Smart grid fault self-healing method and system based on generative artificial intelligence

Abstract

The present invention discloses a method and system for smart grid fault self-healing based on generative artificial intelligence, which relates to the field of smart grid fault self-healing. The method includes: obtaining and preprocessing multi-source heterogeneous data; generating a pre-trained language model and performing domain adaptability fine-tuning, adding a task-specific output layer; performing feature fusion using a multimodal data fusion method, and using annotated historical fault data sets for supervised training to obtain fault diagnosis results. Fault self-healing is modeled as a Markov decision process, and a deep reinforcement learning algorithm is used to interactively learn the optimal decision-making strategy with a power grid simulation environment, and an optimal fault self-healing plan is generated and converted into specific operation instructions. The present invention integrates multi-source heterogeneous data, performs accurate positioning and description generation, improves the reliability and self-healing ability of the smart grid, and ensures rapid recovery of normal operation when a power grid fails.

Description

Technical Field

[0001] The present invention relates to the field of smart grid fault self-healing, and in particular to a smart grid fault self-healing method and system based on generative artificial intelligence. Background Art

[0002] The rapid development of smart grid technology has greatly improved the automation and intelligence level of distribution networks, enabling more efficient and stable grid operation. However, despite significant progress, existing fault diagnosis and self-healing systems still face many challenges when dealing with complex and sudden grid faults.

[0003] Existing fault diagnosis methods primarily rely on expert experience and pre-set rule bases. This approach performs reasonably well for routine faults, but its diagnostic accuracy and reliability significantly decline when faced with multiple faults, cascading faults, or new, complex faults. While expert experience is valuable, its update speed struggles to keep pace with the increasingly complex nature of the power grid, while rule bases are often too rigid to cover all possible fault scenarios. Furthermore, existing fault self-healing systems typically handle faults through pre-set emergency plans, which are often based on historical data and experience from past fault cases and lack real-time and adaptability. Such systems struggle to learn from new fault cases and adjust response strategies in real time, resulting in a very limited ability to adapt to dynamic changes in the power grid environment. With the increasing complexity of power grid structures and the ever-changing operating environment, traditional fault diagnosis and self-healing methods are no longer able to meet the demanding requirements of modern smart grids. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to improve the accuracy and reliability of fault diagnosis, especially for complex multiple faults and new faults, and give the system the ability to continuously learn and self-improve, so that it can continuously adapt to changes in the power grid environment.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a smart grid fault self-healing method based on generative artificial intelligence, comprising:

[0008] Acquire multi-source heterogeneous data and perform data preprocessing;

[0009] Based on the preprocessed multi-source heterogeneous data, a pre-trained language model is generated and fine-tuned for domain adaptation, adding a task-specific output layer.

[0010] Based on the pre-trained language model, a multimodal data fusion method is used to fuse features and generate a comprehensive feature representation. Supervised training is performed using annotated historical fault datasets. Fault diagnosis results are obtained based on the pre-trained language model after supervised training.

[0011] The fault self-healing process is modeled as a Markov decision process. Using a deep reinforcement learning algorithm, the optimal decision strategy is learned through interaction with the power grid simulation environment. Based on the fault diagnosis results and the current power grid status, the optimal fault self-healing solution is generated.

[0012] Convert the optimal fault self-recovery plan into specific equipment operation instructions, and schedule the execution of each operation according to priority and timing requirements.

[0013] As a preferred solution for the smart grid fault self-healing method based on generative artificial intelligence, it:

[0014] The domain adaptation fine-tuning includes:

[0015] Corpus collection: Screening text materials in the power field, including professional literature, technical reports, fault records, and equipment manuals;

[0016] Data preprocessing: Clean, segment, and standardize the collected text to remove irrelevant information and noise;

[0017] Domain vocabulary expansion: Expand the model vocabulary based on professional terms in the power field;

[0018] Continuous pre-training: Using processed power sector corpus, the pre-trained model is further trained to adapt it to the domain-specific language patterns and knowledge structure.

[0019] Evaluation and tuning: Use domain-related evaluation tasks to evaluate the domain adaptability of the model and perform fine-tuning.

