Intelligent arrangement method, system and equipment for operation and maintenance tasks of power transformation equipment and medium

Abstract

The invention relates to a power transformation equipment operation and maintenance task intelligent arrangement method and system, equipment and a medium, and relates to the technical field of electric power. The method comprises the steps of obtaining real-time operation data and historical operation and maintenance data of power transformation equipment, constructing input information of a large electric power model by adopting a Prompt technology enhanced by a knowledge graph, and predicting an operation and maintenance task by utilizing the large electric power model; and generating an operation and maintenance task arrangement scheme, performing dynamic optimization of the operation and maintenance task arrangement scheme based on reinforcement learning, generating and issuing an operation and maintenance work ticket, and feeding back task execution state and equipment state information to the large electric power model in real time to optimize the operation and maintenance task arrangement scheme. Compared with the prior art, the intelligent arrangement of the operation and maintenance tasks of the power transformation equipment is realized, the operation and maintenance efficiency and the equipment reliability are improved, the human resource consumption is reduced, and the scientificity of operation and maintenance decision is enhanced.

Description

Technical Field

[0001] This invention relates to power equipment technology, and in particular to a method, system, equipment, and medium for intelligent scheduling of operation and maintenance tasks for substation equipment. Background Technology

[0002] Substation equipment is a critical component of the power system, and its safe and stable operation is of great significance to ensuring the reliability of power supply. With the continuous expansion of the power grid and the improvement of its intelligence level, the types and quantities of substation equipment are showing a rapid growth trend, posing significant challenges to traditional substation equipment operation and maintenance management. Traditional substation equipment operation and maintenance mainly relies on fixed maintenance cycles and experience-based judgment, lacking scientific assessment and accurate prediction of the actual operating status of the equipment, resulting in unreasonable allocation of operation and maintenance resources and low maintenance efficiency.

[0003] The existing scheduling of substation equipment operation and maintenance tasks mainly relies on manual experience and fixed cycles, lacking the ability to accurately perceive and analyze the actual status of the equipment. It is impossible to dynamically adjust the operation and maintenance strategy according to the real-time operating status of the equipment, resulting in over-maintenance of some equipment and under-maintenance of others, causing waste of operation and maintenance resources and potential safety hazards.

[0004] With the development of artificial intelligence technology, especially the maturity of technologies such as large models, knowledge graphs, and reinforcement learning, new technical means have been provided for the intelligent operation and maintenance of substation equipment. These technologies can effectively solve the problems of traditional methods struggling to effectively integrate and utilize professional knowledge and experience in power systems, and lacking the ability to deeply mine and learn from historical operation and maintenance data. Chinese Patent CN119847802A discloses an intelligent operation and maintenance management and alarm system based on a large model intelligent agent, comprising: a data acquisition and fusion module for processing raw data collected from IT systems to obtain an operation and maintenance data graph, wherein the raw data includes at least: logs, indicators, and link tracing data; an anomaly detection module for analyzing pre-processed data and time-series features to identify abnormal data; an alarm optimization module for analyzing the semantic information corresponding to abnormal data to determine the priority of corresponding alarms; a root cause analysis module for performing fault reasoning on abnormal data based on a pre-built knowledge graph and generating fault explanations based on the data corresponding to the abnormal data; and an automated operation and maintenance decision-making module for generating automatic repair strategies based on the abnormal data and the fault explanations, and executing the automatic repair strategies to repair the abnormal data. However, this method emphasizes anomaly detection and root cause analysis and automatically generates repair strategies, but it does not involve effective solutions for operation and maintenance task orchestration and lacks a dynamic optimization process, making it difficult to adapt to the complex and ever-changing power grid operating environment and equipment status changes. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a method, system, equipment and medium for intelligent scheduling of substation operation and maintenance tasks. By integrating real-time operation data, historical operation and maintenance data and expert knowledge, it can achieve accurate prediction and optimized scheduling of substation operation and maintenance tasks, thereby improving operation and maintenance efficiency and equipment reliability.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] According to a first aspect of the present invention, an intelligent scheduling method for operation and maintenance tasks of substation equipment is provided. The method acquires real-time operation data and historical operation and maintenance data of substation equipment, constructs input information of a power big model using knowledge graph-enhanced Prompt technology, uses the power big model to predict operation and maintenance tasks, generates an operation and maintenance task scheduling scheme, dynamically optimizes the operation and maintenance task scheduling scheme based on reinforcement learning, generates and issues operation and maintenance work tickets, and simultaneously feeds back task execution status and equipment status information to the power big model in real time to optimize the operation and maintenance task scheduling scheme.

[0008] As a preferred technical solution, the method includes the following steps:

[0009] Acquire real-time operating data and historical maintenance data of the power equipment. The real-time operating data includes the operating parameters, environmental parameters and alarm information of the power equipment. The historical maintenance data includes the inspection records, fault records and maintenance records of the power equipment.

[0010] Based on the acquired real-time operation data and historical operation and maintenance data, the input information of the power large model is constructed using knowledge graph-enhanced Prompt technology. The knowledge graph includes the topological relationship of substation equipment, equipment attributes, operation and maintenance procedures and expert experience. By semantically associating the knowledge graph with real-time operation data and historical operation and maintenance data, a structured operation and maintenance task description is generated.

[0011] The structured operation and maintenance task description is input into the power big model, and the power big model outputs the operation and maintenance task prediction results of the substation equipment. The operation and maintenance task prediction results include the type, priority and expected workload of the operation and maintenance task. An operation and maintenance task scheduling scheme is generated based on the operation and maintenance task prediction results.

[0012] The operation and maintenance task orchestration scheme is dynamically optimized based on reinforcement learning methods. Specifically, it includes: constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives, using the actual execution feedback of operation and maintenance personnel as the environmental state, and obtaining an operation and maintenance task orchestration scheme that meets multiple constraints through iterative optimization. The multiple constraints include personnel qualification requirements, tool configuration requirements, safety interval requirements, and time window requirements.

[0013] Based on the optimized operation and maintenance task orchestration scheme, an operation and maintenance work ticket is generated, which includes work items, work content, work procedures, security measures, and precautions.

[0014] Operation and maintenance work orders are issued to operation and maintenance personnel, and task execution status and equipment status information are collected in real time during the execution of operation and maintenance tasks. The task execution status and equipment status information are fed back to the power big data model to continuously optimize the operation and maintenance task scheduling scheme.

[0015] As a preferred technical solution, the process of generating a structured operation and maintenance task description includes:

[0016] Construct a knowledge graph, which includes entity nodes of substation equipment and relational edges. The entity nodes include topology relation nodes, equipment attribute nodes, operation and maintenance procedure nodes, and expert experience nodes of substation equipment. The relational edges are used to represent the association relationships between entity nodes.

[0017] The entity nodes in the knowledge graph are vectorized based on a deep learning model. Key features are extracted from the real-time operation data and historical operation and maintenance data through an attention mechanism. The extracted key features are semantically aligned with the vector representations of the entity nodes.

[0018] A knowledge-enhanced Prompt template is constructed using a multi-head self-attention mechanism. The Prompt template includes a task type description, a device status description, a historical operation and maintenance description, and an expert knowledge description. Semantically aligned feature information is then filled into the corresponding positions in the Prompt template.

