An agent-based power distribution system health monitoring and prediction method

Abstract

The application discloses a power distribution system health monitoring and prediction method based on an intelligent agent, relates to the intelligent monitoring and diagnosis field of power equipment, and specifically comprises the following steps: establishing a cross-level intelligent agent cooperation mechanism, identifying a device fault source of a power distribution system through autonomous judgment of a single-node intelligent agent, formulating a strategy instruction of dispatching and maintenance according to global information, feeding back the strategy instruction to global nodes of the power distribution system through cooperation between levels, predicting a fault risk by using a system dispatching intelligent agent based on feedback results of the strategy instruction of dispatching and maintenance, providing an early warning level based on a prediction result, optimizing resource dispatching and maintenance configuration through an intelligent dispatching algorithm, and self-learning based on a monitoring deviation between an actual situation and the prediction result. The application realizes real-time identification of faults, formulation of strategies, prediction of risks, optimization of dispatching, and autonomous learning through cross-level intelligent agent cooperation, and improves the intelligent monitoring and operation and maintenance capability of the power distribution system.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and diagnosis of power equipment, and specifically to a method for health monitoring and prediction of power distribution systems based on intelligent agents. Background Technology

[0002] Currently, fault monitoring and prediction in power distribution systems mostly employ methods based on single sensors or static models. These methods collect real-time equipment operating status data and use threshold detection or simple data analysis techniques to identify faults. However, they typically rely on manually set rules or empirical parameters, lacking effective adaptability to dynamic system changes. Furthermore, while some methods incorporate edge computing and cloud platform architectures to alleviate data transmission latency, data transmission bottlenecks still exist, limiting the real-time nature of fault response. Traditional monitoring methods primarily focus on the local state of equipment, lacking the fusion of multi-level, multi-source information, making it difficult to quickly and accurately locate the root cause of faults. They also lack system-level causal relationship analysis and risk path identification capabilities, resulting in limited advance warning and accuracy.

[0003] Existing technologies suffer from the following shortcomings: They lack the integration of multi-level, multi-source information, resulting in insufficient comprehensiveness and accuracy in monitoring, making it difficult to quickly locate and analyze complex faults. They lack the ability to proactively adapt to dynamic changes in the power distribution system, hindering early warning and leading to slow fault response times and delays in intervention. Traditional technologies rely heavily on cloud processing for data transmission, resulting in data transmission delays and network dependencies, reducing the real-time performance of fault detection. Furthermore, they struggle to accurately identify the root causes of faults and potential risks, impacting the safe and stable operation of the overall system.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for health monitoring and prediction of power distribution systems based on intelligent agents, so as to solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for health monitoring and prediction of a power distribution system based on intelligent agents, specifically comprising:

[0007] Establish a cross-level intelligent agent collaboration mechanism, and identify the equipment fault source of the power distribution system through the autonomous judgment of a single node intelligent agent, and package and report the node status and fault information to the macro level.

[0008] Based on global information, formulate scheduling and maintenance strategies and instructions, and feed these strategies and instructions back to the global nodes of the power distribution system through inter-level collaboration.

[0009] Based on the feedback results of the execution scheduling and maintenance strategy instructions, the system scheduling agent is used to predict the risk of failure.

[0010] Based on the prediction results, early warning levels are provided, and resource scheduling and maintenance configurations are optimized through intelligent scheduling algorithms;

[0011] It learns from the monitoring deviation between the actual situation and the prediction results.

[0012] As a preferred embodiment of the agent-based power distribution system health monitoring and prediction method described in this invention, wherein:

[0013] The cross-level intelligent agent collaboration mechanism specifically includes,

[0014] The system is divided into micro-level and macro-level. The micro-level includes single-node intelligent agents that monitor the node status of the power distribution system in real time, while the macro-level includes system scheduling intelligent agents that analyze the overall power distribution network status in real time.

[0015] The method of identifying power distribution system equipment faults through the autonomous judgment of a single-node intelligent agent specifically includes:

[0016] By dynamically registering and connecting different device sensors through a plug-in API architecture, and using an event-driven model to support asynchronous mechanisms to collect and preprocess the operating status data of devices in the power distribution system in real time;

[0017] By combining the historical operating status data of equipment in the power distribution system with a self-correcting feature algorithm, time-frequency features are dynamically extracted to reflect the fault information of the current power distribution system.