[0020] As a preferred solution for the smart grid fault self-healing method based on generative artificial intelligence, it:

[0021] The multimodal data fusion method includes:

[0022] Time series feature extraction: processing time series data;

[0023] Waveform feature analysis: Apply convolutional neural networks to extract features from fault waveforms and identify fault characteristic signals;

[0024] Text feature processing: Use pre-trained language models to process text information including fault reports and operation logs and extract semantic features;

[0025] Graph structure features: Use graph neural networks to model the power grid topology and capture the propagation characteristics of faults in the network;

[0026] Feature fusion: Design attention mechanisms or gating units to dynamically fuse features from different modalities to generate a comprehensive representation;

[0027] Model training: Supervised training is performed using a labeled historical fault dataset to optimize the joint loss function of multiple tasks.

[0028] As a preferred solution for the smart grid fault self-healing method based on generative artificial intelligence, it:

[0029] The adding of a task-specific output layer comprises:

[0030] The task-specific output layers are used for fault classification, fault location, and fault description generation respectively;

[0031] Fault classification: classify faults into different types;

[0032] Fault location: determine the specific location where the fault occurred;

[0033] Fault description generation: automatically generate detailed fault description;

[0034] The task-specific output layer also includes fault duration prediction: estimating the duration of the fault to provide a reference for subsequent repair and scheduling;

[0035] Fault impact assessment: Assess the impact of the fault on grid stability and user power supply.

[0036] As a preferred solution for the smart grid fault self-healing method based on generative artificial intelligence, it:

[0037] The multi-source heterogeneous data includes: time series data: a time series of voltage, current, and power measurement values; waveform data: voltage and current waveforms recorded by a fault recorder; text data: unstructured text including fault reports and equipment maintenance records; topology data: power grid structure and equipment connection relationships.

[0038] As a preferred solution for the smart grid fault self-healing method based on generative artificial intelligence, it:

[0039] Modeling the fault self-healing process as a Markov decision process includes:

[0040] The Markov decision process consists of a state space, an action space, a transition probability function, a reward function, and a discount factor, expressed as a five-tuple (S, A, P, R, γ):

[0041] The state space is represented as:

[0042] S=(T,D,L)

[0043] Where: T represents the grid topology, D represents the device status, and L represents the load condition;

[0044] The action vector is expressed as:

[0045] a=(O,T,G,R,P)

[0046] Where: O is the switch operation vector, T is the load transfer matrix, G is the generation adjustment vector, R is the reconstruction operation, and P is the protection setting adjustment vector;

[0047] The state transition probability function is expressed as:

[0048] P(s′|s,a)=Pr(S t+1 =s′|St=s,A t =a)

[0049] The reward function is expressed as:

[0050] R(s,a,s′)=-αT-βA-γC

[0051] Where: α, β, γ are weight factors used to balance the importance of different factors; T, A, C represent power restoration speed, number of affected users, and operation cost, respectively;

[0052] Express the discount factor as:

[0053] γ∈[0,1].

[0054] As a preferred solution for the smart grid fault self-healing method based on generative artificial intelligence, it:

[0055] Converting the optimal fault self-healing solution into specific equipment operation instructions includes:

[0056] The instruction parsing is represented as:

[0057] I=f(S)

[0058] Where: I is the generated operation instruction set; S is the self-healing solution; f is the instruction parsing function;

[0059] Express the execution schedule as:

[0060] E=g(I,P,T)

[0061] Where: E is the execution plan; I is the set of operation instructions; P is the priority of the operation; T is the timing requirement of the operation; g is the execution scheduling function;

[0062] Express real-time monitoring as:

[0063] M=h(S t )

[0064] Where: M is the monitoring report; S t is the grid state at time t; h is the monitoring function;

[0065] Express exception handling as:

[0066] A=i(E,M)

[0067] Among them: A is the exception handling action; E is the execution plan; M is the monitoring report; i is the exception handling function.

[0068] In a second aspect, an embodiment of the present invention provides a smart grid fault self-healing system based on generative artificial intelligence, comprising:

[0069] Data preprocessing module, used to obtain multi-source heterogeneous data and perform data preprocessing;

[0070] The pre-trained model generation module is used to generate a pre-trained language model based on pre-processed multi-source heterogeneous data, perform domain-adaptive fine-tuning, and add task-specific output layers;

[0071] The fault diagnosis module is used to perform feature fusion using a multimodal data fusion method based on a pre-trained language model to generate a comprehensive feature representation; supervised training is performed using annotated historical fault datasets; and fault diagnosis results are obtained based on the pre-trained language model after supervised training.

