Power distribution cabinet maintenance system based on artificial intelligence

Abstract

The invention discloses a power distribution cabinet maintenance system based on artificial intelligence, and belongs to the technical field of intelligent power distribution cabinet diagnosis. Normalization, time sequence feature extraction and denoising are carried out on the collected data, an electrical topological graph is automatically constructed, node states are vectorized, text semantics are coded by utilizing a BERT class model, and a maintenance knowledge graph is generated; through fusion of multi-source data, a multi-branch neural network is constructed, state perception and risk determination are realized, and a fault trend and structure degradation are identified. Calculating a potential fault risk coefficient, and triggering a structure health monitoring mechanism; analyzing the structural health based on a topological graph and a graph neural network, and starting semantic strategy retrieval when the structure is abnormal; comparing the semantic conformity between the state and the historical strategy, and assisting in generating a precise maintenance strategy; according to the system, dynamic monitoring, intelligent early warning and strategy recommendation of the operation state of the power distribution cabinet are realized, and the operation and maintenance efficiency and the equipment reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent power distribution cabinet diagnostic technology, specifically to an artificial intelligence-based power distribution cabinet maintenance system. Background Technology

[0002] With the continuous intelligent development of power systems, distribution cabinets, as key nodes in the power distribution system, directly affect the safety and reliability of power supply. Currently, distribution cabinets widely used in industrial parks, residential communities, and important public facilities typically face the following technical challenges: Firstly, traditional distribution cabinet maintenance methods heavily rely on manual inspections, resulting in problems such as untimely response, insufficient coverage, and high labor costs. Manual operation, especially in high-temperature, high-pressure, and hazardous environments, poses significant safety hazards. Secondly, distribution cabinets may experience various faults during operation, such as overload, short circuit, arc discharge, loose contacts, and abnormal temperature. Failure to accurately identify these faults in their early stages can lead to equipment damage or even fires, causing significant losses.

[0003] In existing technologies, some sensor-based real-time monitoring systems can collect and alarm parameters such as voltage, current, temperature and humidity, but they generally suffer from problems such as isolated monitoring data, lack of in-depth correlation analysis, and difficulty in predicting fault trends and solutions in advance. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based power distribution cabinet maintenance system to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: an artificial intelligence-based power distribution cabinet maintenance system, comprising a data acquisition module, a data processing module, an AI intelligent diagnostic model establishment module, a fault precursor monitoring module, a structural health monitoring module, and a semantic strategy alignment monitoring module; The data acquisition module is used to collect and acquire power distribution operation data, including topology data and semantic data. The data processing module is used to extract time-series features from the collected data using normalization and sliding window, and to denoise the data using IQR and DBSCAN; to automatically construct an electrical topology diagram and quantize the node status; and to use a BERT-like model to perform text semantic encoding and alignment to generate a maintenance knowledge graph. The AI ​​intelligent diagnostic model building module is used to construct a multi-branch neural network model by fusing time-series signals, topological structure, and semantic policy features; through joint training with historical data of power distribution operation and maintenance, it has the ability to perceive state and determine risk; extract multimodal key feature tensors to identify fault trends, structural degradation, and policy deviation; and improve the model's ability to provide early warning and policy recommendation in complex environments through secondary reinforcement training. Finally, it is deployed at the operation and maintenance end to realize intelligent diagnosis and auxiliary decision-making. The fault precursor monitoring module is used to perform weighted aggregation calculation on the standardized time series formed by real-time monitoring of the power distribution operation data of the power distribution cabinet, calculate the potential fault risk coefficient FX, and compare it with the first threshold Q1 to determine whether the operation of the power distribution cabinet components is stable. If it is unstable, the structural health monitoring mechanism is activated. The structural health monitoring module is used to construct a topology map based on electrical components, extract deep features by combining the key node state input graph neural network, perform similarity analysis with the reference state vector, calculate and obtain the structural health coefficient JG, and compare it with the second threshold Q2 to determine whether the distribution cabinet topology is healthy. If it is unhealthy, the semantic strategy retrieval mechanism is activated. The semantic policy alignment monitoring module is used to calculate the semantic policy conformity coefficient YC by encoding the state features of the power distribution cabinet into a semantic vector and comparing it with the semantic vector of each historical policy in the knowledge graph through cosine similarity comparison. The coefficient YC is then compared with the third threshold Q3 to determine whether the matching degree between the power distribution cabinet state and the knowledge base policy is qualified, and a policy is then given.