[0018] By embedding single-node intelligent agents into local inference, reliance on the cloud is avoided, ensuring rapid response. Based on the equipment relationship diagram of the power distribution system, the root cause of the fault is analyzed through causal relationship inference algorithm.

[0019] As a preferred embodiment of the agent-based power distribution system health monitoring and prediction method described in this invention, wherein:

[0020] The preprocessing of the operating status data of equipment in the power distribution system specifically includes,

[0021] Based on the adaptive connection mechanism of edge computing nodes, the acquisition frequency and acquisition type of equipment sensors are automatically adjusted according to the data changes of the power distribution system, and data denoising is performed using sliding window and Kalman filtering.

[0022] The specific formula for the self-correcting feature algorithm is as follows:

[0023]

[0024] in, This indicates the time-frequency features extracted after self-correction. This represents the time-frequency features extracted after correction from the previous time point. This indicates the time-frequency features extracted from the current running status data. This represents historical operational status data. The self-correcting learning rate, representing the degree of control feature fusion, is dynamically adjusted to adapt to device changes. This represents the fault penalty coefficient, used to adjust the degree of influence of historical operating status data. The larger the deviation, the higher the corresponding β(t), which increases the dependence on historical operating status data.

[0025] As a preferred embodiment of the agent-based power distribution system health monitoring and prediction method described in this invention, wherein:

[0026] The equipment relationship diagram based on the power distribution system analyzes the root cause of faults using a causal relationship inference algorithm, specifically including:

[0027] The device sensors are treated as nodes in the causal relationship, and the impact of the fault is treated as an edge in the causal relationship.

[0028] We introduce particles to represent the existence of each causal relationship edge;

[0029] A custom prior distribution is defined based on historical operating status data, and particle weights are updated according to real-time collected operating status data of equipment in the power distribution system.

[0030] Resampling is applied, and the hierarchical structure of the equipment relationship graph of the power distribution system is encoded, including the bottom layer as the state of single-node equipment, the middle layer as the equipment influence path, and the top layer as the root cause of power distribution system faults. The hierarchical structure of the relationship graph is used to constrain the particle state.

[0031] By recursively estimating the path strength information of each fault's impact at different time points through virtual intervention, and combining this with changes in the operating status of the power distribution system's equipment, the reliability of the causal relationship between the equipment in the power distribution system is dynamically adjusted.

[0032] The specific formula for recursively estimating the path strength information of each fault's impact at different time points is expressed as follows:

[0033]

[0034] in, Indicates path strength information, This indicates the total number of paths affected by the fault. Indicates particle index, This represents the influence weight of the k-th particle at time point t after virtual intervention. This represents the effect of virtual intervention on the k-th particle at time point t;

[0035] Impact scores are calculated for particle sets along all fault-affected paths, and potential fault sources in the power distribution system are identified by ranking them by impact.

[0036] As a preferred embodiment of the agent-based power distribution system health monitoring and prediction method described in this invention, wherein:

[0037] The process of formulating scheduling and maintenance strategy instructions based on global information, and feeding these instructions back to the global nodes of the power distribution system through inter-level collaboration, specifically includes:

[0038] Reinforcement learning is used to derive the optimal scheduling instruction based on the paths affected by all failures;

[0039] The global node decomposes the scheduling policy instructions into sub-policies and transmits them to potential fault sources of equipment in the power distribution system through hierarchical relationships;

[0040] The hierarchical structure allows for upward feedback, reporting the strategy execution effectiveness, actual deviations, and potential risks to the global node.

[0041] Set up a closed-loop collaboration between the global, regional, and device levels to automatically coordinate the execution of scheduling and maintenance strategy instructions by the neighboring single-node intelligent agents of the currently faulty node.

[0042] The robustness of the AI-based scenario-based virtual verification strategy under different failure conditions was demonstrated.

[0043] As a preferred embodiment of the agent-based power distribution system health monitoring and prediction method described in this invention, wherein:

[0044] The feedback results of the strategy instructions based on execution scheduling and maintenance are used by the system scheduling agent to predict fault risks, specifically including:

[0045] Multi-scale time-frequency decomposition is used to extract multi-level features from the feedback data;

[0046] The graph neural network model is used to perform causal learning on the equipment relationship diagram of the power distribution system. The feedback results are used as reward signals to input into the reinforcement learning model to dynamically learn the causal relationship between maintenance effect and failure risk.