[0072] The self-healing solution generation module is used to model the fault self-healing process as a Markov decision process. It uses a deep reinforcement learning algorithm to learn the optimal decision strategy through interaction with the power grid simulation environment. It generates the optimal fault self-healing solution based on the fault diagnosis results and the current power grid status.

[0073] The decision-making execution module is used to convert the optimal fault self-recovery plan into specific equipment operation instructions and schedule the execution of various operations according to priority and timing requirements.

[0074] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0075] memory and processor;

[0076] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the smart grid fault self-healing method based on generative artificial intelligence as described in any embodiment of the present invention.

[0077] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the generative artificial intelligence-based smart grid fault self-healing method.

[0078] Beneficial effects of the present invention: The present invention utilizes large-scale language models and deep reinforcement learning technology to realize an intelligent and adaptive fault diagnosis and self-healing decision-making framework; integrates multi-source heterogeneous data to accurately identify and locate faults; adopts a multi-task learning framework, combined with pre-trained language models and feature fusion technology, to realize fault classification, location, description generation, duration prediction and impact assessment; can improve the reliability and self-healing capabilities of smart grids, and ensure that the grid can quickly resume normal operation when a fault occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0080] Figure 1 This is an overall flow chart of the smart grid fault self-healing method based on generative artificial intelligence described in the first embodiment of the present invention. DETAILED DESCRIPTION

[0081] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0082] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0083] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0084] Example 1

[0085] Reference Figure 1 , which is the first embodiment of the present invention, provides a smart grid fault self-healing method based on generative artificial intelligence, comprising:

[0086] S1: Acquire multi-source heterogeneous data and perform data preprocessing;

[0087] S2: Based on the preprocessed multi-source heterogeneous data, a pre-trained language model is generated and fine-tuned for domain adaptation, adding a task-specific output layer.

[0088] S3: Based on the pre-trained language model, multimodal data fusion methods are used to perform feature fusion to generate a comprehensive feature representation; supervised training is performed using annotated historical fault datasets; and fault diagnosis results are obtained based on the pre-trained language model after supervised training.

[0089] S4: Model the fault self-healing process as a Markov decision process. Utilize a deep reinforcement learning algorithm to learn the optimal decision strategy through interaction with the power grid simulation environment. Generate the optimal fault self-healing solution based on the fault diagnosis results and the current power grid status.

[0090] S5: Convert the optimal fault self-recovery plan into specific equipment operation instructions and schedule the execution of each operation according to priority and timing requirements.

[0091] The multi-source heterogeneous data in the above step S1 include: time series data: time series of measured values ​​of voltage, current, and power; waveform data: voltage and current waveforms recorded by the fault recorder; text data: unstructured text including fault reports and equipment maintenance records; topology data: grid structure and equipment connection relationship.

[0092] In an embodiment of the present application, multi-source heterogeneous data can be obtained from smart grid sensors, measurement equipment, and historical databases.

[0093] The data preprocessing in step S1 above includes: data cleaning: removing outliers, filling missing values, and eliminating noise; data alignment: unifying the timestamps and sampling rates of data from different sources; feature extraction: extracting representative features from the original data; data standardization: normalizing data of different scales to a unified range.

[0094] The domain adaptation fine-tuning in step S2 above includes:

[0095] Corpus collection: Screening of professional literature, technical reports, fault records, equipment manuals, and other text materials in the power field;

[0096] Data preprocessing: Clean, segment, and standardize the collected text to remove irrelevant information and noise;

[0097] Domain vocabulary expansion: Expand the model vocabulary based on professional terms in the power field;

[0098] Continuous pre-training: Using processed power sector corpus, the pre-trained model is further trained to adapt it to the domain-specific language patterns and knowledge structure.

[0099] Evaluation and tuning: Use domain-related evaluation tasks to evaluate the domain adaptability of the model and perform fine-tuning.

[0100] In the embodiment of the present application, the domain-related evaluation tasks may be professional terminology recognition and relationship extraction;

[0101] In another possible implementation, the domain-related evaluation task may also include named entity recognition:

[0102] Identify and classify specific entities in the power sector (such as equipment name, model, manufacturer, fault type, etc.). This helps the model understand and analyze power-related text data more accurately.