[0006] Preferably, the data acquisition module includes an operational data acquisition unit, a topological data acquisition unit, and a semantic data acquisition unit; The operational data acquisition unit is used to collect real-time power distribution operation data, including: three-phase current value, three-phase voltage value, active power, reactive power, frequency fluctuation, harmonic distortion, waveform abnormality, temperature value inside the distribution cabinet, humidity value inside the distribution cabinet, noise spectrum, low-frequency vibration, and contactor abnormal noise, by installing multi-channel current sensors, voltage sensors, temperature sensors, humidity values ​​inside the distribution cabinet, noise spectrum, low-frequency vibration, and contactor abnormal noise. The topology data acquisition unit is used to acquire topology data in real time by installing infrared thermal imaging sensors, digital switch status sensors, and arc detection sensors at key nodes of buses, circuit breakers, contactors, relays, and transformers. The data includes: switch quantity matrix, electrical connection topology, and node status vector. The semantic data acquisition unit is used to collect semantic data, including maintenance work order content, fault case descriptions, expert knowledge rules, and user terminal fault reports, by integrating maintenance database and user terminal information.

[0007] Preferably, the data processing module is used to standardize the physical quantities of current and voltage by using Min-Max normalization, construct a fixed-dimensional time-series input vector group using sliding time window technology, identify and eliminate sudden noise points caused by physical interference using the IQR method and DBSCAN algorithm; automatically construct a topological adjacency graph of electrical components based on equipment wiring diagrams and SCADA system records, with the state of each component node encoded as a vector; use BERT and RoBERTa language models to embed the semantics of work order text, case text, and expert suggestions into high-dimensional semantic vectors, calculate text alignment through semantic matching algorithms, and construct a maintenance suggestion knowledge graph using semantic graph extraction technology.

[0008] Preferably, the AI ​​intelligent diagnostic model building module is used to construct an initial risk assessment model based on power distribution operation data. It employs a multi-branch neural network structure to fuse power distribution operation sequence signal features, topology graph features, and semantic strategy embedding features to form an AI intelligent diagnostic model for power distribution cabinet fault identification and maintenance decision-making. The model is jointly trained and validated using multiple batches of historical power distribution operation data and maintenance record data to obtain an AI intelligent diagnostic model with basic state perception and risk judgment capabilities. Trend drift features in electrical time-series signals, node anomaly embedding vector change features in the topology graph, and historical strategy matching error patterns in the semantic data are extracted to construct a model for potential fault trend judgment. A key intermediate layer multimodal joint feature tensor for positioning, structural deviation tracking, and strategy conformity analysis is used to identify fault initiation modes, local structural degradation paths, and strategy adaptation deviations. Based on the feature tensor, the AI ​​intelligent diagnostic model is further enhanced through secondary training to improve its early warning identification capability, local structural damage inference capability, and dynamic strategy matching adaptability in complex multi-source data environments. The trained AI intelligent diagnostic model is then deployed at the operation and maintenance end of the power distribution cabinet system to achieve real-time risk scoring of power distribution status, structural health tracking, and intelligent strategy recommendation. This constructs an integrated intelligent decision support system for on-site maintenance assistance and fault early warning response, improving system safety and maintenance efficiency.

[0009] Preferably, the fault precursor monitoring module includes a first calculation unit and a first analysis unit; The first calculation unit is used to perform weighted aggregation calculation on the standardized time series formed by real-time monitoring of the power distribution operation data of the distribution cabinet, and after dimensionless processing, calculate and obtain the potential fault risk coefficient FX.

[0010] Preferably, the first analysis unit is used to pre-set a first threshold Q1 and compare the potential failure risk coefficient FX with the first threshold Q1 to obtain a first evaluation result, including: When the potential fault risk coefficient FX < the first threshold Q1, it indicates that the components of the distribution cabinet are operating stably and there is no potential fault risk, and continuous monitoring is required. When the potential fault risk coefficient FX ≥ the first threshold Q1, it indicates that the components of the distribution cabinet are operating unstablely and there is a potential fault risk, triggering the first early warning instruction and generating the first strategy: start the structural health monitoring mechanism.

[0011] Preferably, the structural health monitoring module includes a second computing unit and a second analysis unit; The second calculation unit is used to construct a topology graph structure by the components of the busbar, circuit breaker, contactor, relay and transformer when the first early warning command is received. It then inputs the real-time status information of the key nodes into the graph neural network (GNN) to obtain the deep feature vector of each node. After similarity comparison with the reference status vector and dimensionless processing, the structural health coefficient JG is calculated and obtained.

[0012] Preferably, the second analysis unit is used to pre-set a second threshold Q2 and compare the structural health coefficient JG with the second threshold Q2 to obtain a second evaluation result, including: When the structural health coefficient JG ≥ the second threshold Q2, it indicates that the topology of the distribution cabinet is healthy and should be continuously monitored. When the structural health coefficient JG < the second threshold Q2, it indicates that the topology of the distribution cabinet is unhealthy, triggering the second early warning instruction and generating the second strategy: activating the semantic strategy retrieval mechanism.