[0047] Automatically generate a fault risk warning index based on time, frequency, and causal structure;

[0048] The specific formula for generating the fault risk warning index is as follows:

[0049]

[0050] in, This indicates a fault risk warning index. The activation function represents the risk level determination. This represents the dynamic adjustment weight of the i-th multi-level feature at time t. This represents the value of the i-th multi-scale feature at time t.

[0051] On the other hand, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements the steps of an agent-based power distribution system health monitoring and prediction method as described above.

[0052] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the steps of an agent-based power distribution system health monitoring and prediction method as described above.

[0053] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0054] By constructing a cross-level intelligent agent collaboration mechanism, single-node intelligent agents at the micro-level can collect equipment status in real time under event-driven and API interface support. Combined with a self-correcting feature algorithm, time-frequency features are dynamically extracted, enabling rapid and accurate identification of potential equipment faults. Furthermore, causal relationships are used to infer and analyze the root causes of faults, significantly improving the sensitivity of fault detection and the accuracy of fault location. At the macro-level, the system scheduling intelligent agent, based on global information, automatically derives the optimal scheduling strategy using reinforcement learning. Combined with multi-scale time-frequency features, risk assessment is performed, and maintenance plans are dynamically adjusted to form closed-loop management, effectively improving scheduling response speed and maintenance efficiency. Through the fusion of multi-source heterogeneous data and virtual intervention technology, equipment status is combined with causal paths to recursively estimate fault path strength, achieving accurate location of potential fault sources. The autonomous learning mechanism continuously optimizes model parameters and causal relationships based on feedback deviations, improving overall predictive capability and robustness. This significantly improves the monitoring, prediction, and scheduling levels of the power distribution system, achieving the goal of safe, stable, and intelligent power distribution network operation. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0056] Figure 1 This is a flowchart of a method for monitoring and predicting the health of a power distribution system based on intelligent agents, according to the present invention. Detailed Implementation

[0057] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0058] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for health monitoring and prediction of a power distribution system based on intelligent agents, specifically including:

[0059] S1. Establish a cross-level intelligent agent collaboration mechanism. Through the autonomous judgment of a single node intelligent agent, identify the equipment fault source of the power distribution system and package and report the node status and fault information to the macro level.

[0060] The cross-level intelligent agent collaboration mechanism specifically includes,

[0061] The system is divided into micro-level and macro-level. The micro-level includes single-node intelligent agents that monitor the node status of the power distribution system in real time, while the macro-level includes system scheduling intelligent agents that analyze the overall power distribution network status in real time.

[0062] It should be noted that the equipment sensors of the power distribution network system are defined as nodes. The built-in single-node intelligent agent dynamically registers and connects to the sensors (temperature, current, voltage, vibration, sound waves, etc.) of different devices (transformers, switches, circuit breakers, cables, etc.) through a plug-in API architecture. The event-driven model supports the asynchronous mechanism to collect the operating status data of the equipment in the power distribution system in real time and perform preprocessing.

[0063] By combining the historical operating status data of equipment in the power distribution system with a self-correcting feature algorithm, time-frequency features are dynamically extracted to reflect the fault information of the current power distribution system.

[0064] By embedding single-node intelligent agents into local inference, reliance on the cloud is avoided, ensuring rapid response. Based on the equipment relationship diagram of the power distribution system, the root cause of the fault is analyzed through causal relationship inference algorithm.

[0065] The preprocessing of the operating status data of equipment in the power distribution system specifically includes,

[0066] Based on the adaptive connection mechanism of edge computing nodes, the acquisition frequency and acquisition type of equipment sensors are automatically adjusted according to the data changes of the power distribution system, and data denoising is performed using sliding window and Kalman filtering.