[0103] In another possible implementation, the domain-related evaluation task may also include text classification:

[0104] The text data in the power sector (such as fault reports, maintenance records, etc.) are classified according to preset categories (such as fault type, severity, treatment suggestions, etc.). This helps to achieve rapid screening and classification of large amounts of text data.

[0105] In another possible implementation, the domain-related evaluation task may also include sentiment analysis:

[0106] Sentiment analysis of text data in the power sector can identify positive, negative, or neutral sentiment expressed in the text. This is crucial for assessing customer satisfaction, equipment operating status, and troubleshooting effectiveness.

[0107] Adding a task-specific output layer in step S2 above includes:

[0108] The task-specific output layers are used for fault classification, fault location, and fault description generation. Fault classification divides faults into different types. Fault location determines the specific location of the fault. Fault description generation automatically generates a detailed description of the fault situation. It also includes fault duration prediction, which estimates the duration of the fault to provide a reference for subsequent repair and scheduling. Fault impact assessment assesses the impact of the fault on grid stability and user power supply.

[0109] In the embodiment of the present application, faults are classified into different types such as short circuit, overload, and equipment failure; the specific location of the fault is determined, including the substation, line section, or specific equipment; and a detailed description of the fault situation is automatically generated, including the cause of the fault, the scope of impact, and the severity;

[0110] In another possible implementation, classifying the faults into different types may further include:

[0111] Ground fault: including single-phase ground fault, two-phase ground fault, etc.

[0112] Circuit breaker fault: such as line break, switch tripping, etc.

[0113] Overvoltage fault: such as lightning overvoltage, operational overvoltage, etc.

[0114] Harmonic fault: equipment overheating, protection malfunction, etc. caused by harmonics.

[0115] Communication failures: such as interruptions or errors in signal transmission in smart grid communication networks.

[0116] Human operational errors: such as equipment failure or system instability caused by incorrect operation.

[0117] Failures caused by natural disasters: Equipment damage caused by natural disasters such as earthquakes, floods, and storms.

[0118] In another possible implementation, determining the specific location where the fault occurs may further include:

[0119] Distribution network level: such as specific locations in low-voltage distribution lines and medium-voltage distribution lines.

[0120] User-side equipment: such as transformers, circuit breakers, switch cabinets and other specific equipment on the user side.

[0121] Distributed energy systems: such as fault points in photovoltaic power generation systems and wind power generation systems.

[0122] Smart terminal devices: such as fault points in smart meters and smart home devices.

[0123] Communication network nodes: such as fiber optic communication nodes, wireless communication base stations, and other specific locations.

[0124] In another possible implementation, automatically generating a detailed fault description may further include:

[0125] Time and date of fault occurrence: Record the specific time when the fault occurred, including year, month, day, hour, minute, and second.

[0126] System status before the fault occurs: describes the system's working status before the fault occurs, such as load conditions, voltage level, frequency, etc.

[0127] Environmental conditions when the fault occurs: such as weather conditions, temperature, humidity, and other environmental factors that may affect the fault.

[0128] Troubleshooting process records: including the troubleshooting steps, tools and equipment used, personnel involved in the handling and their operation records.

[0129] Follow-up measures for troubleshooting: such as temporary measures taken, subsequent repair plans, suggestions for preventing similar failures from happening again, etc.

[0130] Impact of the fault on users: Record the impact of the fault on users' power supply, such as the scope of the power outage, duration of the power outage, number of affected users, etc.

[0131] Economic loss assessment of failures: Evaluate the direct and indirect economic losses caused by failures to provide data support for accident analysis and cost control.

[0132] The multimodal data fusion method in step S3 includes:

[0133] Time series feature extraction: processing time series data;

[0134] In the embodiments of the present application, a long short-term memory network (LSTM) or a Transformer model is used to process time series data; the formula of the LSTM unit is:

[0135] h t =σ(W ih x t +b ih +W hh h t-1 +b hh +W ig c t-1 +b ig )

[0136] c t =f t *c t-1 +i t *tanh(W ic x t +b ic )

[0137] Where: h t is the hidden state at time step t; c t is the cell state at time step t; x t is the input at time step t; W and b are model parameters; σ is the sigmoid activation function; f t and i t They are the forget gate and the input gate respectively.

[0138] Waveform feature analysis: Apply convolutional neural networks to extract features from fault waveforms and identify fault characteristic signals.

