adjusting the order of implementation steps, or supplementing auxiliary intervention methods, such as increasing the investment in backup equipment. The fine-tuned operation and maintenance intervention strategy is then output.

[0147] Step S140-7: Perform full-scenario performance verification on the fine-tuned operation and maintenance intervention strategy again, output the verification results, and determine the optimal operation and maintenance intervention strategy that covers the evolution process of operation and maintenance risks based on the verification results.

[0148] The fine-tuned operation and maintenance intervention strategy is then validated again in a full-scenario simulation environment. Its intervention effect, operational disturbance, resource consumption, and implementation efficiency are evaluated using the same methods as before, and the validation results are output. If the validation results show that the strategy achieves good performance in all scenarios, then it is determined to be the optimal operation and maintenance intervention strategy covering the entire risk evolution process.

[0149] Step S150: The optimal operation and maintenance intervention strategy is transformed into a standardized sequence of execution instructions, pushed to the power grid dispatch automation operation and maintenance execution terminal, and the entire execution process is tracked and the operation and maintenance intervention results are output.

[0150] The selected optimal operation and maintenance intervention strategy is transformed into a standardized instruction sequence that can be recognized and executed by the power grid dispatch automation operation and maintenance execution terminal, and then pushed for execution. At the same time, the execution process is tracked, and finally the operation and maintenance intervention result is output.

[0151] Step S150-1: Analyze the optimal operation and maintenance intervention strategy, extract the core execution elements in the optimal operation and maintenance intervention strategy, and convert the core execution elements into structured instruction fragments in accordance with the instruction receiving format requirements of the power grid dispatch automation operation and maintenance execution terminal. Each instruction fragment contains instruction type, operation object, operation parameters, execution time, and key information on completion standards.

[0152] A detailed analysis of the optimal operation and maintenance intervention strategy is conducted, extracting its core execution elements, such as the type of intervention measure, the targeted equipment, the parameters to be adjusted, the execution time requirements, and the completion criteria. Referring to the instruction reception format requirements of power grid dispatch automation operation and maintenance execution terminals, such as the RTU in a substation and the execution module of the dispatch master station, these core execution elements are transformed into structured instruction fragments. Each instruction fragment includes: instruction type (e.g., control instruction, regulation instruction); operation object (e.g., specific equipment number); operation parameters (e.g., the voltage and current values ​​to be adjusted); execution time (the specific time the instruction needs to be executed); and completion criteria (the state or indicator to be achieved after the instruction is executed, such as the equipment temperature dropping below a certain value).

[0153] Step S150-2: According to the implementation sequence requirements of intervention measures in the optimal operation and maintenance intervention strategy, sort the structured instruction fragments, mark the execution priority of each instruction and its relationship with other instructions, and output the preliminary execution instruction sequence.

[0154] The intervention measures in the optimal operation and maintenance intervention strategy have certain implementation timing requirements, such as certain instructions needing to be executed first and others needing to be executed later. Based on these timing requirements, the structured instruction fragments are sorted. Simultaneously, the execution priority of each instruction is marked, such as urgent instructions having high priority and ordinary instructions having low priority; and its relationship with other instructions is also marked, such as instruction A must be executed only after instruction B is completed, or instructions C and D can be executed in parallel, outputting a preliminary sequence of execution instructions.

[0155] Step S150-3: Standardize the initial execution instruction sequence and output a standardized execution instruction sequence that meets the requirements of the execution end.

[0156] The initial execution instruction sequence undergoes standardization, including format unification, syntax checking, and semantic verification. This ensures that each instruction in the sequence conforms to the receiving requirements of the power grid dispatch automation operation and maintenance execution terminal, such as instruction encoding format, data type, and length limits. After standardization, a standardized execution instruction sequence is output.

[0157] Step S150-4: Push a standardized execution instruction sequence to the power grid dispatch automation operation and maintenance execution terminal. During the push process, monitor the data transmission status in real time, output data transmission status information, and after receiving the instruction reception confirmation information from the power grid dispatch automation operation and maintenance execution terminal, start the execution status monitoring process and collect the instruction execution monitoring information from the power grid dispatch automation operation and maintenance execution terminal in real time.