[0067] The specific formula for the self-correcting feature algorithm is as follows:

[0068]

[0069] in, This indicates the time-frequency features extracted after self-correction. This represents the time-frequency features extracted after correction from the previous time point. This indicates the time-frequency features extracted from the current running status data. This represents historical operational status data. The self-correcting learning rate, representing the degree of control feature fusion, is dynamically adjusted to adapt to device changes. This represents the fault penalty coefficient, used to adjust the degree of influence of historical operating status data. The larger the deviation, the higher the corresponding β(t), which increases the dependence on historical operating status data.

[0070] The equipment relationship diagram based on the power distribution system analyzes the root cause of faults using a causal relationship inference algorithm, specifically including:

[0071] The device sensors are treated as nodes in the causal relationship, and the impact of the fault is treated as an edge in the causal relationship.

[0072] We introduce particles to represent the existence of each causal relationship edge;

[0073] A custom prior distribution is defined based on historical operating status data, and particle weights are updated according to real-time collected operating status data of equipment in the power distribution system.

[0074] It should be noted that the historical operating status data was cleaned, noise was reduced, and mean features were extracted.

[0075] It should be noted that the probability density function is estimated nonparametrically from historical operating state data, and the particle weights are adjusted using a Bayesian update mechanism based on the current operating state data.

[0076] Extract the particle set of the custom prior distribution and assign initial weights, define the state observation of the current single-node agent, correct the original historical prior distribution through Bayesian recursive state, and dynamically update the prior parameters according to the state mean characteristics of the particle set.

[0077] Resampling is applied, and the hierarchical structure of the equipment relationship graph of the power distribution system is encoded, including the bottom layer as the state of single-node equipment, the middle layer as the equipment influence path, and the top layer as the root cause of power distribution system faults. The hierarchical structure of the relationship graph is used to constrain the particle state.

[0078] It should be noted that the device impact path refers to the sequence of power transmission lines in the power distribution system that start from a certain fault source (a node in the causal relationship), pass through a series of intermediate nodes, and reach the point of impact.

[0079] By recursively estimating the path strength information of each fault's impact at different time points through virtual intervention, and combining this with changes in the operating status of the power distribution system's equipment, the reliability of the causal relationship between the equipment in the power distribution system is dynamically adjusted.

[0080] The specific formula for recursively estimating the path strength information of each fault's impact at different time points is expressed as follows:

[0081]

[0082] in, Indicates path strength information, This indicates the total number of paths affected by the fault. Indicates particle index, This represents the influence weight of the k-th particle at time point t after virtual intervention. This represents the effect of virtual intervention on the k-th particle at time point t;

[0083] Impact scores are calculated for particle sets along all fault-affected paths, and potential fault sources in the power distribution system are identified by ranking them by impact.

[0084] It should be noted that each path affected by a fault is defined to correspond to a potentially faulty device, forming a path-device mapping table;

[0085] Based on the path impact estimate of the particle set, measure the degree of impact of each path on the system in the current state;

[0086] The impact level is assigned to the corresponding device, the impact of all paths passing through that device is accumulated, and the devices ranked at the top are identified as potential sources of failure after being sorted by magnitude.

[0087] S2. Formulate scheduling and maintenance strategy instructions based on global information, and feed the strategy instructions back to the global nodes of the power distribution system through inter-level collaboration;

[0088] It should be noted that reinforcement learning (such as deep Q-networks) is used to derive the optimal scheduling instructions based on the paths affected by all faults;

[0089] It should be noted that by using reinforcement learning for single-node agents and system scheduling agents, the optimal scheduling decision (such as adjusting equipment switching status, load adjustment, voltage adjustment, etc.) is output based on the current state and the path affected by the fault, in order to minimize the fault risk and improve system stability.

[0090] The global node decomposes the scheduling policy instructions into sub-policies and transmits them to potential fault sources of equipment in the power distribution system through hierarchical relationships;

[0091] It should be noted that in the power distribution system, after the global node formulates the overall scheduling strategy, it decomposes it into multiple sub-strategies, which are then communicated and distributed to the potential fault sources of the equipment in the power distribution system through hierarchical relationships, so as to achieve coordinated scheduling and fault control.

[0092] The hierarchical relationship is a hierarchical control architecture used to transmit complex overall scheduling strategies in a hierarchical and fine-grained manner.

[0093] The decomposition sub-strategy maps the global strategy to sub-strategies for the sensor sub-regions of the power distribution system equipment according to the topology of the power distribution system. For each potential fault source equipment, a dedicated scheduling instruction is generated and decomposed using rule matching.