[0013] Preferably, the semantic policy alignment monitoring module includes a third computing unit and a third analysis unit: The third calculation unit is used to encode the status features of the power distribution cabinet into a semantic vector when it receives the second early warning command. With each historical strategy semantic vector in the knowledge graph Cosine similarity comparison is performed, and after dimensionless processing, the semantic strategy conformity coefficient YC is calculated and obtained.

[0014] Preferably, the third analysis unit is used to pre-set a third threshold Q3, and compare the semantic strategy conformity coefficient YC with the third threshold Q3 to obtain a third evaluation result, including: When the semantic policy conformity coefficient YC ≥ the third threshold Q3, it means that the matching degree between the power distribution cabinet status and the knowledge base policy is qualified. The maintenance task is sent directly, the historical policy that is closest to the semantics of the power distribution cabinet status is called, the work plan is generated, and continuous monitoring is performed. When the semantic policy conformity coefficient YC < the third threshold Q3, it indicates that the matching degree between the power distribution cabinet status and the knowledge base policy is not qualified, triggering the third early warning instruction and generating the third policy: a new policy is generated by combining the knowledge graph and expert rules, which is executed after manual review and confirmation, and recorded.

[0015] This invention provides an artificial intelligence-based power distribution cabinet maintenance system. It has the following beneficial effects: (1) The present invention integrates operation data, topology data and semantic data, and utilizes multimodal feature extraction and joint modeling technology to effectively identify potential fault trends and abnormal modes of distribution cabinets, realize early warning monitoring of faults, and significantly improve the early warning response capability of power distribution system.

[0016] (2) This kind of power distribution cabinet maintenance system based on artificial intelligence constructs an electrical component topology diagram and introduces a graph neural network (GNN) to extract deep structural features. It compares the structural health coefficient with the reference state to realize intelligent analysis of the health status of key components such as busbars, circuit breakers, and contactors, thereby improving the accuracy of locating local structural degradation and hidden dangers.

[0017] (3) This kind of power distribution cabinet maintenance system based on artificial intelligence uses BERT-type models to perform high-dimensional semantic encoding of maintenance work orders, cases and expert rules, and constructs a maintenance knowledge graph. After a fault occurs, semantic strategy matching and optimization recommendation can be realized, which significantly improves the matching degree and practicality of strategy push and enhances the automation and intelligence level of fault handling.

[0018] (4) This invention is a power distribution cabinet maintenance system based on artificial intelligence. Through early warning monitoring, health analysis, strategy alignment and a three-level judgment system, it realizes closed-loop intelligent diagnosis from risk assessment, structural monitoring to maintenance decision-making. It can dynamically adapt to the field conditions in a complex multi-source data environment, effectively reduce the failure rate and the frequency of manual intervention, and significantly improve the safety and operation efficiency of the power distribution system. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an artificial intelligence-based power distribution cabinet maintenance system according to the present invention. Detailed Implementation

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Example 1 Please see Figure 1 This invention provides an artificial intelligence-based power distribution cabinet maintenance system, including a data acquisition module, a data processing module, an AI intelligent diagnostic model establishment module, a fault precursor monitoring module, a structural health monitoring module, and a semantic strategy alignment monitoring module; The data acquisition module is used to collect and acquire power distribution operation data, including topology data and semantic data. The data processing module is used to extract time-series features from the collected data using normalization and sliding window, and to denoise the data using IQR and DBSCAN; to automatically construct an electrical topology diagram and quantize the node status; and to use a BERT-like model to perform text semantic encoding and alignment to generate a maintenance knowledge graph. The AI ​​intelligent diagnostic model building module is used to construct a multi-branch neural network model by fusing time-series signals, topological structure, and semantic policy features; through joint training with historical data of power distribution operation and maintenance, it has the ability to perceive state and determine risk; extract multimodal key feature tensors to identify fault trends, structural degradation, and policy deviation; and improve the model's ability to provide early warning and policy recommendation in complex environments through secondary reinforcement training. Finally, it is deployed at the operation and maintenance end to realize intelligent diagnosis and auxiliary decision-making. The fault precursor monitoring module is used to perform weighted aggregation calculation on the standardized time series formed by real-time monitoring of the power distribution operation data of the power distribution cabinet, calculate the potential fault risk coefficient FX, and compare it with the first threshold Q1 to determine whether the operation of the power distribution cabinet components is stable. If it is unstable, the structural health monitoring mechanism is activated. The structural health monitoring module is used to construct a topology map based on electrical components, extract deep features by combining the key node state input graph neural network, perform similarity analysis with the reference state vector, calculate and obtain the structural health coefficient JG, and compare it with the second threshold Q2 to determine whether the distribution cabinet topology is healthy. If it is unhealthy, the semantic strategy retrieval mechanism is activated. The semantic policy alignment monitoring module is used to calculate the semantic policy conformity coefficient YC by encoding the state features of the power distribution cabinet into a semantic vector and comparing it with the semantic vector of each historical policy in the knowledge graph through cosine similarity comparison. The coefficient YC is then compared with the third threshold Q3 to determine whether the matching degree between the power distribution cabinet state and the knowledge base policy is qualified, and a policy is then given.