[0139] In the embodiment of the present application, the formula of the convolutional layer is:

[0140] f(x)=max(0,x*wb)

[0141] Where: x is the input image; w is the convolution kernel; b is the bias term; * represents the convolution operation.

[0142] Text feature processing: Use pre-trained language models to process text information including fault reports and operation logs and extract semantic features;

[0143] In the embodiment of the present application, the pre-trained language model used is BERT, and the self-attention mechanism formula is:

[0144]

[0145] Where: Q is the query matrix; K is the key matrix; V is the value matrix; d k is the dimension of the key vector

[0146] In another possible implementation, the pre-trained language model used may also be RoBERTa (Robustly Optimized BERTPre-training Approach):

[0147] RoBERTa is an optimized version of BERT, optimized with a larger dataset, longer training time, and more training steps. It is suitable for large-scale text data processing and high-precision text classification tasks. The self-attention mechanism formula is the same as BERT, but the parameter settings and data processing during training are more sophisticated.

[0148] In another possible implementation, the pre-trained language model used may also be XLNet:

[0149] XLNet combines the advantages of Transformer-XL and pre-trained language models, using a permutation language model (PLM) to capture contextual information. It is suitable for tasks that require capturing complex contextual relationships, such as semantic understanding of long texts and question-answering systems. The self-attention mechanism formula is similar to BERT, but the ability to capture contextual information is enhanced by using a permutation language model.

[0150] In another possible implementation, the pre-trained language model used may also be ALBERT (ALiteBERT):

[0151] ALBERT achieves more efficient training and inference by reducing model parameters and optimizing the training process; it is suitable for resource-constrained environments, such as natural language processing tasks in mobile devices or embedded systems; the self-attention mechanism formula is the same as BERT, but the model size is reduced through techniques such as parameter sharing and matrix decomposition.

[0152] Graph structure features: Use graph neural networks to model the power grid topology and capture the propagation characteristics of faults in the network.

[0153] In the embodiment of the present application, the graph neural network used is GNN, and the formula of the graph convolution layer is:

[0154]

[0155] in: is the feature of node v at layer l; W (l) is the weight matrix of the lth layer; N(v) is the set of neighbor nodes of node v.

[0156] Feature fusion: Design an attention mechanism or gating unit to dynamically fuse features of different modalities to generate a comprehensive representation. In this embodiment of the application, the gating unit formula is:

[0157] g=σ(W g x)

[0158] c=g*h1+(1-g)*h2

[0159] Where: g is the gating coefficient; h1 and h2 are the features of different modes; σ is the sigmoid activation function.

[0160] Model training: Supervised training is performed using a labeled historical fault dataset to optimize the joint loss function of multiple tasks.

[0161] In step S4 above, the fault self-healing process is modeled as a Markov decision process, which includes:

[0162] The Markov decision process consists of a state space, an action space, a transition probability function, a reward function, and a discount factor, expressed as a five-tuple (S, A, P, R, γ):

[0163] The state space is represented as:

[0164] S=(T,D,L)

[0165] Where: T represents the grid topology, D represents the device status, and L represents the load condition.

[0166] It should be noted that the state space S is the set of all possible situations that a smart device may encounter. In a smart grid fault self-healing system, the state space includes information such as the grid topology, device status, and load conditions. This information collectively describes the current state of the grid.

[0167] The action vector is expressed as:

[0168] a=(O,T,G,R,P)

[0169] Where: O is the switch operation vector, T is the load transfer matrix, G is the generation adjustment vector, R is the reconstruction operation, and P is the protection setting adjustment vector.

[0170] It should be noted that the action space A includes possible self-healing operations such as switching operations, load transfer, and power scheduling.

[0171] The state transition probability function is expressed as:

[0172] P(s′|s,a)=Pr(S t+1 =s′|St=s,A t =a)

[0173] It should be noted that the state transition probability function P describes the probability of transitioning from state s to state s′ after executing action a. In smart grid fault self-healing systems, state transitions may be affected by various factors, such as equipment failure rate, load variation, and operation success rate.

[0174] The reward function is expressed as:

[0175] R(s,a,s′)=-αT-βA-γC

[0176] Where: α, β, and γ are weight factors used to balance the importance of different factors. T, A, and C represent power restoration speed, number of affected users, and operation cost, respectively.