[0094] The hierarchical structure allows for upward feedback, reporting the strategy execution effectiveness, actual deviations, and potential risks to the global node.

[0095] Set up a closed-loop collaboration between the global, regional, and device levels to automatically coordinate the execution of scheduling and maintenance strategy instructions by the neighboring single-node intelligent agents of the currently faulty node.

[0096] The robustness of the AI-based scenario-based virtual verification strategy under different fault conditions was demonstrated.

[0097] It should be noted that a virtual environment containing the power distribution system is established through power simulation software, and the real system parameters, equipment sensors and fault information are simulated.

[0098] Load the scheduling policy to be verified into the virtual environment;

[0099] By simulating different fault source scenarios in turn through an automated framework, the system outputs policy instructions, operating status data and fault handling time, and evaluates system stability, recovery speed, fault impact range and success rate based on the deviation of the power distribution system.

[0100] S3. Based on the feedback results of the execution scheduling and maintenance strategy instructions, the system scheduling agent is used to predict the risk of failure.

[0101] Multi-scale time-frequency decomposition (such as multi-level wavelet transform or multi-scale short-time Fourier transform) is used to extract multi-level features from the feedback data.

[0102] The graph neural network model is used to perform causal learning on the equipment relationship diagram of the power distribution system. The feedback results are used as reward signals to input into the reinforcement learning model to dynamically learn the causal relationship between maintenance effect and failure risk.

[0103] It should be noted that the training graph neural model performs causal learning on the equipment relationship graph of the power distribution system and initializes the reinforcement learning policy network;

[0104] In each decision cycle, the graph neural network model is used to dynamically update the causal relationship parameters between devices based on the latest communication relationships, forming the current state.

[0105] It should be noted that the reward signal is designed as follows:

[0106] 1. Rewards increase when maintenance operations successfully reduce the probability of failure or shorten the duration of failure;

[0107] 2. If maintenance fails to improve the situation or even causes secondary failures, the reward will be reduced;

[0108] Integrate reward signals into reinforcement learning models;

[0109] By iteratively repeating the above steps, the causal understanding of the maintenance schedule and the robustness of the strategy are gradually improved.

[0110] Automatically generate a fault risk warning index based on time, frequency, and causal structure;

[0111] The specific formula for generating the fault risk warning index is as follows:

[0112]

[0113] in, This indicates a fault risk warning index. The activation function represents the risk level determination. This represents the dynamic adjustment weight of the i-th multi-level feature at time t. This represents the value of the i-th multi-scale feature at time t.

[0114] S4. Provide early warning levels based on prediction results, and optimize resource scheduling and maintenance configuration through intelligent scheduling algorithms;

[0115] It should be noted that the warning levels are divided as follows:

[0116] 1. Green (Low Risk): Failure probability <10%;

[0117] 2. Yellow (Medium Risk): 10% ≤ probability of failure < 30%;

[0118] Orange (High Risk): 30% ≤ probability of failure < 50%;

[0119] Red (Emergency): Fault probability ≥ 50%;

[0120] Furthermore, different warning levels are assigned to different devices or areas;

[0121] It should be noted that a dynamic maintenance priority list is established, including priority scheduling and maintenance for equipment in red zones, preventative maintenance for orange zones, and routine maintenance for green zones.

[0122] Optimize resource scheduling and maintenance configuration using a greedy algorithm;

[0123] When resources are limited, maximize the probability of maintaining high-risk equipment;

[0124] Maintain the overall stable operation of the power distribution system;

[0125] Minimize maintenance costs and power distribution system downtime.

[0126] S5. Self-learning based on the monitoring deviation between the actual situation and the prediction results;

[0127] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0128] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of implementing an agent-based power distribution system health monitoring and prediction method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0129] Example 2

[0130] The following is another embodiment of the present invention, which provides a method for health monitoring and prediction of power distribution system based on intelligent agents. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0131] To verify the effectiveness of this invention, actual operating data from a city's power distribution network was used in an experiment. The experimental background included: deploying 50 device sensors (including transformers, switches, circuit breakers, etc.) as single-node intelligent agents in the power distribution system; employing a multi-level intelligent agent collaborative mechanism for fault monitoring, causal analysis, and predictive early warning. Real-time operating status data (temperature, current, voltage, vibration) of the equipment in the power distribution system was collected every second. After integration and processing, the measured data was used for feature extraction and fault root cause identification.