[0158] Standardized execution command sequences are pushed to the power grid dispatch automation operation and maintenance execution terminal via a dedicated power communication network. During the push process, the data transmission status is monitored in real time, such as whether the data transmission is normal, whether there is packet loss or delay, and the data transmission status information is output. When the execution terminal receives the command reception confirmation information, that is, the execution terminal has successfully received the command sequence, the execution status monitoring process is initiated. Through communication with the execution terminal, command execution monitoring information is collected in real time, such as the command execution progress, execution status (in progress, completed, execution failed, etc.), and equipment operating parameters during the execution process.

[0159] Step S150-5: Compare the instruction execution monitoring information with the completion criteria in the standardized execution instruction sequence. If an instruction execution abnormality is detected or the completion criteria are not met, an abnormality handling instruction is pushed and the abnormality handling instruction push result is output.

[0160] The system compares the real-time collected command execution monitoring information with the completion criteria in the standardized command sequence. If an execution anomaly is detected, such as execution time exceeding the preset time, equipment operating parameters not changing as expected, or failure to meet the completion criteria (e.g., equipment temperature not decreasing to the target value), an anomaly handling command is immediately pushed to the execution terminal. This command can include adjusting operating parameters, restarting the execution step, or switching to a backup plan. The system outputs the result of the anomaly handling command push, indicating whether the command was successfully pushed and whether the execution terminal received it.

[0161] Step S150-6: After all standardized execution instructions have been completed, collect the operation status data and residual risk characteristic data of the power grid dispatching system, and output the operation and maintenance intervention results.

[0162] After all standardized execution instructions have been completed, the system collects operational status data from the power grid dispatching system, such as parameters like voltage, current, and temperature of each device, and indicators like system load and frequency; as well as residual risk characteristic data, i.e., whether any risk characteristics remain in the risk evolution chain, and the strength of those residual characteristics. This data is then processed and analyzed to generate operation and maintenance intervention results, such as whether the risk has been successfully controlled and whether the power grid operation has returned to normal, and these results are then output.

[0163] Based on the same inventive concept, please refer to Figure 2 This paper shows a schematic block diagram of a power grid dispatch automation intelligent operation and maintenance system 100 based on a multimodal AI large model, which is provided in an embodiment of this application for executing the above-described power grid dispatch automation intelligent operation and maintenance method based on a multimodal AI large model. The power grid dispatch automation intelligent operation and maintenance system 100 based on a multimodal AI large model may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0164] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located within the power grid dispatch automation intelligent operation and maintenance system 100 based on a multimodal AI large model and are separately configured. However, it should be understood that the machine-readable storage medium 120 may also be independent of the power grid dispatch automation intelligent operation and maintenance system 100 based on a multimodal AI large model and may be accessed by the processor 130 via a bus interface. Alternatively, the machine-readable storage medium 120 may also be integrated into the processor 130 and may communicate and interact with external systems through the communication unit 110.

[0165] The processor 130 is the control center of the power grid dispatch automation intelligent operation and maintenance system 100 based on a multimodal AI big data model. It connects various parts of the system via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the machine-readable storage medium 120, and by calling data stored in the machine-readable storage medium 120, thereby providing overall monitoring of the system. Optionally, the processor 130 may include one or more processing cores; for example, the processor 130 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor. The machine-readable storage medium 120 is used to store machine-executable instructions for executing the scheme of this application, and the processor 130 is used to execute the machine-executable instructions stored in the machine-readable storage medium 120 to realize the power grid dispatching automated intelligent operation and maintenance method based on a multimodal AI large model provided in the aforementioned method embodiments.