[0022] In this embodiment, by constructing an intelligent diagnostic system that integrates multi-source data acquisition, graph neural network structure modeling, and semantic strategy alignment analysis, an automatic closed-loop processing flow can be realized during the operation of the power distribution cabinet, from potential fault precursor identification and structural health analysis to maintenance strategy recommendation. This significantly improves the accuracy and response efficiency of fault warnings and reduces the frequency of manual intervention and maintenance costs.

[0023] Example 2 This embodiment is an explanation of Embodiment 1. Specifically, the data acquisition module includes a running data acquisition unit, a topological data acquisition unit, and a semantic data acquisition unit. The operational data acquisition unit is used to collect real-time power distribution operation data, including: three-phase current value, three-phase voltage value, active power, reactive power, frequency fluctuation, harmonic distortion, waveform abnormality, temperature value inside the distribution cabinet, humidity value inside the distribution cabinet, noise spectrum, low-frequency vibration, and contactor abnormal noise, by installing multi-channel current sensors, voltage sensors, temperature sensors, humidity values ​​inside the distribution cabinet, noise spectrum, low-frequency vibration, and contactor abnormal noise. The topology data acquisition unit is used to acquire topology data in real time by installing infrared thermal imaging sensors, digital switch status sensors, and arc detection sensors at key nodes of buses, circuit breakers, contactors, relays, and transformers. The data includes: switch quantity matrix, electrical connection topology, and node status vector. The semantic data acquisition unit is used to collect semantic data, including maintenance work order content, fault case descriptions, expert knowledge rules, and user terminal fault reports, by integrating maintenance database and user terminal information.

[0024] In this embodiment, by setting up three types of acquisition units—operational data, topological data, and semantic data—comprehensive perception and information fusion of the multi-dimensional operating status of the power distribution cabinet can be achieved. This not only improves the accuracy and completeness of fault identification but also provides high-quality, structured data support for subsequent AI diagnostic models, significantly enhancing the system's ability to identify and process fault modes under complex operating conditions.

[0025] Example 3 This embodiment is an explanation of Embodiment 2. Specifically, the data processing module is used to standardize the physical quantities of current and voltage by using Min-Max normalization, construct a fixed-dimensional time-series input vector group using sliding time window technology, identify and eliminate sudden noise points caused by physical interference using the IQR method and DBSCAN algorithm, automatically construct a topological adjacency graph of electrical components based on equipment wiring diagrams and SCADA system records, and encode the state of each component node as a vector; use BERT and RoBERTa language models to embed the semantics of work order text, case text, and expert suggestions into high-dimensional semantic vectors, calculate text alignment through semantic matching algorithms, and construct a maintenance suggestion knowledge graph using semantic graph extraction technology.

[0026] In this embodiment, by introducing Min-Max normalization, sliding time window, IQR and DBSCAN denoising methods into the data processing module, combined with automatic topology graph construction and semantic encoding mechanisms, not only is the temporal consistency, robustness and structural expressiveness of multi-source data effectively improved, but also high-dimensional alignment and graph representation of maintenance semantic information are achieved with the help of BERT-like models, providing clear and accurate multimodal input for subsequent AI models, and significantly enhancing the intelligence level and inference depth of fault diagnosis.

[0027] Example 4 This embodiment is an explanation of Embodiment 3. Specifically, the AI ​​intelligent diagnostic model building module is used to construct an initial risk assessment model based on power distribution operation data. It employs a multi-branch neural network structure to fuse power distribution operation sequence signal features, topology graph features, and semantic policy embedding features to form an AI intelligent diagnostic model for power distribution cabinet fault identification and maintenance decision-making. The model is jointly trained and validated using multiple batches of historical power distribution operation data and maintenance record data to obtain an AI intelligent diagnostic model with basic state perception and risk judgment capabilities. It extracts trend drift features from electrical time-series signals, node anomaly embedding vector change features from the topology graph, and historical policy matching error patterns from the semantic data to construct... A key intermediate layer multimodal joint feature tensor is constructed for potential fault trend determination, structural deviation tracking, and strategy conformity analysis. This tensor is used to identify fault initiation modes, local structural degradation paths, and strategy adaptation deviations. Based on the feature tensor, the AI ​​intelligent diagnostic model is further enhanced through secondary training to improve its early warning identification capability, local structural damage inference capability, and dynamic strategy matching adaptability in complex multi-source data environments. The trained AI intelligent diagnostic model is then deployed at the operation and maintenance end of the power distribution cabinet system to achieve real-time risk scoring of power distribution status, structural health tracking, and intelligent strategy recommendation. This constructs an integrated intelligent decision support system for on-site maintenance assistance and fault early warning response, improving system safety and maintenance efficiency.