[0132] During the experiment, an average of 60,000 data points were collected daily (i.e., once per second, for approximately 16.7 hours). Before the experiment, real-time load and operating status data for a specific day were selected as training data to construct a time-frequency characteristic model of the equipment in the power distribution system. This data was used to train a self-correcting feature algorithm and a causal relationship inference model, and combined with historical fault data to analyze the root causes of faults.

[0133] In online monitoring, the fault information of the current power distribution system is identified by using the self-correcting feature formula. After continuous monitoring, the power distribution system extracts time-frequency features in real time (such as third-order wavelet energy and spectrum features after Butterworth filtering). When the temperature of a transformer suddenly rises to 75℃ (the normal range is 60℃-70℃), the single-node intelligent agent is triggered to judge it as a potential fault source.

[0134] Within a 30-minute monitoring window (1800 data points in total), the characteristics of this transformer deviated from the historical average by more than 5 times, and the agent reported the fault source as "transformer T1". Converted to a fault probability, this reached 94%, which is about 10% higher than the 85% without a self-correcting algorithm, significantly improving fault detection sensitivity.

[0135] Using a causal inference algorithm, taking the "transformer T1" node as the starting point, and through a particle filter model, 30 consecutive virtual intervention simulations were conducted to obtain path strength information.

[0136] 1. Path A: Transformer T1 → Distribution switch S3 → Distribution transformer T4, with a total impact score of 0.92;

[0137] 2. Path B: Transformer T1 → Surge Arrester R2 → Busbar Connection Point, Impact Score 0.76;

[0138] The high impact score of path A indicates that the T1 fault is most likely caused by internal coil overheating, while the impact of path B is relatively low. The path strength recursive formula shows that after virtual intervention, the path strength from T1 to S3 continues to rise, which is consistent with the actual temperature anomaly.

[0139] After analysis, the agent uses reinforcement learning to assess the impact of fault paths and determine the optimal repair sequence, thus scheduling the maintenance of transformer T1 in advance. The scheduling instructions are automatically generated by the global agent.

[0140] 1. Arrange for maintenance personnel to replace the temperature control switch of transformer T1 before 2:00 AM the following day to reduce the risk of overheating;

[0141] 2. Adjust the load distribution of neighboring nodes to alleviate the load pressure on T1, thereby reducing the probability of failure again.

[0142] During implementation, the agent continuously monitors changes in equipment status and the execution of scheduling strategies based on actual performance, dynamically adjusting the impact scores of causal paths and the reliability of identifying potential fault sources. For example, after overhauling transformer T1, if the temperature gradually returns to within the normal range (60℃–70℃), the agent automatically reduces its path impact score, thereby lowering the priority of T1's fault assessment. Furthermore, if a decrease in frequency fluctuations due to load redistribution is detected during scheduling execution (e.g., frequency deviation decreasing from 0.12Hz to 0.05Hz), the agent will optimize future scheduling strategies accordingly.

[0143] Meanwhile, based on feedback data, a self-correcting feature algorithm is used to continuously adjust the anomaly threshold and path influence model parameters to improve the accuracy of fault prediction. For example, within 30 minutes of continuous monitoring, the fault confirmation rate increased from 94% to 96%, corresponding to a 15% decrease in the false alarm rate. Through these dynamic adjustments, the fault location accuracy and dispatch response speed of the entire power distribution system are significantly enhanced, enabling early warning and efficient repair of potential faults, and ensuring the safe and stable supply of power.