[0019] A structured operation and maintenance task description is generated based on the populated Prompt template. The structured operation and maintenance task description includes device identification information, task triggering conditions, task execution requirements, and task completion standards.

[0020] As a preferred technical solution, the step of inputting the structured operation and maintenance task description into the power system large model and outputting the operation and maintenance task prediction results of the substation equipment from the power system large model includes:

[0021] A large power model is constructed using a multimodal fusion pre-training method. The multimodal fusion pre-training method includes: inputting text data, time series data, and image data of power equipment into the corresponding encoders for feature extraction; aligning and fusing features of different modalities through a cross-modal attention mechanism; and training the large power model based on mask autoencoder and contrastive learning strategies.

[0022] The structured operation and maintenance task description is input into the power big model to obtain the potential failure modes and health status of the substation equipment, and the operation and maintenance task prediction results of the substation equipment are output.

[0023] As a preferred technical solution, the aforementioned large-scale power model analyzes the potential fault modes and equipment health status based on a hierarchical task classification framework, and outputs the operation and maintenance task prediction results for substation equipment, wherein:

[0024] At the first level, the type of operation and maintenance task is determined through the equipment status assessment module. The types of operation and maintenance tasks include preventive maintenance tasks, fault repair tasks, and condition-based maintenance tasks.

[0025] At the second level, the task urgency score is calculated through the task priority evaluation module, and the priority of the operation and maintenance task is determined based on the task urgency score.

[0026] At the third level, the workload estimation module predicts the number of personnel, equipment configuration, and working time required for maintenance tasks based on historical execution data of similar tasks.

[0027] The calculation results from the first level, the second level, and the third level are combined to generate the operation and maintenance task prediction results.

[0028] As a preferred technical solution, the reward function specifically includes:

[0029] Operational efficiency reward Re(s,a):

[0030] Re(s,a)=w1·(1 / T c )+w2·(N u / N t )+w3·(1 / T d );

[0031] Among them, T c N represents the total time to complete the task. u N represents the actual working hours of the personnel. t T represents the total available time. d This represents the equipment downtime, where w1, w2, and w3 are weighting coefficients.

[0032] Equipment reliability bonus Rr(s,a):

[0033] Rr(s,a)=α·H(s)+β·(1-F(s))+γ·Q(s);

[0034] Where H(s) represents the equipment health function, F(s) represents the failure risk function, Q(s) represents the maintenance quality scoring function, and α, β, and γ are weighting coefficients;

[0035] Constraint violation penalty Rp(s,a):

[0036] Rp(s,a)=-λ1·max(0,1-M p)-λ2·max(0,1-M r )-λ3·max(0,1-M s )-λ4·max(0,1-M t );

[0037] Among them, M p M represents the degree of matching between personnel qualifications. r Indicating the rationality of tool and equipment configuration, M s M represents the degree of safety interval satisfaction. t The time window compliance is represented by λ1, λ2, λ3, and λ4, which are penalty coefficients.

[0038] As a preferred technical solution, the reinforcement learning method is based on the dual Q-learning algorithm for policy optimization:

[0039] Construct the state-action value function Q(s,a) and the target value function Q'(s,a):

[0040] Q(s,a)←Q(s,a)+η[R(s,a)+γ·maxQ'(s',a')-Q(s,a)];

[0041] Where η represents the learning rate, γ represents the discount factor, and R(s,a) represents the reward function;

[0042] An ε-greedy strategy is used for action selection:

[0043] π(a|s)={1-ε+ε / |A|,if a=argmaxQ(s,a);ε / |A|,otherwise};

[0044] Where π(a|s) represents the probability of choosing action a in state s, ε represents the exploration probability, A represents the action space, and |A| represents the size of the action space;

[0045] The optimal operation and maintenance task orchestration scheme will be output when the following convergence conditions are met:

[0046] |Q(k+1)(s,a)-Q(k)(s,a)|<δ;

[0047] Where k represents the number of iterations and δ represents the convergence threshold.

[0048] According to a second aspect of the present invention, a substation equipment operation and maintenance task intelligent scheduling system is provided for implementing the method, the system comprising:

[0049] Data acquisition module: Acquires real-time operating data and historical maintenance data of the power equipment. The real-time operating data includes the operating parameters, environmental parameters and alarm information of the power equipment. The historical maintenance data includes the inspection records, fault records and maintenance records of the power equipment.

[0050] Input information construction module: Based on the acquired real-time operation data and historical operation and maintenance data, the input information of the power large model is constructed using knowledge graph-enhanced Prompt technology. The knowledge graph includes the topological relationship of substation equipment, equipment attributes, operation and maintenance procedures and expert experience. By semantically associating the knowledge graph with real-time operation data and historical operation and maintenance data, a structured operation and maintenance task description is generated.

[0051] Operation and maintenance task orchestration scheme generation module: The structured operation and maintenance task description is input into the power big model, and the power big model outputs the operation and maintenance task prediction results of the substation equipment. The operation and maintenance task prediction results include the type, priority and expected workload of the operation and maintenance task. An operation and maintenance task orchestration scheme is generated based on the operation and maintenance task prediction results.

[0052] The operation and maintenance task orchestration scheme optimization module: dynamically optimizes the prediction results of the operation and maintenance tasks based on reinforcement learning methods. Specifically, it includes: constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives, using the actual execution feedback of operation and maintenance personnel as the environmental state, and obtaining an operation and maintenance task orchestration scheme that meets multiple constraints through iterative optimization. The multiple constraints include personnel qualification requirements, tool configuration requirements, safety interval requirements, and time window requirements.

[0053] Operation and maintenance work ticket generation module: Based on the optimized operation and maintenance task orchestration scheme, it generates operation and maintenance work tickets, which include work items, work content, work procedures, security measures and precautions;

[0054] Feedback optimization module: Issues maintenance work orders to maintenance personnel and collects task execution status and equipment status information in real time during the execution of maintenance tasks. It then feeds the task execution status and equipment status information back to the power big data model to continuously optimize the maintenance task scheduling scheme.

[0055] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0056] processor;

[0057] Memory used to store processor-executable instructions;

[0058] The processor is configured to invoke instructions stored in the memory to execute the method.

[0059] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having computer program instructions stored thereon, which, when executed by a processor, implement the method described thereon.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] (1) The intelligent scheduling method for operation and maintenance tasks of substation equipment provided by the present invention constructs input information through knowledge graph-enhanced Prompt technology, effectively integrates the topological relationship, attributes, operation and maintenance procedures and expert experience of substation equipment, significantly improves the accuracy and completeness of operation and maintenance task description, and solves the problem of task planning deviation caused by incomplete information in the traditional method of relying on experience judgment.

[0062] (2) This invention uses reinforcement learning to dynamically optimize the prediction results of operation and maintenance tasks, and constructs a reward function with operation and maintenance efficiency and equipment reliability as the objectives. It fully considers multi-dimensional constraints such as personnel qualifications, tool configuration, safety interval and time window, and realizes the optimal allocation of operation and maintenance resources, improves resource utilization and reduces operation and maintenance costs.