[0028] In this embodiment, by introducing a multi-branch neural network architecture into the AI ​​intelligent diagnostic model building module, multi-modal features of time-series signals, topology, and semantic strategies are integrated. Combined with the construction of key intermediate layer feature tensors and a secondary reinforcement training mechanism, the model has stronger fault trend recognition, structural deviation tracking, and semantic strategy deviation discrimination capabilities. This significantly improves the perception accuracy of early fault precursors and the intelligent recommendation level of maintenance strategies under complex operating conditions, thereby achieving highly reliable operation and efficient maintenance decision support for the power distribution system.

[0029] Example 5 This embodiment is an explanation of embodiment 4. Specifically, the fault precursor monitoring module includes a first calculation unit and a first analysis unit. The first calculation unit is used to perform weighted aggregation calculations on a standardized time series formed by real-time monitoring of the power distribution operation data of the distribution cabinet. After dimensionless processing, the potential fault risk coefficient FX is calculated and obtained, as shown in the following formula: ; In the formula, Indicates the length of the time window. This represents the normalized time-series input at time i. Represented as the Sigmoid function, Represents the weight matrix. This indicates the bias top scalar. Represents the weighting function; ; In the formula, This represents the three-phase current value at time i. This represents the three-phase voltage value at time i. This represents the temperature value at time i. This represents the humidity value at time i. This represents the total harmonic distortion rate at time i. This represents the noise value at time i. ; In the formula, e represents the base of the natural constant. This indicates the adjustment weight offset coefficient.

[0030] In this embodiment, the first calculation unit in the fault precursor monitoring module introduces a weighted aggregation calculation mechanism of standardized time series, combined with the Sigmoid function to achieve dimensionless mapping and nonlinear enhancement processing, accurately extracting the subtle abnormal change trends of various operating parameters, including current, voltage, temperature, humidity, harmonic distortion, and noise in the time dimension, thereby effectively calculating the potential fault risk coefficient FX, improving the early identification capability of potential fault initiation stages of the distribution cabinet, and providing a more timely and accurate decision-making basis for subsequent structural health assessment and strategic intervention.

[0031] Example 6 This embodiment is an explanation of embodiment 5. Specifically, the first analysis unit is used to preset a first threshold Q1 and compare the potential failure risk coefficient FX with the first threshold Q1 to obtain a first evaluation result, including: When the potential fault risk coefficient FX < the first threshold Q1, it indicates that the components of the distribution cabinet are operating stably and there is no potential fault risk, and continuous monitoring is required. When the potential fault risk coefficient FX ≥ the first threshold Q1, it indicates that the components of the distribution cabinet are operating unstablely and there is a potential fault risk, triggering the first early warning instruction and generating the first strategy: start the structural health monitoring mechanism.

[0032] In this embodiment, a first threshold Q1 is set by the first analysis unit in the fault precursor monitoring module, and the potential fault risk coefficient FX is dynamically compared and analyzed with this threshold. This realizes an intelligent judgment and hierarchical response mechanism for the operating status of the distribution cabinet components. When FX exceeds the threshold, the system can automatically determine that there is a potential fault risk and immediately trigger the structural health monitoring mechanism to intervene in advance and prevent the fault from escalating. When FX is below the threshold, continuous monitoring is performed to prevent false alarms. This design significantly improves the system's response accuracy and early warning capability, effectively enhancing the stability and safety of the power distribution system. Specific implementation data are shown in the table below:

[0033] Example 7 This embodiment is an explanation of Embodiment 6. Specifically, the structural health monitoring module includes a second calculation unit and a second analysis unit. The second calculation unit, upon receiving the first early warning command, constructs a topology graph structure from the components of the busbar, circuit breaker, contactor, relay, and transformer. It then inputs the real-time status information of key nodes into a graph neural network (GNN) to obtain the deep feature vector of each node. This feature vector is then compared with a reference status vector, and after dimensionless processing, the structural health coefficient JG is calculated using the following formula: ; In the formula, This indicates the total number of nodes in the power distribution cabinet structure diagram. This represents the embedding vector of node v after aggregation in the k-th layer of the GNN. Represents the normal reference state vector. This represents a constant used to prevent the denominator from being zero.