[0166] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model, characterized in that, The method includes: Collect multimodal information on power grid dispatching and maintenance, and perform coupling processing on the multimodal information on power grid dispatching and maintenance to output a coupling feature set. The multimodal information on power grid dispatching and maintenance includes power grid dispatching instruction information, equipment operating status information, system interaction information, and environmental impact information. The coupled feature set is input into a multimodal AI model to perform operation and maintenance risk chain mining and preliminary inference, and outputs risk evolution chain and full-stage risk status description information; Based on the risk evolution chain and the full-stage risk status description information, dynamic intervention strategy simulation is performed, and multiple operation and maintenance intervention strategies adapted to different risk evolution nodes are output. Perform full-scenario performance simulation verification on multiple operation and maintenance intervention strategies, and output the optimal operation and maintenance intervention strategy covering the entire risk evolution process; The optimal operation and maintenance intervention strategy is transformed into a standardized sequence of execution instructions, which is then pushed to the power grid dispatch automation operation and maintenance execution terminal and the entire execution process is tracked to output the operation and maintenance intervention results. The process of inputting the coupled feature set into a multimodal AI large model to perform operation and maintenance risk chain mining and preliminary inference, and outputting risk evolution chain and full-stage risk status description information, includes: The coupled feature set is input into the feature recognition unit of the multimodal AI large model. Through the preset risk feature comparison rules, coupled features with risk triggering tendency are screened out, risk triggering source features are determined, the power grid dispatching and maintenance scenarios corresponding to the risk triggering source features are marked, the initial parameters, occurrence time and related data of the risk triggering source features are recorded, and the risk triggering source features and related information are output. Based on the risk trigger source characteristics and related information, the association mining unit of the multimodal AI model is used to mine the transmission relationship between the risk trigger source and other features in the coupled feature set, analyze the influence intensity and transmission time between different features, determine the initial transmission direction after the risk is triggered, and output the initial risk transmission path. Import historical risk data of power grid dispatch automation operation and maintenance, screen historical risk cases that are consistent with the characteristics of risk triggering source, extract risk transmission process data from the historical risk cases, supplement them into the initial risk transmission path, and output the improved risk transmission path; All features in the improved risk transmission path are sorted according to the time sequence of risk transmission. A risk transmission sequence is constructed with the risk trigger source feature as the starting point, the final risk impact feature as the ending point, and the intermediate transmission features as nodes. Each node is labeled with feature type, feature strength, and occurrence time. The transmission link between nodes is labeled with transmission rate and impact degree. Based on the risk transmission sequence, an operation and maintenance risk evolution chain is constructed, and the correlation logic and transmission rules of each node in the operation and maintenance risk evolution chain are constructed, and a visualized operation and maintenance risk evolution chain structure is output. By simulating the entire evolution process of the operation and maintenance risk evolution chain through a multimodal AI large model, different power grid dispatch operation scenario parameters are set to deduce the characteristic changes of each node in the operation and maintenance risk evolution chain under different scenarios, and output operation and maintenance risk evolution simulation data under different scenarios. Record time node data, feature intensity change data, and impact range expansion data in the simulation data of operation and maintenance risk evolution under different scenarios, extract key nodes in the risk evolution process of the operation and maintenance risk evolution chain, and output key node information of operation and maintenance risk evolution. By integrating simulation data of operation and maintenance risk evolution under different scenarios with key node information of operation and maintenance risk evolution, risk status description information covering the entire stage of risk triggering, transmission, diffusion and attenuation is formed. The risk status description information is associated and bound with the operation and maintenance risk evolution chain, so that each node in the operation and maintenance risk evolution chain corresponds to risk status data, and the associated operation and maintenance risk evolution chain and risk status description information of the entire stage are output.

2. The method for automated intelligent operation and maintenance of power grid dispatching based on a multimodal AI large model according to claim 1, characterized in that, The process involves collecting multimodal information on power grid dispatching and maintenance, and coupling this information to output a coupled feature set, including: Data collection nodes are deployed to collect multi-dimensional information to cover the entire scenario of power grid dispatching and maintenance. The data collection nodes are respectively set in the power grid dispatching command issuing terminal, equipment operation status monitoring port, system interaction data interface and environmental parameter sensing area. The collection scope includes the issuance time, transmission path and execution feedback of power grid dispatching commands, voltage parameters, current parameters, temperature parameters and vibration parameters of equipment operation, data transmission rate, interaction frequency and data integrity of system interaction, and environmental impacts such as temperature change, humidity change, wind force level, precipitation intensity and surrounding electromagnetic interference intensity. After all data collection nodes are synchronously connected to a unified time base for data collection, multi-dimensional collected information is output. The system performs full-process trajectory restoration processing on the power grid dispatch instruction information collected from multiple dimensions, tracks the complete transmission link of each power grid dispatch instruction from the issuing entity to the execution terminal, records the instruction reception time, response delay and data verification results of each node in the complete transmission link, extracts the associated parameters in the instruction execution process, and outputs the full-process trajectory data of instruction execution. The system performs correlation analysis on the equipment operation status information collected from multiple dimensions, selects the core equipment in the power grid dispatching system as the analysis subject, tracks the status parameter change curves of a single device in different operating periods, compares the parameter fluctuation correlation of different devices under the same dispatching task, extracts the collaborative change characteristics and abnormal fluctuation correlation characteristics of the equipment operation status, and outputs the equipment operation correlation status data. Perform transmission path mining processing on the system interaction information collected from multiple dimensions, extract the interaction links between the main power grid dispatching system and subsystems, and between subsystems, record the interaction data type, data transmission volume, interaction response time and interaction failure records of each interaction link, analyze the transmission impact of any system interaction anomaly on other related systems, extract the transmission correlation characteristics of system interaction, and output system interaction transmission path data; The system performs effect analysis and processing on environmental impact information collected from multiple dimensions, distinguishes the impact of natural and anthropogenic environmental factors on power grid dispatching and operation, tracks the fluctuation of power grid equipment operating parameters when a single environmental factor changes, analyzes the comprehensive impact effect of multiple environmental factors superimposed, extracts the correlation characteristics between environmental factors and power grid operating status, and outputs environmental impact data. Standardize the entire process trajectory data of instruction execution, equipment operation status data, system interaction transmission path data, and environmental impact data. Then, use a multimodal coupling algorithm to mine the interaction relationships between the standardized data and extract the coupling features formed by different data combinations. The extracted coupling features are subjected to temporal filtering to retain dynamic coupling features that change with time and remove static coupling features that have no risk. The output is a set of coupling features containing various dynamic correlation information.