[0063] (3) This invention realizes closed-loop management of the entire operation and maintenance task process. By collecting task execution status and equipment status information in real time and feeding it back to the power big model, it supports the continuous optimization and iterative update of operation and maintenance schemes, effectively improves the intelligence level of substation equipment operation and maintenance, reduces equipment failure rate by about 40%, extends equipment service life by about 15%, and provides a strong guarantee for the safe and stable operation of the power grid. Attached Figure Description

[0064] Figure 1 This is a flowchart of the method of the present invention;

[0065] Figure 2 This is a complete flowchart of constructing a large power model based on Prompt technology in an embodiment of the present invention;

[0066] Figure 3 This is a schematic diagram comparing the performance of different types of operation and maintenance tasks under the hierarchical task classification framework in this embodiment of the invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0068] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0069] Example 1

[0070] This embodiment provides an intelligent scheduling method for substation equipment operation and maintenance tasks. The method acquires real-time operation data and historical operation and maintenance data of substation equipment, uses knowledge graph-enhanced Prompt technology to construct input information for a large power model, uses the large power model to predict operation and maintenance tasks, generates an operation and maintenance task scheduling scheme, and dynamically optimizes the operation and maintenance task scheduling scheme based on reinforcement learning. It generates and issues operation and maintenance work tickets, and simultaneously feeds back the task execution status and equipment status information to the large power model in real time to optimize the operation and maintenance task scheduling scheme.

[0071] Specifically, such as Figure 1 As shown, the method includes the following steps:

[0072] S1, acquire real-time operating data and historical maintenance data of the power equipment. The real-time operating data includes the operating parameters, environmental parameters and alarm information of the power equipment. The historical maintenance data includes the inspection records, fault records and maintenance records of the power equipment.

[0073] S2. Based on the acquired real-time operation data and historical operation and maintenance data, the input information of the power large model is constructed using knowledge graph-enhanced Prompt technology. The knowledge graph includes the topological relationships of substation equipment, equipment attributes, operation and maintenance procedures, and expert experience. By semantically associating the knowledge graph with the real-time operation data and historical operation and maintenance data, a structured operation and maintenance task description is generated.

[0074] In one alternative implementation, such as Figure 2 As shown, the process of generating a structured operation and maintenance task description includes:

[0075] Construct a knowledge graph, which includes entity nodes of substation equipment and relational edges. The entity nodes include topology relation nodes, equipment attribute nodes, operation and maintenance procedure nodes, and expert experience nodes of substation equipment. The relational edges are used to represent the association relationships between entity nodes.

[0076] The entity nodes in the knowledge graph are vectorized based on a deep learning model. Key features are extracted from the real-time operation data and historical operation and maintenance data through an attention mechanism. The extracted key features are semantically aligned with the vector representations of the entity nodes.

[0077] A knowledge-enhanced Prompt template is constructed using a multi-head self-attention mechanism. The Prompt template includes a task type description, a device status description, a historical operation and maintenance description, and an expert knowledge description. Semantically aligned feature information is then filled into the corresponding positions in the Prompt template.

[0078] A structured operation and maintenance task description is generated based on the populated Prompt template. The structured operation and maintenance task description includes device identification information, task triggering conditions, task execution requirements, and task completion standards.

[0079] S3. Input the structured operation and maintenance task description into the power big model, and output the operation and maintenance task prediction results of the substation equipment from the power big model. The operation and maintenance task prediction results include the type, priority and expected workload of the operation and maintenance tasks. Generate an operation and maintenance task scheduling scheme based on the operation and maintenance task prediction results.

[0080] In one optional implementation, the step of inputting the structured operation and maintenance task description into the power system large model, and outputting the operation and maintenance task prediction results of the substation equipment from the power system large model, includes:

[0081] A large power model is constructed using a multimodal fusion pre-training method. The multimodal fusion pre-training method includes: inputting text data, time series data, and image data of power equipment into the corresponding encoders for feature extraction; aligning and fusing features of different modalities through a cross-modal attention mechanism; and training the large power model based on mask autoencoder and contrastive learning strategies.

[0082] The structured operation and maintenance task description is input into the power big model to obtain the potential failure modes and health status of the substation equipment, and the operation and maintenance task prediction results of the substation equipment are output.

[0083] For example, the large power model of this application is based on the Transformer encoder-decoder structure and combines the knowledge graph-enhanced Prompt technique, adopts a multi-head self-attention mechanism to process multi-source heterogeneous data, introduces gated recurrent units (GRU) to enhance the temporal feature extraction capability, and uses residual connections and layer normalization to optimize the model training stability.

[0084] include:

[0085] Multimodal data fusion layer: integrates heterogeneous data such as text, numerical data, and images;

[0086] Knowledge graph embedding layer: Incorporating domain-specific knowledge constraints;

[0087] Temporal feature extraction layer: captures the temporal variation patterns of device parameters;

[0088] Feature interaction layer: Models the relationships between different features.

[0089] In one optional implementation, the power big data model analyzes the potential fault modes and equipment health status based on a hierarchical task classification framework, and outputs the operation and maintenance task prediction results of the substation equipment, wherein:

[0090] At the first level, the type of operation and maintenance task is determined through the equipment status assessment module. The types of operation and maintenance tasks include preventive maintenance tasks, fault repair tasks, and condition-based maintenance tasks.

[0091] At the second level, the task urgency score is calculated through the task priority evaluation module, and the priority of the operation and maintenance task is determined based on the task urgency score.

[0092] At the third level, the workload estimation module predicts the number of personnel, equipment configuration, and working time required for maintenance tasks based on historical execution data of similar tasks.

[0093] The calculation results from the first level, the second level, and the third level are combined to generate the operation and maintenance task prediction results.

[0094] S4. Dynamically optimize the operation and maintenance task orchestration scheme based on reinforcement learning method, specifically including: constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives, using the actual execution feedback of operation and maintenance personnel as the environment state, and obtaining an operation and maintenance task orchestration scheme that meets multiple constraints through iterative optimization. The multiple constraints include personnel qualification requirements, tool configuration requirements, safety interval requirements, and time window requirements.

[0095] In one alternative implementation, the reinforcement learning method optimizes the policy based on the dual Q-learning algorithm:

[0096] Construct the state-action value function Q(s,a) and the target value function Q'(s,a):

[0097] Q(s,a)←Q(s,a)+η[R(s,a)+γ·maxQ'(s',a')-Q(s,a)];

[0098] Where η represents the learning rate, γ represents the discount factor, and R(s,a) represents the reward function;

[0099] An ε-greedy strategy is used for action selection:

[0100] π(a|s)={1-ε+ε / |A|,if a=argmaxQ(s,a);ε / |A|,otherwise};

[0101] Where π(a|s) represents the probability of choosing action a in state s, ε represents the exploration probability, A represents the action space, and |A| represents the size of the action space;

[0102] The optimal operation and maintenance task orchestration scheme will be output when the following convergence conditions are met:

[0103] |Q(k+1)(s,a)-Q(k)(s,a)|<δ;

[0104] Where k represents the number of iterations and δ represents the convergence threshold.

[0105] S5. Based on the optimized operation and maintenance task orchestration scheme, generate an operation and maintenance work ticket. The operation and maintenance work ticket includes work items, work content, work procedures, security measures, and precautions.