[0034] In this embodiment, the core electrical components of the distribution cabinet are constructed into a topology graph through the second computing unit in the structural health monitoring module. A graph neural network (GNN) is then used to extract deep state features of each node, and similarity analysis is performed with the normal reference state vector to calculate the structural health coefficient JG. This achieves a multi-dimensional, graph-structured intelligent assessment of the internal structural operating status of the distribution cabinet. This method not only overcomes the limitations of traditional monitoring methods in identifying complex electrical connections but also accurately captures local node anomalies or structural degradation trends, effectively improving the accuracy of fault location and the sensitivity of structural risk identification. This significantly enhances the overall health monitoring capability and operational decision-making efficiency of the power distribution system.

[0035] Example 8 This embodiment is an explanation of embodiment 7. Specifically, the second analysis unit is used to preset a second threshold Q2 and compare the structural health coefficient JG with the second threshold Q2 to obtain a second evaluation result, including: When the structural health coefficient JG ≥ the second threshold Q2, it indicates that the topology of the distribution cabinet is healthy and should be continuously monitored. When the structural health coefficient JG < the second threshold Q2, it indicates that the topology of the distribution cabinet is unhealthy, triggering the second early warning instruction and generating the second strategy: activating the semantic strategy retrieval mechanism.

[0036] In this embodiment, the structural health threshold Q2 is set by the second analysis unit in the structural health monitoring module, and the calculated structural health coefficient JG is dynamically compared, realizing a quantitative assessment and early warning triggering mechanism for the health status of the distribution cabinet topology. When the structural index is lower than the safety threshold, the system can automatically identify structural anomaly risks, promptly trigger the second early warning command, and initiate the semantic strategy retrieval mechanism, thereby effectively avoiding major electrical faults caused by undetected structural degradation, improving the stability, safety, and proactive response of the electrical system. Specific implementation data are shown in the table below:

[0037] Example 9 This embodiment is an explanation of embodiment 8. Specifically, the semantic strategy alignment monitoring module includes a third calculation unit and a third analysis unit: The third calculation unit is used to encode the status features of the power distribution cabinet into a semantic vector when it receives the second early warning command. With each historical strategy semantic vector in the knowledge graph After performing cosine similarity comparison and dimensionless processing, the semantic strategy consistency coefficient YC is calculated as follows: ; In the formula, Semantic vector representation of the status of the power distribution cabinet. Let j represent the semantic representation of the j-th maintenance strategy in the knowledge base. Indicates the weighting factor for historical time difference; ; In the formula, This represents the present time of the j-th strategy.

[0038] In this embodiment, a semantic vector representation and cosine similarity comparison mechanism are introduced through the third computing unit in the semantic policy alignment monitoring module. Combined with a historical time difference weighting factor, this effectively achieves deep semantic alignment and matching measurement between the current state of the distribution cabinet and the maintenance strategy in the knowledge graph. This design can dynamically identify the historical strategy that is closest to the current operating state, ensuring that the recommended strategy has high semantic relevance and timeliness, thereby improving the accuracy of fault response and the pertinence of decision recommendations. It avoids misjudgment or improper handling caused by strategy aging or semantic deviation, and enhances the system's intelligent response capability and evolutionary adaptation capability.

[0039] Example 10 This embodiment is an explanation of embodiment 9. Specifically, the third analysis unit is used to preset a third threshold Q3 in advance, and compares and analyzes the semantic strategy conformity coefficient YC with the third threshold Q3 to obtain the third evaluation result, including: When the semantic policy conformity coefficient YC ≥ the third threshold Q3, it means that the matching degree between the power distribution cabinet status and the knowledge base policy is qualified. The maintenance task is sent directly, the historical policy that is closest to the semantics of the power distribution cabinet status is called, the work plan is generated, and continuous monitoring is performed. When the semantic policy conformity coefficient YC < the third threshold Q3, it indicates that the matching degree between the power distribution cabinet status and the knowledge base policy is not qualified, triggering the third early warning instruction and generating the third policy: a new policy is generated by combining the knowledge graph and expert rules, which is executed after manual review and confirmation, and recorded.