[0177] It should be noted that the reward function R considers factors such as power restoration speed, number of affected users, and operation cost to evaluate the effectiveness of the action.

[0178] Express the discount factor as:

[0179] γ∈[0,1]

[0180] It should be noted that the discount factor γ is a parameter between 0 and 1 that determines the importance of future rewards relative to immediate rewards. The discount factor allows the agent to balance short-term and long-term gains.

[0181] It should also be noted that by modeling the fault self-healing process as a Markov decision process, the smart grid fault self-healing system can apply reinforcement learning algorithms to learn optimal decision-making strategies. By interacting with the grid simulation environment, the system continuously optimizes the decision-making process to achieve rapid and effective fault self-healing.

[0182] In step S5, converting the optimal fault self-healing solution into specific equipment operation instructions includes:

[0183] The instruction parsing is represented as:

[0184] I=f(S)

[0185] Where: I is the generated operation instruction set; S is the self-healing solution; f is the instruction parsing function.

[0186] For example, if the self-healing plan requires "isolating the fault area and reconfiguring the grid to restore power supply", operation instructions such as closing specific circuit breakers, adjusting the tap position of transformers, and reconfiguring the setting values ​​of protection relays will be generated.

[0187] Express the execution schedule as:

[0188] E=g(I,P,T)

[0189] Among them: E is the execution plan; I is the operation instruction set; P is the priority of the operation; T is the timing requirement of the operation; g is the execution scheduling function.

[0190] It should be noted that execution scheduling involves scheduling the execution of various operations based on priority and timing requirements. In grid fault self-healing, some operations may need to be executed first, while others can be performed later. This requires considering the dependencies between operations and the real-time state of the grid.

[0191] Express real-time monitoring as:

[0192] M=h(S t )

[0193] Where: M is the monitoring report; S t is the grid status at time t; h is the monitoring function.

[0194] It should be noted that real-time monitoring involves monitoring changes in grid status during the execution of the self-healing solution. This step is crucial for ensuring the solution is executed as planned and for responding to any unexpected situations in a timely manner. Real-time monitoring can be achieved through real-time data collection and status assessment.

[0195] Express exception handling as:

[0196] A=i(E,M)

[0197] Among them: A is the exception handling action; E is the execution plan; M is the monitoring report; i is the exception handling function.

[0198] It should be noted that exception handling is the process of detecting and handling abnormal situations that may occur during execution. These may include equipment failure, operational failure, or external interference. It is necessary to be able to identify abnormal situations and take appropriate remedial measures.

[0199] In the embodiment of this application, a continuous learning module can also be added to enable the system to continuously learn and optimize from new fault cases and self-healing experiences, thereby improving the adaptability and efficiency of the system. The main functions include:

[0200] Experience replay: This involves storing and managing historical failure cases and self-healing experiences in an experience database. This data, including past failure scenarios, measures taken, and their effects, is used to train and optimize models. The experience replay process can be represented as follows:

[0201] E={(e1,r1),(e2,r2),...,(e n ,r n )}

[0202] Among them: E is the experience base; e i is the i-th fault case; r i is the corresponding self-healing result; n is the number of cases in the experience database.

[0203] Knowledge distillation: Knowledge distillation is the process of extracting key knowledge from new cases and updating the system's knowledge base. This process involves extracting valuable information from experience replay and converting it into knowledge that the model can understand and apply. Knowledge distillation can be expressed as:

[0204] K=D(E)

[0205] Among them: K is the knowledge base; D is the knowledge distillation process.

[0206] Model fine-tuning: It is the incremental update of the neural network model of each module based on new data. This process involves fine-tuning the model weights using new empirical data to improve the accuracy and robustness of the model. Model fine-tuning can be expressed as:

[0207] θ n =θ o +Δθ

[0208] Where: θ n are the updated model parameters; θ o is the model parameter before updating; Δθ is the parameter update amount.

[0209] Performance evaluation: This is the process of continuously evaluating the system's performance in new scenarios and identifying areas for improvement. This process involves using various evaluation metrics to measure the model's performance and making adjustments based on the evaluation results. Performance evaluation can be expressed as:

[0210] P=f(M,D)

[0211] Where: P is the performance evaluation result; M is the model; D is the evaluation dataset; f is the evaluation function.