[0106] S6. Issue maintenance work orders to maintenance personnel and collect task execution status and equipment status information in real time during the execution of maintenance tasks. Feed the task execution status and equipment status information back to the power big data model to continuously optimize the maintenance task scheduling scheme.

[0107] Example 2

[0108] This embodiment provides a further detailed description of step S2 based on embodiment 1.

[0109] This method first constructs a knowledge graph of substation equipment, including various types of substation equipment entity nodes and the edges connecting them. Entity nodes mainly cover four types of information: substation equipment topology relationships, equipment attributes, operation and maintenance procedures, and expert experience. For example, for a specific model of circuit breaker CB-1001, its topology relationship node records the connection relationship between the circuit breaker and bus A and protection device PR-501; the equipment attribute node contains information such as its rated voltage of 500kV, rated current of 2000A, and manufacturing year of 2015; the operation and maintenance procedure node records the circuit breaker's periodic maintenance cycle of 6 months and the method for checking the oil level of the operating mechanism; the expert experience node contains experiential knowledge such as "the oil level of the operating mechanism needs to be replenished in time when it is below 60%". Relationship edges, such as "connected to", "has attributes", "follows procedures", and "refers to experience", are used to represent the semantic associations between different nodes.

[0110] After the knowledge graph is constructed, a deep learning model is used to vectorize the entity nodes in the graph. Specifically, a graph neural network model is employed, which captures the structural relationships between nodes and generates 128-dimensional entity node vector representations. For example, the vector representation of circuit breaker CB-1001 incorporates multi-dimensional information such as its equipment type, connection relationships, and attribute characteristics. Simultaneously, key features are extracted from real-time operational data and historical maintenance data. For real-time operational data, such as the fact that circuit breaker CB-1001 has been operated 1500 times and the last three operation times were 78ms, 82ms, and 87ms respectively, an attention mechanism is used to identify the key feature of an upward trend in operation time. For historical maintenance data, such as the fact that this type of circuit breaker experienced a failure to operate when the operation time reached 90ms, this type of related information is given special attention. The extracted features are also converted into vector representations and semantically aligned with the entity node vectors in the knowledge graph. A similarity matrix is ​​calculated to identify the most relevant entity nodes between the data and the knowledge base.

[0111] Based on semantic alignment results, a knowledge-enhanced Prompt template is constructed. This template comprises four main parts: task type description, equipment status description, historical maintenance description, and expert knowledge description. A multi-head self-attention mechanism is employed to handle the relationships between different parts, assigning different attention weights to different input features. For example, for the maintenance task of circuit breaker CB-1001, the generated Prompt template includes: the task type description section filled with "Circuit breaker operating mechanism overhaul"; the equipment status description section filled with "Circuit breaker CB-1001 operating time has shown an upward trend for three consecutive times, currently at 87ms, approaching the warning value of 90ms"; the historical maintenance description section filled with "This type of circuit breaker experienced a failure to operate when the operating time reached 90ms, and the last maintenance date was October 15, 2023"; and the expert knowledge description section filled with "When the circuit breaker operating time continues to increase, the operating mechanism oil level and mechanical transmission components should be checked."

[0112] Finally, a structured maintenance task description is generated based on the filled-in Prompt template. This description includes four core elements: equipment identification information, task triggering conditions, task execution requirements, and task completion criteria. For the above case, the generated structured maintenance task description is: Equipment identification information: "Circuit breaker CB-1001, located in bay 3 of site S001"; Task triggering conditions: "Operation time increases three times consecutively, and the current value of 87ms is close to the warning value of 90ms"; Task execution requirements: "Inspect the circuit breaker operating mechanism, focusing on whether the oil level is below the 60% mark, whether there is any jamming in the transmission components, and whether the spring energy storage mechanism is normal"; Task completion criteria: "Operation time recovers to below 80ms, all components of the operating mechanism are in normal condition, record the maintenance process, and update the equipment maintenance file." Such a structured description can be directly input into the power system model to generate more accurate maintenance suggestions and solutions.

[0113] This method demonstrates high efficiency and accuracy in practical applications. For example, during equipment maintenance at a 500kV substation, this implementation successfully identified a potential fault risk in an aging circuit breaker. By analyzing the circuit breaker's operating time trends, historical maintenance records, and similar equipment failure cases, a targeted maintenance task description was generated. Maintenance personnel followed the description to identify and resolve the issue of insufficient oil pressure in the operating mechanism, preventing a potential circuit breaker failure to operate. Statistical data shows that after adopting this method, the unplanned outage rate of substation equipment decreased by 23%, equipment maintenance efficiency increased by 31%, and the reliability and safety of power grid operation were significantly improved.

[0114] Example 3

[0115] This embodiment provides a further detailed description of step S3 based on embodiment 1.

[0116] The construction of the large-scale power model employs a multimodal fusion pre-training method, which integrates textual, temporal, and image data of substation equipment for feature extraction and fusion. For textual data, an encoder based on a Transformer architecture is used to process textual information such as equipment nameplate information, historical fault records, and maintenance records. The encoder contains 12 layers of self-attention modules, with a hidden layer dimension of 768 and 12 attention heads, effectively capturing semantic features of the text. For example, for a textual description such as "110kV main transformer B-phase bushing rupture and oil leakage," the encoder can extract key features such as equipment type, faulty component, and fault phenomenon.

[0117] The time-series data processing employs a temporal convolutional network structure with causal convolution. This network contains 8 convolutional layers with a kernel size of 3 and dilation coefficients increasing progressively from 1 to 128, enabling the extraction of features at different time scales. This structure can process time-series monitoring data such as temperature, vibration, and partial discharge of power equipment. For example, for the oil temperature data of the main transformer, the model can identify an abnormal pattern where the temperature gradually rises from 35°C to 75°C and remains at a high temperature.

[0118] Image data is processed by an improved visual encoder, optimized based on the ResNet50 architecture, containing five residual blocks and increasing the feature dimension from 64 to 2048. An attention enhancement module is added to target local details in the equipment images, enabling the model to focus on visual features such as insulator flashover and metal component corrosion. For example, in circuit breaker contact images, the model can identify carbonization marks and ablation conditions.

[0119] Multimodal feature fusion employs a cross-modal attention mechanism to achieve alignment and interaction between different modalities. For each modal feature vector, its attention weights with other modal features are calculated to generate a cross-modal context vector, which is then concatenated with the original features and passed through a feedforward network to obtain a fused representation. In the specific implementation, attention is calculated using a dot product approach, and the weights are normalized using a softmax function. This mechanism enables the model to associate "casing breakage" in the text with the corresponding abnormal temperature curve and crack features in the image.

[0120] The pre-training phase employs a masking autoencoder and contrastive learning strategy. Masking autoencoder randomly masks 15% of the input data, requiring the model to reconstruct the masked portion. For example, masking temperature values ​​for a specific time period in time-series data or local regions in an image requires the model to predict this missing information. Contrastive learning enhances the model's ability to recognize similar samples by constructing positive and negative sample pairs. Positive sample pairs include data from different modalities of the same device, while negative sample pairs consist of data from different devices. The model pre-training uses a device dataset containing 500,000 text records, 300,000 time-series data sets, and 200,000 device images. It is trained for 300 epochs using the Adam optimizer with a learning rate of 0.0001 and a batch size of 32.