[0040] In this embodiment, a semantic policy conformity evaluation mechanism is set up through the third analysis unit in the semantic policy alignment monitoring module. Based on a threshold Q3, the matching results are classified and judged, enabling intelligent judgment and graded response to the maintenance policy matching degree. When the semantic matching degree is high, the system can automatically call the historical best policy and generate a work plan, achieving automated task scheduling and rapid handling. When the matching degree is insufficient, an intelligent policy generation mechanism is triggered, combining knowledge graphs and expert rules to adaptively construct new policies and introduce manual review, ensuring a safe and reliable processing flow even in complex or novel fault scenarios. This design significantly improves the system's semantic adaptability and policy response flexibility, helping to reduce the misjudgment rate, enhance intelligent decision-making capabilities, and improve the efficiency of the operation and maintenance closed loop. Specific implementation data is shown in the table below:

[0041] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0042] The above formulas are all derived from software simulation using a large amount of data, and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based power distribution cabinet maintenance system, characterized in that, It includes a data acquisition module, a data processing module, an AI intelligent diagnostic model building module, a fault precursor monitoring module, a structural health monitoring module, and a semantic strategy alignment monitoring module; The data acquisition module is used to collect and acquire power distribution operation data, including topology data and semantic data. The data processing module is used to extract time-series features from the collected data using normalization and sliding window, and to denoise the data using IQR and DBSCAN; to automatically construct an electrical topology diagram and quantize the node status; and to use a BERT-like model to perform text semantic encoding and alignment to generate a maintenance knowledge graph. The AI ​​intelligent diagnostic model building module is used to construct a multi-branch neural network model by fusing time-series signals, topological structure and semantic strategy features; through joint training with historical data of power distribution operation and maintenance, it has the ability to perceive state and determine risk; and extracts multimodal key feature tensors to identify fault trends, structural degradation and strategy deviation. The model is enhanced through secondary reinforcement training to improve its early warning and strategy recommendation capabilities in complex environments, and is finally deployed on the operation and maintenance end to achieve intelligent diagnosis and auxiliary decision-making. The fault precursor monitoring module is used to perform weighted aggregation calculation on the standardized time series formed by real-time monitoring of the power distribution operation data of the power distribution cabinet, calculate the potential fault risk coefficient FX, and compare it with the first threshold Q1 to determine whether the operation of the power distribution cabinet components is stable. If it is unstable, the structural health monitoring mechanism is activated. The structural health monitoring module is used to construct a topology map based on electrical components, extract deep features by combining the key node state input graph neural network, perform similarity analysis with the reference state vector, calculate and obtain the structural health coefficient JG, and compare it with the second threshold Q2 to determine whether the distribution cabinet topology is healthy. If it is unhealthy, the semantic strategy retrieval mechanism is activated. The semantic policy alignment monitoring module is used to calculate the semantic policy conformity coefficient YC by encoding the state features of the power distribution cabinet into a semantic vector and comparing it with the semantic vector of each historical policy in the knowledge graph through cosine similarity comparison. The coefficient YC is then compared with the third threshold Q3 to determine whether the matching degree between the power distribution cabinet state and the knowledge base policy is qualified, and a policy is then given.

2. The power distribution cabinet maintenance system based on artificial intelligence according to claim 1, characterized in that, The data acquisition module includes a runtime data acquisition unit, a topological data acquisition unit, and a semantic data acquisition unit; The operational data acquisition unit is used to collect real-time power distribution operation data, including: three-phase current value, three-phase voltage value, active power, reactive power, frequency fluctuation, harmonic distortion, waveform abnormality, temperature value inside the distribution cabinet, humidity value inside the distribution cabinet, noise spectrum, low-frequency vibration, and contactor abnormal noise, by installing multi-channel current sensors, voltage sensors, temperature sensors, humidity values ​​inside the distribution cabinet, noise spectrum, low-frequency vibration, and contactor abnormal noise. The topology data acquisition unit is used to acquire topology data in real time by installing infrared thermal imaging sensors, digital switch status sensors, and arc detection sensors at key nodes of buses, circuit breakers, contactors, relays, and transformers. The data includes: switch quantity matrix, electrical connection topology, and node status vector. The semantic data acquisition unit is used to collect semantic data, including maintenance work order content, fault case descriptions, expert knowledge rules, and user terminal fault reports, by integrating maintenance database and user terminal information.

3. The power distribution cabinet maintenance system based on artificial intelligence according to claim 2, characterized in that, The data processing module is used to standardize the physical quantities of current and voltage using Min-Max normalization, construct a fixed-dimensional time-series input vector group using sliding time window technology, identify and eliminate sudden noise points caused by physical interference using the IQR method and DBSCAN algorithm, automatically construct a topological adjacency graph of electrical components based on equipment wiring diagrams and SCADA system records, and encode the state of each component node as a vector; use BERT and RoBERTa language models to embed the semantics of work order text, case text, and expert suggestions into high-dimensional semantic vectors, calculate text alignment through semantic matching algorithms, and construct a maintenance suggestion knowledge graph using semantic graph extraction technology.