3. The method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model according to claim 1, characterized in that, The dynamic intervention strategy simulation based on the risk evolution chain and full-stage risk status description information outputs multiple operation and maintenance intervention strategies adapted to different risk evolution nodes, including: The system analyzes the risk evolution chain and the risk status description information of the entire stage of the operation and maintenance risk. It divides the risk evolution of the operation and maintenance risk chain into different stages based on the key nodes in the chain. It extracts the risk characteristic parameters, risk transmission rate, risk impact range and related equipment and systems corresponding to each stage, and outputs the characteristic information of the risk evolution stage of each operation and maintenance risk chain. Based on the characteristic information of the risk triggering stage in the characteristic information of each operation and maintenance risk evolution chain, we analyze the formation cause and initial diffusion trend of the risk triggering source characteristics, determine the core direction of intervention for the risk triggering stage, select intervention methods, extract the operation object, operation steps and implementation time window of each intervention method, and output the intervention elements of the risk triggering stage. Based on the characteristic information of the initial transmission stage in the risk evolution stage of each operation and maintenance risk evolution chain, we analyze the key nodes and transmission intensity of the risk transmission path of the operation and maintenance risk evolution chain, determine the core direction of intervention in the initial transmission stage, select intervention methods, extract the implementation sequence and mutual cooperation logic of the intervention methods, and output the intervention elements of the initial transmission stage. Based on the characteristic information of the rapid diffusion stage in the risk evolution stage of each operation and maintenance risk evolution chain, we analyze the expansion pattern of the risk impact scope and the core impact objects, determine the core direction of intervention in the rapid diffusion stage, select intervention methods, construct the emergency implementation process and resource allocation plan for intervention methods, and output the intervention elements of the rapid diffusion stage. Based on the characteristic information of the stable decay stage in the risk evolution stage of each operation and maintenance risk evolution chain, we analyze the decay law of risk intensity and residual risk characteristics, determine the core direction of intervention in the stable decay stage, select intervention methods, extract the implementation cycle and effect evaluation criteria of the intervention methods, and output the intervention elements of the stable decay stage. Based on intervention elements for the risk triggering stage, initial propagation stage, rapid diffusion stage, and stable decay stage, and combined with the technical specifications for automated operation and maintenance of power grid dispatch, initial intervention strategies are generated to adapt to different risk evolution stages of the operation and maintenance risk evolution chain. Each draft contains core content such as intervention objectives, intervention measures, implementation steps, resource requirements, and time requirements, and outputs multiple initial intervention strategies. Multiple initial intervention strategies are input into a multimodal AI model for simulation and verification. A simulation environment consistent with the actual power grid dispatch and operation and maintenance scenario is built to simulate the changes in the operation and maintenance risk evolution chain during the implementation of the draft. Data on the impact of intervention measures on risk characteristics, disturbance data on normal power grid operation, and resource consumption data are recorded, and simulation and verification data are output. Based on the simulation and verification data, optimize the initial intervention strategy, adjust the measures for intervention effect data that do not meet the preset target, optimize the steps for disturbance data that exceed the preset target during implementation, adjust the schemes for resource consumption data that exceed the preset target, and output the optimized operation and maintenance intervention strategy. By integrating the optimized operation and maintenance intervention strategies for each stage of risk evolution, multiple operation and maintenance intervention strategies covering different risk evolution nodes are formed. Each operation and maintenance intervention strategy includes the appropriate risk evolution stage, core intervention objectives, specific implementation steps, and effect evaluation criteria.