[0121] In the operation and maintenance task prediction phase, a structured description of the operation and maintenance task is input into the power system's large-scale model. This structured description includes fields such as equipment identifier, equipment type, operating parameters, and abnormal phenomena. For example: "Equipment ID: TR-1103, Type: 220kV oil-immersed transformer, Operating duration: 15 years, Abnormal phenomena: Phase B temperature is 12℃ higher than other phases, there is a slight abnormal noise, and the oil color has darkened." After processing this input, the model outputs the operation and maintenance task prediction results, including task type (e.g., inspection, replacement, maintenance), priority (urgent, high, medium, low), and estimated workload (man-hours, personnel allocation).

[0122] To identify potential failure modes of power equipment, the model analyzes the similarity between input information and historical cases, and combines this with an equipment health assessment algorithm to generate a failure probability distribution. Equipment health status assessment is based on a comprehensive analysis of multi-dimensional features, including the equipment's life cycle location, the degree of deviation of key parameters, and historical failure modes. For example, for the transformer described in the input, the model might predict its failure mode as "localized overheating due to winding insulation aging," with a health status score of 65 (out of 100). It would recommend performing "partial discharge testing" and "oil chromatography analysis" with a "high" priority, and the estimated workload would be "2 people / 6 hours."

[0123] Figure 3 The performance comparison results for different operation and maintenance tasks are presented. It is evident that the method provided in this embodiment achieves good results in task completion rate, resource utilization efficiency, response timeliness, and fault prediction accuracy for various tasks. The model evaluation, based on 10,000 real operation and maintenance records, achieves an accuracy rate of 87.5%, which is 15 percentage points better than traditional rule-based expert systems. Regarding fault prediction lead time, potential faults can be detected an average of 7.2 days in advance, a 40% improvement compared to traditional expert experience-based judgment methods, effectively reducing unplanned equipment downtime.

[0124] This specific embodiment provides a detailed technical description of an operation and maintenance task prediction method based on a hierarchical task classification framework. This method mainly revolves around the analysis of potential failure modes and equipment health status, determining the type, priority, and required workload of operation and maintenance tasks through three levels of evaluation modules.

[0125] First, the system receives equipment monitoring data, including but not limited to real-time parameters such as vibration, temperature, pressure, and current data collected by sensors, as well as historical operation and maintenance records. In a real-world application scenario, the main transformer of a substation might collect hourly data such as oil temperature, load rate, and gas content.

[0126] At the first level, the equipment condition assessment module identifies potential fault modes and assesses equipment health by analyzing these monitoring data. This module employs feature extraction technology to extract key features from the raw data, such as the peak value, mean, and standard deviation of transformer oil temperature data. Subsequently, the module uses a pre-trained equipment condition recognition model to map the extracted features to predefined equipment condition categories. For example, if the transformer oil temperature exceeds 85°C for three consecutive days and gas analysis shows an acetylene content of 150 ppm, a potential winding overheating fault mode in the transformer may be identified.

[0127] Based on the identified fault modes and health status, the equipment condition assessment module determines the type of maintenance task. If the equipment is in normal operation but nearing its preset maintenance cycle, such as a circuit breaker that has been in operation for 11 months and is approaching its annual maintenance deadline, a preventative maintenance task is generated. If a fault is detected, such as a circuit breaker that refuses to operate and cannot be remotely reset, a fault repair task is generated. If the equipment is in an abnormal state but has not yet affected normal operation, such as a transformer whose oil temperature fluctuates abnormally but is still within a safe range, a condition-based maintenance task is generated.

[0128] At the second level, the task priority assessment module calculates the task urgency score. This module comprehensively considers multiple factors: equipment importance weight, fault impact range, fault development speed, safety risk level, and operating environment conditions. For example, for transformer equipment in the main power supply system, its importance weight might be set to 0.9 (out of 1), while backup equipment might be set to 0.5; the fault impact range might involve 10 important downstream users, scoring 0.8; the fault development speed, predicted based on historical data, is medium-speed, scoring 0.6; the safety risk level is high, scoring 0.9; and the operating environment conditions are normal, scoring 0.3. These sub-scores are then weighted and averaged to obtain the final urgency score of 0.75.

[0129] Based on the calculated urgency score, the task priority assessment module categorizes maintenance tasks into different priorities. For example, tasks with an urgency score greater than 0.8 are marked as "urgent" and need to be processed within 24 hours; tasks with scores between 0.5 and 0.8 are marked as "important" and need to be processed within 72 hours; tasks with scores less than 0.5 are marked as "routine" and can be scheduled for processing within one week. In the example, the transformer maintenance task with a score of 0.75 is marked as "important" priority.

[0130] At the third level, the workload estimation module predicts the resources required for the current maintenance task based on historical execution data of similar tasks. This module first retrieves historical records similar to the current task from the task database and extracts the workload information for these tasks. For example, for an identified transformer winding overheating maintenance task, it might find 15 similar maintenance records completed in the past two years.

[0131] By analyzing these historical records, the workload estimation module calculates the average execution time for this type of task to be 6.5 hours, with a standard deviation of 1.2 hours. The required personnel are typically three, including one senior technician and two general technicians. Required equipment includes specialized devices such as insulation resistance testers, transformer ratio testers, and infrared thermometers. The module also considers the specific characteristics of the current task, such as equipment capacity, installation location, and fault severity, adjusting the basic estimate accordingly. If the current transformer capacity is 20% greater than the historical average, the estimated working time may need to be increased by 15%, reaching approximately 7.5 hours.

[0132] After completing the analysis at the three levels mentioned above, a complete operation and maintenance task prediction result is generated, including task type (condition-based maintenance task), priority (important, to be processed within 72 hours) and estimated workload (3 technicians, 7.5 hours of work time, with a specific list of tools).

[0133] Finally, these operation and maintenance task prediction results are output to the operation and maintenance task orchestration module. This module can perform subsequent task scheduling and resource allocation based on the prediction results. For example, it can arrange the execution order according to task priority, reasonably allocate personnel and tool resources according to the estimated workload, and generate detailed execution plans and work instructions to ensure that operation and maintenance work is carried out efficiently and orderly.

[0134] Example 4

[0135] This embodiment provides a further detailed description of step S4 based on embodiment 1.

[0136] This embodiment provides a method for constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives. The method uses the actual execution feedback of operation and maintenance personnel as the environmental state, and obtains an operation and maintenance task orchestration scheme that satisfies multiple constraints through iterative optimization.

[0137] In the operation and maintenance of power equipment, a multi-objective reward function R(s,a) is first constructed. This function consists of three parts: operation and maintenance efficiency reward Re(s,a), equipment reliability reward Rr(s,a), and constraint violation penalty Rp(s,a).

[0138] Operation and maintenance efficiency bonus Re(s,a)=w1·(1 / T) c )+w2·(N u / N t )+w3·(1 / T d );

[0139] Among them, T c N represents the total time to complete the task. u N represents the actual working hours of the personnel. t T represents the total available time. d This represents the equipment downtime, where w1, w2, and w3 are weighting coefficients.