4. The power distribution cabinet maintenance system based on artificial intelligence according to claim 3, characterized in that, The AI ​​intelligent diagnostic model building module is used to construct an initial risk assessment model based on power distribution operation data. It employs a multi-branch neural network structure to fuse power distribution operation sequence signal features, topology graph features, and semantic policy embedding features to form an AI intelligent diagnostic model for distribution cabinet fault identification and maintenance decision-making. The model is jointly trained and validated using multiple batches of historical power distribution operation data and maintenance record data to obtain an AI intelligent diagnostic model with basic state perception and risk judgment capabilities. It extracts trend drift features from electrical time-series signals, node anomaly embedding vector change features from the topology graph, and historical policy matching error patterns from semantic data to construct a system for potential fault trend determination. The key intermediate layer of the structural deviation tracking and strategy conformity analysis uses a multimodal joint feature tensor to identify fault initiation modes, local structural degradation paths, and strategy adaptation deviations. Based on this feature tensor, the AI ​​intelligent diagnostic model undergoes secondary reinforcement training to improve its early warning recognition capabilities, local structural damage inference capabilities, and dynamic strategy matching adaptability in complex multi-source data environments. The trained AI intelligent diagnostic model is then deployed at the distribution cabinet system's operation and maintenance end to achieve real-time risk scoring of power distribution status, structural health tracking, and intelligent strategy recommendation. This constructs an integrated intelligent decision support system for on-site maintenance assistance and fault early warning response, improving system safety and maintenance efficiency.

5. The power distribution cabinet maintenance system based on artificial intelligence according to claim 4, characterized in that, The fault precursor monitoring module includes a first calculation unit and a first analysis unit; The first calculation unit is used to perform weighted aggregation calculation on the standardized time series formed by real-time monitoring of the power distribution operation data of the power distribution cabinet, and after dimensionless processing, calculate and obtain the potential fault risk coefficient FX.

6. The power distribution cabinet maintenance system based on artificial intelligence according to claim 5, characterized in that, The first analysis unit is used to pre-set a first threshold Q1 and compare the potential failure risk coefficient FX with the first threshold Q1 to obtain a first evaluation result, including: When the potential fault risk coefficient FX < the first threshold Q1, it indicates that the components of the distribution cabinet are operating stably and there is no potential fault risk, and continuous monitoring is required. When the potential fault risk coefficient FX ≥ the first threshold Q1, it indicates that the components of the distribution cabinet are operating unstablely and there is a potential fault risk, triggering the first early warning instruction and generating the first strategy: start the structural health monitoring mechanism.

7. The power distribution cabinet maintenance system based on artificial intelligence according to claim 6, characterized in that, The structural health monitoring module includes a second computing unit and a second analysis unit; The second calculation unit is used to construct a topology graph structure by the components of the busbar, circuit breaker, contactor, relay and transformer when the first early warning command is received. It then inputs the real-time status information of the key nodes into the graph neural network (GNN) to obtain the deep feature vector of each node. After similarity comparison with the reference status vector and dimensionless processing, the structural health coefficient JG is calculated and obtained.

8. The power distribution cabinet maintenance system based on artificial intelligence according to claim 7, characterized in that, The second analysis unit is used to pre-set a second threshold Q2 and compare the structural health coefficient JG with the second threshold Q2 to obtain a second evaluation result, including: When the structural health coefficient JG ≥ the second threshold Q2, it indicates that the topology of the distribution cabinet is healthy and should be continuously monitored. When the structural health coefficient JG < the second threshold Q2, it indicates that the topology of the distribution cabinet is unhealthy, triggering the second early warning instruction and generating the second strategy: activating the semantic strategy retrieval mechanism.

9. The power distribution cabinet maintenance system based on artificial intelligence according to claim 8, characterized in that, The semantic strategy alignment monitoring module includes a third computing unit and a third analysis unit: The third calculation unit is used to encode the status features of the power distribution cabinet into a semantic vector when it receives the second early warning command. With each historical strategy semantic vector in the knowledge graph Cosine similarity comparison is performed, and after dimensionless processing, the semantic strategy conformity coefficient YC is calculated and obtained.

10. The power distribution cabinet maintenance system based on artificial intelligence according to claim 9, characterized in that, The third analysis unit is used to pre-set a third threshold Q3, and compare the semantic strategy conformity coefficient YC with the third threshold Q3 to obtain the third evaluation result, including: When the semantic policy conformity coefficient YC ≥ the third threshold Q3, it means that the matching degree between the power distribution cabinet status and the knowledge base policy is qualified. The maintenance task is sent directly, the historical policy that is closest to the semantics of the power distribution cabinet status is called, the work plan is generated, and continuous monitoring is performed. When the semantic policy conformity coefficient YC < the third threshold Q3, it indicates that the matching degree between the power distribution cabinet status and the knowledge base policy is not qualified, triggering the third early warning instruction and generating the third policy: a new policy is generated by combining the knowledge graph and expert rules, which is executed after manual review and confirmation, and recorded.

Images