[0212] It should be noted that the continuous learning module, through functions such as experience replay, knowledge distillation, model fine-tuning, and performance evaluation, enables the smart grid fault self-healing system to continuously improve and adapt to new situations. This generative AI-based approach not only improves the system's self-healing efficiency but also enhances its resilience to unknown faults. Through continuous learning and optimization, the system can better ensure the stable and reliable operation of the smart grid.

[0213] The above is a schematic diagram of the generative artificial intelligence-based smart grid fault self-healing method of this embodiment. It should be noted that the technical solution of the generative artificial intelligence-based smart grid fault self-healing system and the technical solution of the generative artificial intelligence-based smart grid fault self-healing method described above are based on the same concept. For details not described in detail in the technical solution of the generative artificial intelligence-based smart grid fault self-healing system of this embodiment, please refer to the description of the technical solution of the generative artificial intelligence-based smart grid fault self-healing method described above.

[0214] The smart grid fault self-healing system based on generative artificial intelligence in this embodiment includes:

[0215] Data preprocessing module, used to obtain multi-source heterogeneous data and perform data preprocessing;

[0216] The pre-trained model generation module is used to generate a pre-trained language model based on pre-processed multi-source heterogeneous data, perform domain-adaptive fine-tuning, and add task-specific output layers;

[0217] The fault diagnosis module is used to perform feature fusion using a multimodal data fusion method based on a pre-trained language model to generate a comprehensive feature representation; supervised training is performed using annotated historical fault datasets; and fault diagnosis results are obtained based on the pre-trained language model after supervised training.

[0218] The self-healing solution generation module is used to model the fault self-healing process as a Markov decision process. It uses a deep reinforcement learning algorithm to learn the optimal decision strategy through interaction with the power grid simulation environment. It generates the optimal fault self-healing solution based on the fault diagnosis results and the current power grid status.

[0219] The decision-making execution module is used to convert the optimal fault self-recovery plan into specific equipment operation instructions and schedule the execution of various operations according to priority and timing requirements.

[0220] This embodiment further provides a computing device applicable to a smart grid fault self-healing method based on generative artificial intelligence, including:

[0221] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the smart grid fault self-healing method based on generative artificial intelligence as proposed in the above embodiment.

[0222] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the smart grid fault self-healing method based on generative artificial intelligence as proposed in the above embodiment.

[0223] The storage medium proposed in this embodiment and the smart grid fault self-healing method based on generative artificial intelligence proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0224] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A smart grid fault self-healing method based on generative artificial intelligence, characterized in that: include: Acquire multi-source heterogeneous data and perform data preprocessing; Based on the preprocessed multi-source heterogeneous data, a pre-trained language model is generated and fine-tuned for domain adaptation, and a task-specific output layer is added; Based on the pre-trained language model, the multimodal data fusion method is used to perform feature fusion and generate comprehensive feature representation; Use annotated historical fault datasets for supervised training; Obtain fault diagnosis results based on the pre-trained language model after supervised training; The fault self-healing process is modeled as a Markov decision process, and the deep reinforcement learning algorithm is used to learn the optimal decision strategy through interaction with the power grid simulation environment; Generate the optimal fault self-healing plan based on the fault diagnosis results and the current power grid status; Convert the optimal fault self-healing plan into specific equipment operation instructions, and schedule the execution of various operations according to priority and timing requirements.

2. The method for self-healing faults of smart grid based on generative artificial intelligence according to claim 1, characterized in that: The domain adaptation fine-tuning includes: Corpus collection: Screening text materials in the power field including professional literature, technical reports, fault records and equipment manuals; Data preprocessing: Clean, segment and standardize the collected text to remove irrelevant information and noise; Domain vocabulary expansion: expand the model vocabulary based on professional terms in the power field; Continuous pre-training: Use the processed power field corpus to further train the pre-trained model to adapt it to the domain-specific language patterns and knowledge structure; Evaluation and tuning: Through domain-related evaluation tasks, the domain adaptability of the model is evaluated and tuned.