[0140] Equipment reliability bonus Rr(s,a)=α·H(s)+β·(1-F(s))+γ·Q(s);

[0141] Where H(s) represents the equipment health function, F(s) represents the failure risk function, Q(s) represents the maintenance quality scoring function, and α, β, and γ are weighting coefficients;

[0142] Restriction and Penalty

[0143] Rp(s,a)=-λ1·max(0,1-M p )-λ2·max(0,1-M r )-λ3·max(0,1-M s )-λ4·max(0,1-M t );

[0144] Among them, M p M represents the degree of matching between personnel qualifications. r Indicating the rationality of tool and equipment configuration, M s M represents the degree of safety interval satisfaction. t The time window compliance is represented by λ1, λ2, λ3, and λ4, which are penalty coefficients.

[0145] The system obtains the overall reward function by weighting and combining these three parts: R(s,a)=ω e ·Re(s,a)+ωr ·Rr(s,a)+Rp(s,a), where ω e and ω r These are the weighting coefficients for operational efficiency and equipment reliability, respectively, and can be adjusted according to business needs. For example, ω can be set. e =0.4, ω r =0.6, which makes the system focus more on equipment reliability.

[0146] The operation and maintenance efficiency reward Re(s,a) considers three key indicators: total task completion time, personnel work efficiency, and equipment downtime. The system calculates the operation and maintenance efficiency reward using a formula, where weighting coefficients w1, w2, and w3 are allocated according to business importance. For example, for highly important power generation equipment, w1 = 0.3, w2 = 0.2, and w3 = 0.5 can be set to focus more on reducing equipment downtime. In practical applications, if the total completion time for a transformer maintenance task is 4 hours, the actual personnel working time is 3 hours (total available time is 8 hours), and the equipment downtime is 2 hours, then the operation and maintenance efficiency reward is calculated as: Re(s,a) = 0.3 × (1 / 4) + 0.2 × (3 / 8) + 0.5 × (1 / 2) = 0.075 + 0.075 + 0.25 = 0.4.

[0147] The equipment reliability bonus Rr(s,a) focuses on three aspects: equipment health, failure risk, and maintenance quality. The equipment health function H(s) is calculated by collecting equipment operating parameters, ranging from 0 to 1; a higher value indicates better equipment health. For example, for a power transformer, H(s) can be calculated based on a weighted average of 10 indicators, including insulating oil temperature, load rate, and vibration value; the H(s) value for a normally operating transformer is approximately 0.85 to 0.95. The failure risk function F(s) is based on historical failure data and current operating status assessment, ranging from 0 to 1; a lower value indicates lower failure risk. For example, for a circuit breaker that has been in use for 5 years, based on its number of operations and the time of its three most recent operations, F(s) might be 0.25. The maintenance quality scoring function Q(s) is determined by the quality of performance by maintenance personnel and expert evaluation results, ranging from 0 to 1; a higher value indicates higher maintenance quality. The system can set weighting coefficients α = 0.35, β = 0.35, and γ = 0.3 to balance the three factors. Taking substation circuit breaker maintenance as an example, if H(s) = 0.88, F(s) = 0.25, and Q(s) = 0.92, then the equipment reliability bonus is calculated as follows: Rr(s,a) = 0.35 × 0.88 + 0.35 × (1 - 0.25) + 0.3 × 0.92 = 0.308 + 0.2625 + 0.276 = 0.8465.

[0148] The penalty for constraint violation, Rp(s,a), considers four types of constraints: personnel qualification matching degree, rationality of tool and equipment configuration, safety interval satisfaction degree, and time window compliance degree. Personnel qualification matching degree M p Assess whether the skill level of the maintenance personnel meets the task requirements. The value ranges from 0 to 1, with 1 indicating full compliance. For example, high-voltage equipment maintenance requires personnel with a high-voltage electrician's license and at least 3 years of experience. If the assigned personnel only have 2 years of experience, M... p The value may be 0.8. The rationality of tool and equipment configuration (M) r The assessment evaluates whether the necessary tools for the task are complete and meet the standards, with a value ranging from 0 to 1, where 1 indicates full compliance. For example, if a line inspection requires five tools such as an infrared thermometer and an insulation tester, but only four are actually provided, and one of them is an outdated model, M... r It may be 0.75. Safety interval satisfaction M s Assess whether the safety distance between different tasks is sufficient, with a value ranging from 0 to 1, where 1 indicates that the requirement is fully met. For example, live-line work on two adjacent devices should maintain a distance of at least 5 meters. If the actual distance is 4 meters, M s It could be 0.8. Time window compliance M t The assessment determines whether the task execution time is within the allowed time window, with a value ranging from 0 to 1, where 1 indicates full compliance. For example, if transformer maintenance is mandated to be carried out during off-peak hours, but some work actually extends into peak load periods, M... t It could be 0.9. The system can be configured with penalty coefficients λ1 = 10, λ2 = 8, λ3 = 15, and λ4 = 12 to reflect the importance of each constraint. Taking substation equipment maintenance as an example, if M... p =0.8, M r =0.9, M s =1.0, M t =0.95, then the constraint violation penalty is calculated as follows:

[0149] Rp(s,a)=-10×max(0,1-0.8)-8×max(0,1-0.9)-15×max(0,1-1.0)-12×max(0,1-0.95)=-10×0.2-8×0.1-15×0-12×0.05=-2-0.8-0-0.6=-3.4.

[0150] During actual operation, the system continuously updates the environmental state 's' by collecting real-time feedback from maintenance personnel and equipment status information, and selects action 'a' based on the current strategy. After each action is completed, the system calculates the reward R(s,a) and updates the strategy network parameters using a reinforcement learning algorithm. After multiple rounds of iterative optimization, the system finally generates a task orchestration scheme that satisfies multiple constraints and optimizes maintenance efficiency and equipment reliability. In the maintenance practice of a power company's substation, after applying this method, equipment downtime was reduced by 23%, maintenance personnel efficiency increased by 18%, and equipment failure rate decreased by 15%, fully validating the effectiveness of the method.

[0151] In a service orchestration method based on dual Q-learning, a state-action space is first constructed, where state s represents the current service deployment status of the system, and action a represents a possible orchestration decision. The state can contain multi-dimensional features, such as current resource utilization, service response time, and system load; actions can be specific operations such as service migration, scaling up / down, and task scheduling. This method optimizes strategies using a dual Q-learning algorithm to cope with complex and ever-changing service orchestration scenarios.

[0152] This method uses two value functions for learning: a state-action value function Q and a target value function Q'. The value function Q(s,a) represents the expected long-term cumulative reward for performing action a in state s, while Q' serves as the target network for stabilizing the learning process. In practice, these functions can be represented using neural networks or lookup tables. During value updates, adjustments are made based on the currently observed reward R(s,a) and the maximum Q' value for the next state s'. For example, when the response time of service A is detected to exceed 500 milliseconds (state s), a scaling decision (action a) might be made, yielding a reward R of 15 units for reducing the response time, after which the system transitions to the new state s'. At this point, the learning rate η is set to 0.05, the discount factor γ is set to 0.9, and the Q-value update amount is 0.05 × (15 + 0.9 × maximum Q' value - current Q value).