[0144] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for health monitoring and prediction of power distribution systems based on intelligent agents, characterized in that, Specifically, it includes: Establish a cross-level intelligent agent collaboration mechanism, and identify the equipment fault source of the power distribution system through the autonomous judgment of a single node intelligent agent, and package and report the node status and fault information to the macro level. By dynamically registering and connecting different device sensors through a plug-in API architecture, and using an event-driven model to support asynchronous mechanisms to collect and preprocess the operating status data of devices in the power distribution system in real time; Time-frequency features are dynamically extracted by combining historical operating status data of equipment in the power distribution system with a self-correcting feature algorithm. The single-node intelligent agent is embedded in local inference, and the root cause of the fault is analyzed by causal relationship inference algorithm based on the equipment relationship graph of the power distribution system. The specific formula for the self-correcting feature algorithm is as follows: ； in, This indicates the time-frequency features extracted after self-correction. This represents the time-frequency features extracted after correction from the previous time point. This indicates the time-frequency features extracted from the current running status data. This represents historical operational status data. The self-correcting learning rate represents the degree of feature fusion. Indicates the fault penalty coefficient; The path strength information of each fault's impact at different time points is estimated recursively through virtual intervention; The specific formula for recursively estimating the path strength information of each fault's impact at different time points is expressed as follows: ； in, Indicates path strength information, This indicates the total number of paths affected by the fault. Indicates particle index, This represents the influence weight of the k-th particle at time point t after virtual intervention. This represents the effect of virtual intervention on the k-th particle at time point t; Based on global information, formulate scheduling and maintenance strategies and instructions, and feed these strategies and instructions back to the global nodes of the power distribution system through inter-level collaboration. Reinforcement learning is used to derive the optimal scheduling instruction based on the paths affected by all faults; The global node decomposes the scheduling policy instructions into sub-policies and transmits them to potential fault sources of equipment in the power distribution system through hierarchical relationships; The hierarchical structure allows for upward feedback, reporting the strategy execution effectiveness, actual deviations, and potential risks to the global node. Set up a closed-loop collaboration between the global, regional, and device levels to automatically coordinate the execution of scheduling and maintenance strategy instructions by the neighboring single-node intelligent agents of the currently faulty node. The robustness of the AI-based scenario-based virtual verification strategy under different fault conditions was demonstrated. Based on the feedback results of the execution scheduling and maintenance strategy instructions, the system scheduling agent is used to predict the risk of failure. Based on the prediction results, early warning levels are provided, and resource scheduling and maintenance configurations are optimized through intelligent scheduling algorithms; It learns from the monitoring deviation between the actual situation and the prediction results.

2. The method for health monitoring and prediction of a power distribution system based on intelligent agents according to claim 1, characterized in that: The equipment relationship diagram based on the power distribution system analyzes the root cause of faults using a causal relationship inference algorithm, specifically including: The device sensors are treated as nodes in the causal relationship, and the impact of the fault is treated as an edge in the causal relationship. We introduce particles to represent the existence of each causal relationship edge; A custom prior distribution is defined based on historical operating status data, and particle weights are updated according to real-time collected operating status data of equipment in the power distribution system. Resampling is applied, and the hierarchical structure of the equipment relationship diagram of the power distribution system is encoded; By recursively estimating the path strength information of each fault's impact at different time points through virtual intervention, and combining this with changes in the operating status of the power distribution system's equipment, the reliability of the causal relationship between the equipment in the power distribution system is dynamically adjusted. Impact scores are calculated for particle sets along all fault-affected paths, and potential fault sources in the power distribution system are identified by ranking them by impact.

3. The method for health monitoring and prediction of a power distribution system based on intelligent agents according to claim 1, characterized in that: The feedback results of the strategy instructions based on execution scheduling and maintenance are used by the system scheduling agent to predict fault risks, specifically including: Multi-scale time-frequency decomposition (such as multi-level wavelet transform or multi-scale short-time Fourier transform) is used to extract multi-level features from the feedback data. The graph neural network model is used to perform causal learning on the equipment relationship diagram of the power distribution system. The feedback results are used as reward signals to input into the reinforcement learning model to dynamically learn the causal relationship between maintenance effect and failure risk. Automatically generate a fault risk warning index based on time, frequency, and causal structure.

4. The method for health monitoring and prediction of a power distribution system based on intelligent agents according to claim 3, characterized in that: The specific formula for generating the fault risk warning index is as follows: ； in, This indicates a fault risk warning index. The activation function represents the risk level determination. This represents the dynamic adjustment weight of the i-th multi-level feature at time t. This represents the value of the i-th multi-scale feature at time t.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the agent-based power distribution system health monitoring and prediction method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the agent-based power distribution system health monitoring and prediction method as described in any one of claims 1 to 4.

Images