3. The method for self-healing faults of smart grid based on generative artificial intelligence according to claim 2, characterized in that: The multimodal data fusion method comprises: Time series feature extraction: processing time series data; Waveform feature analysis: Apply convolutional neural network to extract features of fault waveform and identify fault feature signals; Text feature processing: Use pre-trained language models to process text information including fault reports and operation logs and extract semantic features; Graph structure features: Use graph neural networks to model the power grid topology and capture the propagation characteristics of faults in the network; Feature fusion: Design attention mechanisms or gating units to dynamically fuse features of different modalities to generate comprehensive representations; Model training: Use annotated historical fault datasets for supervised training and optimize the joint loss function of multiple tasks.

4. The method for self-healing faults of smart grid based on generative artificial intelligence according to claim 3, characterized in that: The adding of a task-specific output layer comprises: The task-specific output layers are used for fault classification, fault location, and fault description generation, respectively; Fault classification: classify faults into different types; Fault location: determine the specific location where the fault occurred; Fault description generation: Automatically generate detailed fault description; The task-specific output layer also includes fault duration prediction: estimating the duration of the fault to provide a reference for subsequent repair and scheduling; Fault impact assessment: Assess the impact of the fault on grid stability and user power supply.

5. The method for self-healing faults of smart grid based on generative artificial intelligence according to claim 4, characterized in that: The multi-source heterogeneous data include: time series data: time series of measured values ​​of voltage, current and power; waveform data: voltage and current waveforms recorded by fault recorders; text data: unstructured text including fault reports and equipment maintenance records; topology data: power grid structure and equipment connection relationship.

6. The method for self-healing faults of smart grid based on generative artificial intelligence according to claim 5, characterized in that: Modeling the fault self-healing process as a Markov decision process includes: The Markov decision process consists of a state space, an action space, a transition probability function, a reward function, and a discount factor, expressed as a five-tuple (S, A, P, R, γ): The state space is represented as: S=(T,D,L) Where: T represents the grid topology, D represents the device status, and L represents the load condition; The action vector is represented as: a=(O,T,G,R,P) Where: O is the switch operation vector, T is the load transfer matrix, G is the generation adjustment vector, R is the reconstruction operation, and P is the protection setting adjustment vector; The state transition probability function is expressed as: P(s′|s,a)=Pr(S t+1 =s′|St=s,A t =a) The reward function is expressed as: R(s,a,s′)=-αT-βA-γC Among them: α, β, γ are weight factors used to balance the importance of different factors; T, A, C represent the power restoration speed, the number of affected users and the operation cost respectively; The discount factor is expressed as: γ∈[0,1].

7. The method for self-healing faults of smart grid based on generative artificial intelligence according to claim 6, characterized in that: The conversion of the optimal fault self-healing solution into specific equipment operation instructions includes: The instruction parsing is represented as: I=f(S) Where: I is the generated operation instruction set; S is the self-healing solution; f is the instruction parsing function; The execution schedule is expressed as: E=g(I,P,T) Where: E is the execution plan; I is the set of operation instructions; P is the priority of the operation; T is the timing requirement of the operation; g is the execution scheduling function; Express real-time monitoring as: M=h(S t ) Where: M is the monitoring report; S t is the grid status at time t; h is the monitoring function; Express exception handling as: A=i(E,M) Among them: A is the exception handling action; E is the execution plan; M is the monitoring report; i is the exception handling function.

8. A system using the smart grid fault self-healing method based on generative artificial intelligence as claimed in any one of claims 1 to 7, characterized in that: include: Data preprocessing module, used to obtain multi-source heterogeneous data and perform data preprocessing; The pre-trained model generation module is used to generate a pre-trained language model based on pre-processed multi-source heterogeneous data, perform domain-adaptive fine-tuning, and add a task-specific output layer; The fault diagnosis module is used to perform feature fusion based on the pre-trained language model using a multimodal data fusion method to generate a comprehensive feature representation; Use annotated historical fault datasets for supervised training; Obtain fault diagnosis results based on the pre-trained language model after supervised training; The self-healing solution generation module is used to model the fault self-healing process as a Markov decision process and use the deep reinforcement learning algorithm to learn the optimal decision strategy through interaction with the power grid simulation environment; Generate the optimal fault self-healing plan based on the fault diagnosis results and the current power grid status; The decision-making execution module is used to convert the optimal fault self-healing plan into specific equipment operation instructions and schedule the execution of various operations according to priority and timing requirements.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the smart grid fault self-healing method based on generative artificial intelligence as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the smart grid fault self-healing method based on generative artificial intelligence as described in any one of claims 1 to 7.

Images