[0153] In the action selection strategy, an ε-greedy strategy is adopted to balance exploration and utilization. In most cases (with a probability of 1-ε), the system selects the action with the highest current Q value; however, it also reserves an ε-probability to randomly select actions to explore potential better solutions. In practice, ε can be set to an initial value of 0.3 and gradually reduced to 0.05 as the learning process progresses, ensuring that the system fully explores in the early stages and makes greater use of the learned knowledge in the later stages. For example, in a specific state with 4 selectable actions and Q values ​​of [10, 15, 8, 12], when ε = 0.2, there is a probability of 0.8 + 0.2 / 4 = 0.85 to select the second action with the highest Q value (value 15), and a probability of 0.05 to select each of the other actions.

[0154] To improve learning efficiency, this method employs an experience replay mechanism. An experience pool D of capacity 10,000 is maintained, storing transition tuples (s, a, r, s'). During each training iteration, 128 samples are randomly selected from D for batch updates, reducing correlation between samples and improving learning stability. For example, when server cluster load fluctuates, the system may collect multiple state transition samples within a short period. Through the experience replay mechanism, these samples can be reused multiple times, accelerating algorithm convergence.

[0155] In actual deployment, the target network Q' is updated using a soft update strategy: every 100 learning steps, Q' = 0.01 × Q + (1 - 0.01) × Q' is executed, which means that 1% of the current network parameters are incorporated into the target network to maintain learning stability.

[0156] During algorithm iteration, the system monitors convergence. When the Q-function update amount is less than the preset threshold δ = 0.01 for 50 consecutive iterations, the algorithm is considered converged, and the optimal orchestration scheme is output. This convergence condition ensures the stability of the scheme and avoids premature stopping or excessive iteration.

[0157] In a test scenario, the initial setup consisted of 3 servers each deploying 5 microservices, with an average response time of 300ms and resource utilization of 65%. After approximately 5000 iterations, the algorithm converged to a solution: centrally deploying high-frequency services and distributed deployments of low-frequency services, while adding 2 instances of critical services. After implementing this solution, the system's average response time decreased to 180ms, resource utilization increased to 78%, and service availability improved from 99.5% to 99.9%.

[0158] This method effectively addresses the decision bias and local optima problems faced by traditional service orchestration by introducing dual Q-learning and an ε-greedy strategy, making it particularly suitable for dynamic service orchestration scenarios in cloud-native environments. Test results show that compared to traditional heuristic algorithms, this method improves resource utilization by 15% and reduces service response time by 40%, significantly optimizing overall system performance.

[0159] Example 5

[0160] This embodiment provides an intelligent scheduling system for substation equipment operation and maintenance tasks, used to implement the method described in Embodiment 1. The system includes:

[0161] Data acquisition module: Acquires real-time operating data and historical maintenance data of the power equipment. The real-time operating data includes the operating parameters, environmental parameters and alarm information of the power equipment. The historical maintenance data includes the inspection records, fault records and maintenance records of the power equipment.

[0162] Input information construction module: Based on the acquired real-time operation data and historical operation and maintenance data, the input information of the power large model is constructed using knowledge graph-enhanced Prompt technology. The knowledge graph includes the topological relationship of substation equipment, equipment attributes, operation and maintenance procedures and expert experience. By semantically associating the knowledge graph with real-time operation data and historical operation and maintenance data, a structured operation and maintenance task description is generated.

[0163] Operation and maintenance task orchestration scheme generation module: The structured operation and maintenance task description is input into the power big model, and the power big model outputs the operation and maintenance task prediction results of the substation equipment. The operation and maintenance task prediction results include the type, priority and expected workload of the operation and maintenance task. An operation and maintenance task orchestration scheme is generated based on the operation and maintenance task prediction results.

[0164] The operation and maintenance task orchestration scheme optimization module: dynamically optimizes the prediction results of the operation and maintenance tasks based on reinforcement learning methods. Specifically, it includes: constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives, using the actual execution feedback of operation and maintenance personnel as the environmental state, and obtaining an operation and maintenance task orchestration scheme that meets multiple constraints through iterative optimization. The multiple constraints include personnel qualification requirements, tool configuration requirements, safety interval requirements, and time window requirements.

[0165] Operation and maintenance work ticket generation module: Based on the optimized operation and maintenance task orchestration scheme, it generates operation and maintenance work tickets, which include work items, work content, work procedures, security measures and precautions;

[0166] Feedback optimization module: Issues maintenance work orders to maintenance personnel and collects task execution status and equipment status information in real time during the execution of maintenance tasks. It then feeds the task execution status and equipment status information back to the power big data model to continuously optimize the maintenance task scheduling scheme.

[0167] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0168] Example 6

[0169] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0170] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0171] The processing unit executes the various methods and processes described above, such as methods S1 to S6. For example, in some embodiments, methods S1 to S6 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1 to S6 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S6 by any other suitable means (e.g., by means of firmware).

[0172] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0173] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0174] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0175] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent scheduling of operation and maintenance tasks for substation equipment, characterized in that, This method acquires real-time operation data and historical maintenance data of substation equipment, uses knowledge graph-enhanced Prompt technology to construct input information for a large power model, uses the large power model to predict maintenance tasks, generates maintenance task scheduling schemes, and dynamically optimizes the maintenance task scheduling schemes based on reinforcement learning, generates and issues maintenance work tickets, and simultaneously feeds back task execution status and equipment status information to the large power model in real time to optimize the maintenance task scheduling schemes.

2. The intelligent scheduling method for operation and maintenance tasks of substation equipment according to claim 1, characterized in that, The method includes the following steps: Acquire real-time operating data and historical maintenance data of the power equipment. The real-time operating data includes the operating parameters, environmental parameters and alarm information of the power equipment. The historical maintenance data includes the inspection records, fault records and maintenance records of the power equipment. Based on the acquired real-time operation data and historical operation and maintenance data, the input information of the power large model is constructed using knowledge graph-enhanced Prompt technology. The knowledge graph includes the topological relationship of substation equipment, equipment attributes, operation and maintenance procedures and expert experience. By semantically associating the knowledge graph with real-time operation data and historical operation and maintenance data, a structured operation and maintenance task description is generated. The structured operation and maintenance task description is input into the power big model, and the power big model outputs the operation and maintenance task prediction results of the substation equipment. The operation and maintenance task prediction results include the type, priority and expected workload of the operation and maintenance task. An operation and maintenance task scheduling scheme is generated based on the operation and maintenance task prediction results. The operation and maintenance task orchestration scheme is dynamically optimized based on reinforcement learning methods. Specifically, it includes: constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives, using the actual execution feedback of operation and maintenance personnel as the environmental state, and obtaining an operation and maintenance task orchestration scheme that meets multiple constraints through iterative optimization. The multiple constraints include personnel qualification requirements, tool configuration requirements, safety interval requirements, and time window requirements. Based on the optimized operation and maintenance task orchestration scheme, an operation and maintenance work ticket is generated, which includes work items, work content, work procedures, security measures, and precautions. Operation and maintenance work orders are issued to operation and maintenance personnel, and task execution status and equipment status information are collected in real time during the execution of operation and maintenance tasks. The task execution status and equipment status information are fed back to the power big data model to continuously optimize the operation and maintenance task scheduling scheme.