4. The method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model according to claim 1, characterized in that, The process involves performing full-scenario performance simulations and verifications on multiple operation and maintenance intervention strategies, outputting the optimal operation and maintenance intervention strategy covering the entire risk evolution process, including: A full-scenario simulation environment for power grid dispatching and maintenance risks is constructed. The full-scenario simulation environment for power grid dispatching and maintenance risks includes various typical scenarios such as normal operation scenario, peak load scenario, extreme environment scenario, and equipment aging scenario. The equipment parameters, system configuration, and historical operation data of the power grid dispatching system are imported to restore the power grid dispatching operation status under different scenarios and output the full-scenario simulation environment for power grid dispatching and maintenance risks. The entire evolution process of the operation and maintenance risk evolution chain is reproduced in the full-scenario simulation environment of power grid dispatch and operation and maintenance risk, and the simulation results of the entire operation and maintenance risk evolution process are output. Multiple operation and maintenance intervention strategies covering different risk evolution nodes of the operation and maintenance risk evolution chain are imported into the full-scenario simulation environment of power grid dispatch operation and maintenance risk. Corresponding intervention measures are implemented in sequence according to the risk evolution stage of the operation and maintenance risk evolution chain. Key data in the implementation process of each operation and maintenance intervention strategy are recorded, and the implementation process data of each operation and maintenance intervention strategy is output. Obtain a set of performance evaluation indicators, which includes intervention effect, operational disturbance, resource consumption, and implementation efficiency; Based on the set of performance evaluation indicators, a comprehensive performance score is given to the implementation process data of each set of operation and maintenance intervention strategies. The weighted sum of the scores of each performance evaluation indicator is calculated, and the comprehensive performance score of each set of operation and maintenance intervention strategies is output. Select the operation and maintenance intervention strategy with the highest comprehensive performance score, verify the adaptability of the operation and maintenance intervention strategy with the highest comprehensive performance score in different scenarios, output the adaptability verification results, and based on the adaptability verification results, fine-tune the intervention measures of the corresponding risk evolution stage of the operation and maintenance risk evolution chain in the operation and maintenance intervention strategy with the highest comprehensive performance score, optimize the intervention parameters, adjust the implementation steps or supplement auxiliary intervention methods, and output the fine-tuned operation and maintenance intervention strategy. The fine-tuned operation and maintenance intervention strategy is re-verified across all scenarios, and the re-verification results are output. Based on the re-verification results, the optimal operation and maintenance intervention strategy covering the entire risk evolution process of the operation and maintenance risk evolution chain is determined.

5. The method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model according to claim 1, characterized in that, The process involves transforming the optimal operation and maintenance intervention strategy into a standardized sequence of execution instructions, pushing it to the power grid dispatch automation operation and maintenance execution terminal, tracking the entire execution process, and outputting the operation and maintenance intervention results, including: The optimal operation and maintenance intervention strategy is analyzed, the core execution elements in the optimal operation and maintenance intervention strategy are extracted, and the core execution elements are transformed into structured instruction fragments in accordance with the instruction receiving format requirements of the power grid dispatch automation operation and maintenance execution terminal. Each instruction fragment contains instruction type, operation object, operation parameters, execution time and key information of completion standard. Based on the implementation sequence requirements of intervention measures in the optimal operation and maintenance intervention strategy, the structured instruction fragments are sorted, the execution priority of each instruction and its relationship with other instructions are marked, and a preliminary execution instruction sequence is output. The initial sequence of execution instructions is standardized to output a standardized sequence of execution instructions that meets the requirements of the execution end. A standardized sequence of execution instructions is pushed to the power grid dispatch automation operation and maintenance execution terminal. During the push process, the data transmission status is monitored in real time, and the data transmission status information is output. After receiving the instruction reception confirmation information from the power grid dispatch automation operation and maintenance execution terminal, the execution status monitoring process is initiated to collect the instruction execution monitoring information from the power grid dispatch automation operation and maintenance execution terminal in real time. If the execution of the instruction is not normal or the completion standard is not met, the execution of the instruction is compared with the completion standard in the standardized execution instruction sequence. If an instruction execution error is detected or the completion standard is not met, an error handling instruction is pushed and the error handling instruction push result is output. After all standardized execution instructions have been completed, the system collects operational status data and residual risk characteristic data of the power grid dispatching system, and outputs operation and maintenance intervention results.