3. The intelligent scheduling method for operation and maintenance tasks of substation equipment according to claim 2, characterized in that, The process of generating a structured description of operation and maintenance tasks includes: Construct a knowledge graph, which includes entity nodes of substation equipment and relational edges. The entity nodes include topology relation nodes, equipment attribute nodes, operation and maintenance procedure nodes, and expert experience nodes of substation equipment. The relational edges are used to represent the association relationships between entity nodes. The entity nodes in the knowledge graph are vectorized based on a deep learning model. Key features are extracted from the real-time operation data and historical operation and maintenance data through an attention mechanism. The extracted key features are semantically aligned with the vector representations of the entity nodes. A knowledge-enhanced Prompt template is constructed using a multi-head self-attention mechanism. The Prompt template includes a task type description, a device status description, a historical operation and maintenance description, and an expert knowledge description. Semantically aligned feature information is then filled into the corresponding positions in the Prompt template. A structured operation and maintenance task description is generated based on the populated Prompt template. The structured operation and maintenance task description includes device identification information, task triggering conditions, task execution requirements, and task completion standards.

4. The intelligent scheduling method for operation and maintenance tasks of substation equipment according to claim 2, characterized in that, The process of inputting the structured operation and maintenance task description into the power system big data model and outputting the operation and maintenance task prediction results of the substation equipment from the power system big data model includes: A large power model is constructed using a multimodal fusion pre-training method. The multimodal fusion pre-training method includes: inputting text data, time series data, and image data of power equipment into the corresponding encoders for feature extraction; aligning and fusing features of different modalities through a cross-modal attention mechanism; and training the large power model based on mask autoencoder and contrastive learning strategies. The structured operation and maintenance task description is input into the power big model to obtain the potential failure modes and health status of the substation equipment, and the operation and maintenance task prediction results of the substation equipment are output.

5. The intelligent scheduling method for operation and maintenance tasks of substation equipment according to claim 4, characterized in that, The aforementioned large-scale power model analyzes the potential fault modes and equipment health status based on a hierarchical task classification framework, and outputs prediction results for the operation and maintenance tasks of substation equipment, wherein: At the first level, the type of operation and maintenance task is determined through the equipment status assessment module. The types of operation and maintenance tasks include preventive maintenance tasks, fault repair tasks, and condition-based maintenance tasks. At the second level, the task urgency score is calculated through the task priority evaluation module, and the priority of the operation and maintenance task is determined based on the task urgency score. At the third level, the workload estimation module predicts the number of personnel, equipment configuration, and working time required for maintenance tasks based on historical execution data of similar tasks. The calculation results from the first level, the second level, and the third level are combined to generate the operation and maintenance task prediction results.

6. The intelligent scheduling method for operation and maintenance tasks of substation equipment according to claim 2, characterized in that, The reward function specifically includes: Operational efficiency reward Re(s,a): Re(s,a)=w1·(1 / T c )+w2·(N u / N t )+w3·(1 / T d ); Among them, T c N represents the total time to complete the task. u N represents the actual working hours of the personnel. t T represents the total available time. d This represents the equipment downtime, where w1, w2, and w3 are weighting coefficients. Equipment reliability bonus Rr(s,a): Rr(s,a)=α·H(s)+β·(1-F(s))+γ·Q(s); Where H(s) represents the equipment health function, F(s) represents the failure risk function, Q(s) represents the maintenance quality scoring function, and α, β, and γ are weighting coefficients; Constraint violation penalty Rp(s,a): Rp(s,a)=-λ1·max(0,1-M p )-λ2·max(0,1-M r )-λ3·max(0,1-M s )-λ4·max(0,1-M t ); Among them, M p M represents the degree of matching between personnel qualifications. r Indicating the rationality of tool and equipment configuration, M s M represents the degree of safety interval satisfaction. t The time window compliance is represented by λ1, λ2, λ3, and λ4, which are penalty coefficients.

7. The intelligent scheduling method for operation and maintenance tasks of substation equipment according to claim 2, characterized in that, The reinforcement learning method described above optimizes the policy based on the dual Q-learning algorithm. Construct the state-action value function Q(s,a) and the target value function Q'(s,a): Q(s,a)←Q(s,a)+η[R(s,a)+γ·maxQ'(s',a')-Q(s,a)]; Where η represents the learning rate, γ represents the discount factor, and R(s,a) represents the reward function; An ε-greedy strategy is used for action selection: π(a|s)={1-ε+ε / |A|,if a=argmaxQ(s,a);ε / |A|,otherwise}; Where π(a|s) represents the probability of choosing action a in state s, ε represents the exploration probability, A represents the action space, and |A| represents the size of the action space; The optimal operation and maintenance task orchestration scheme will be output when the following convergence conditions are met: |Q(k+1)(s,a)-Q(k)(s,a)|<δ; Where k represents the number of iterations and δ represents the convergence threshold.

8. A smart scheduling system for operation and maintenance tasks of substation equipment, characterized in that, For implementing the method as described in any one of claims 1 to 7, the system comprises: Data acquisition module: Acquires real-time operating data and historical maintenance data of the power equipment. The real-time operating data includes the operating parameters, environmental parameters and alarm information of the power equipment. The historical maintenance data includes the inspection records, fault records and maintenance records of the power equipment. Input information construction module: Based on the acquired real-time operation data and historical operation and maintenance data, the input information of the power large model is constructed using knowledge graph-enhanced Prompt technology. The knowledge graph includes the topological relationship of substation equipment, equipment attributes, operation and maintenance procedures and expert experience. By semantically associating the knowledge graph with real-time operation data and historical operation and maintenance data, a structured operation and maintenance task description is generated. Operation and maintenance task orchestration scheme generation module: The structured operation and maintenance task description is input into the power big model, and the power big model outputs the operation and maintenance task prediction results of the substation equipment. The operation and maintenance task prediction results include the type, priority and expected workload of the operation and maintenance task. An operation and maintenance task orchestration scheme is generated based on the operation and maintenance task prediction results. The operation and maintenance task orchestration scheme optimization module: dynamically optimizes the prediction results of the operation and maintenance tasks based on reinforcement learning methods. Specifically, it includes: constructing a reward function with operation and maintenance efficiency and equipment reliability as optimization objectives, using the actual execution feedback of operation and maintenance personnel as the environmental state, and obtaining an operation and maintenance task orchestration scheme that meets multiple constraints through iterative optimization. The multiple constraints include personnel qualification requirements, tool configuration requirements, safety interval requirements, and time window requirements. Operation and maintenance work ticket generation module: Based on the optimized operation and maintenance task orchestration scheme, it generates operation and maintenance work tickets, which include work items, work content, work procedures, security measures and precautions; Feedback optimization module: Issues maintenance work orders to maintenance personnel and collects task execution status and equipment status information in real time during the execution of maintenance tasks. It then feeds the task execution status and equipment status information back to the power big data model to continuously optimize the maintenance task scheduling scheme.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method of any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method of any one of claims 1 to 7.

Images