6. The method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model according to claim 2, characterized in that, The process involves performing full-process trajectory reconstruction processing on the power grid dispatch instruction information collected from multi-dimensional data. This process tracks the complete transmission link of each power grid dispatch instruction from the issuing entity to the execution terminal, records the instruction reception time, response delay, and data verification results of each node in the complete transmission link, extracts relevant parameters during instruction execution, and outputs the full-process trajectory data of instruction execution, including: Extract the unique identifier code of each power grid dispatch instruction from the power grid dispatch instruction information collected from multiple dimensions. Use the unique identifier code as an index to associate the information of each link of instruction issuance, transmission, reception and execution, establish an instruction full-process information association table, and output the instruction full-process information association table. Trace the information of the instruction issuance process, determine the equipment number of the instruction issuing entity, the issuance timestamp, and the power grid operation load status at the time of issuance, and enter the issuance process field of the instruction full process information association table, and output the first instruction full process information association table; Track the information of the instruction transmission process, track all transmission nodes that the power grid dispatch instruction passes through from the issuing body to the execution terminal and record the node sequence, record the equipment number of each transmission node, the timestamp of the power grid dispatch instruction entering the transmission node, the timestamp of the power grid dispatch instruction leaving the transmission node, the data verification result of the transmission node for the power grid dispatch instruction, calculate the instruction response delay of each transmission node, and enter the transmission process field of the first instruction full process information association table, and output the second instruction full process information association table. Extract the instruction receiving information, determine the device number of the instruction receiving terminal, the receiving timestamp, the integrity status of the instruction data at the time of reception, and the feedback information of the receiving terminal. If there is missing or incorrect instruction data, record the specific fields of the missing or incorrect data and the supplementary or corrective measures, and enter the receiving stage field of the second instruction full process information association table, and output the third instruction full process information association table. Collect information on the instruction execution process, track the start execution timestamp after the execution terminal receives the instruction, the timestamp of key operation nodes during the execution process, and the execution completion timestamp, determine the scheduling task type corresponding to the instruction, record the device number of the instruction, the changes in the device's operating parameters during the execution process, and the result feedback information after the execution is completed, calculate the instruction execution completion time, and enter the execution process field of the third instruction full process information association table, and output the fourth instruction full process information association table; Extract core information from the fourth instruction full-process information association table, organize the extracted core information according to the time sequence of instruction execution, and output the organized instruction full-process core information. Based on the timeline, the core information of the entire instruction process is linked together in the order of issuing, transmitting, receiving, and executing to form an instruction execution trajectory diagram. At the same time, a structured data file containing all core information is generated, and the entire instruction execution trajectory data is output.

7. The method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model according to claim 1, characterized in that, The process involves importing historical risk data from the automated operation and maintenance of the power grid dispatching system, screening historical risk cases that match the characteristics of the risk triggering source, extracting risk transmission process data from these historical risk cases, supplementing them into the initial risk transmission path, and outputting a refined risk transmission path, including: Collect historical risk data of power grid dispatch automation operation and maintenance. The historical risk data of power grid dispatch automation operation and maintenance covers various operation and maintenance risk cases that have occurred in the power grid dispatch system, including the time of occurrence, the scenario of occurrence, the triggering characteristics of the risk, the record of the risk transmission process, the scope of the risk impact, the intervention measures and the effects. Historical risk data of power grid dispatch automation operation and maintenance are classified and organized, and divided into multiple categories according to the type of risk triggering characteristics. Each category is further subdivided according to the type of equipment or system affected by the risk, forming a historical risk case classification library. Extract the core parameters from the risk trigger source characteristics and related information, use the extracted core parameters as search conditions to search in the historical risk case classification database, and output the retrieved historical risk cases. Calculate the similarity of the core parameters of the risk trigger features and the risk trigger source features in the retrieved historical risk cases, output the similarity calculation results, filter out historical risk cases with similarity higher than the feature matching threshold in the similarity calculation results, form a set of common source risk cases, sort the cases in the set of common source risk cases, arrange them from high to low similarity, and output the sorted set of common source risk cases. Analyze the top N cases in the sorted set of cases with the same source of risk, extract the risk transmission process record for each case, extract the intermediate features, feature change nodes, and transmission path turning information in the risk transmission process, and output the extracted risk transmission process related information. The extracted information related to the risk transmission process is deduplicated, and the adaptability of the core information of the deduplicated risk transmission process to the initial risk transmission path is analyzed. It is determined whether intermediate features can be integrated into the initial risk transmission path, whether the feature change nodes are consistent with the time logic of the initial risk transmission path, and whether the transmission path turning information conforms to the current risk evolution trend. Information that conforms to the time logic and evolution trend of the initial risk transmission path is retained, and the risk transmission process information that meets the requirements is output. Insert intermediate features from the qualified risk transmission process information into the corresponding time nodes of the initial risk transmission path, supplement the identifiers of feature change nodes and the range of parameter changes, update the turning information of the transmission path, improve the link settings of the transmission path, and output the improved risk transmission path.

8. The method for automated intelligent operation and maintenance of power grid dispatch based on a multimodal AI large model according to claim 3, characterized in that, The risk triggering stage feature information, based on the feature information of each operational risk evolution stage in the risk evolution chain, is analyzed to determine the formation cause and initial diffusion trend of the risk triggering source features, identify the core intervention direction for this risk triggering stage, select intervention methods, extract the operation object, operation steps, and implementation time window for each intervention method, and output the intervention elements of the risk triggering stage, including: By combining the associated data in the coupling feature set, we can trace the direct cause of the formation of risk trigger source features, analyze the specific manifestations and causes of parameter anomalies, the parameter indicators of equipment anomalies, the time point of anomaly occurrence and its associated impact with other equipment, the abnormal fields of interactive data, the fault nodes of the interactive link, the abnormal change magnitude and duration of key environmental factors, and output the analysis results of the causes of risk trigger source features. Based on the analysis of the causes of risk triggering source characteristics, the initial diffusion trend is predicted, including the main direction of diffusion, the possible associated characteristics, the changing pattern of diffusion rate, and the possible characteristic variations during the diffusion process. Combined with the real-time status of power grid dispatch operation, the potential impact of initial diffusion on core equipment, key systems and overall dispatch stability of the power grid is judged, and the initial diffusion trend prediction results are output. Based on the initial diffusion trend prediction results, the core direction of intervention for this risk evolution stage is determined. Based on the core direction of intervention, intervention measures for adjusting the execution parameters of the instruction are selected. The operation objects, operation steps, and implementation time windows of the intervention measures for adjusting the execution parameters of the instruction are extracted, and the elements of the intervention measures for adjusting the execution parameters of the instruction are output. Based on the core direction of the intervention, select intervention methods to suspend the operation of related equipment, extract the operation objects, operation steps, and implementation time windows of the intervention methods to suspend the operation of related equipment, and output the elements of the intervention methods to suspend the operation of related equipment. Based on the core direction of the intervention, select the intervention methods for shielding environmental interference sources, extract the operation objects, operation steps, and implementation time windows of the intervention methods for shielding environmental interference sources, and output the elements of the intervention methods for shielding environmental interference sources. Prioritize intervention measures such as adjusting command execution parameters, suspending the operation of associated equipment, and shielding environmental interference sources; set switching conditions between different intervention measures; and output the priority of intervention measures and switching conditions. Construct a coordinated intervention scheme for adjusting command execution parameters, suspending the operation of associated equipment, and shielding environmental interference sources. This scheme includes time connection rules, operation connection procedures, and data sharing specifications for the implementation of different methods, and outputs intervention elements for the risk triggering stage.

9. A power grid dispatching automation and intelligent operation and maintenance system based on a multimodal AI large model, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the power grid dispatch automation and intelligent operation and maintenance method based on a multimodal AI large model as described in any one of claims 1 to 8 by executing the machine-executable instructions.

Images