Power distribution network todynet model optimization and adaptive adjustment method and device

Abstract

The application provides a power distribution network TodyNet model optimization and adaptive adjustment method, device, equipment and medium, comprising: constructing a TodyNet basic model, which is used for extracting time and space characteristics of power distribution network dynamic graph structure data and outputting fault research and judgment results; constructing a reinforcement learning agent, which comprises a state space, an action space and a reward function; the reinforcement learning agent online collects the operation state of the power distribution network and the operation state of the TodyNet basic model in real time, generates a current state vector, and outputs a target adjustment action according to the current state vector; according to the target adjustment action, the model parameters, the model structure or the feature processing strategy of the TodyNet basic model are adaptively adjusted to obtain an optimized TodyNet model; based on the optimized TodyNet model, real-time fault research and judgment are performed on the power distribution network, and a final fault positioning result is output, so as to solve the problem that the TodyNet model parameters and structure cannot be adaptively adjusted, resulting in a decrease in fault research and judgment accuracy and insufficient real-time performance.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance of power systems and the intersection of artificial intelligence, and in particular to a method, device, equipment and medium for optimizing and adaptively adjusting the TodyNet model of a distribution network. Background Technology

[0002] As a crucial component of the power system, the distribution network is characterized by its complex structure and frequent dynamic changes. Its topology often changes in real time due to events such as switch operations, line switching, and fault isolation. Rapid and accurate assessment and location of distribution network faults, especially low-current grounding faults, are of great significance for improving power supply reliability and shortening fault response time.

[0003] Currently, fault assessment in distribution networks often employs graph neural network models to model the distribution network topology. Among them, the TodyNet model, as an advanced dynamic graph neural network architecture, can effectively model the spatiotemporal dependencies of the distribution network through the collaborative work of graph construction modules, dynamic graph transformation modules, dynamic graph isomorphic network modules, and temporal convolution modules, demonstrating good application potential in fault feature extraction and propagation path tracing.

[0004] However, the existing TodyNet model has the following technical shortcomings in practical applications: (1) Static Model Structure and Parameters: After training, the existing TodyNet model keeps its model parameters, graph sparsification strategy, feature fusion method, etc., fixed, and cannot be dynamically and adaptively adjusted according to the real-time operating status of the distribution network (such as topology changes, load fluctuations, and changes in fault types). When the distribution network topology changes or new fault characteristics appear, the model's fault judgment accuracy drops significantly, and real-time performance is difficult to guarantee.

[0005] (2) Model optimization relies on manual intervention: The optimization and updating of the existing TodyNet model mainly relies on manual parameter tuning and offline retraining. Operation and maintenance personnel need to collect historical fault data regularly, manually adjust the model hyperparameters and retrain the model. The whole process is time-consuming and inefficient, cannot adapt to the real-time changes of distribution network data, and is difficult to meet the engineering requirements for rapid fault handling.

[0006] (3) Lack of online adaptation and performance self-correction capabilities: Existing technical solutions lack real-time monitoring and closed-loop feedback mechanisms for the model's own operating status. When the model's performance deteriorates due to environmental changes (such as increased localization error or increased inference latency), it cannot automatically trigger model adjustment or retraining. The model cannot achieve self-optimization and adaptive iteration, making it difficult to guarantee the long-term stable operation of the fault assessment system.

[0007] Therefore, there is an urgent need for a method for optimizing and adaptively adjusting the TodyNet model of the power distribution network to solve the technical problem that the parameters and structure of the existing TodyNet model cannot be adaptively adjusted, resulting in a decrease in the accuracy of fault diagnosis and insufficient real-time performance. Summary of the Invention

[0008] To overcome the problems existing in related technologies, this disclosure provides a method, device, equipment, and medium for optimizing and adaptively adjusting the TodyNet model in power distribution networks, in order to solve the technical problems in related technologies where the parameters and structure of the TodyNet model cannot be adaptively adjusted, resulting in a decrease in the accuracy of fault diagnosis and insufficient real-time performance.

[0009] This specification provides one or more embodiments of a method for optimizing and adaptively adjusting a TodyNet model for a power distribution network, including the following steps: A TodyNet basic model is constructed. The TodyNet basic model is used to extract spatiotemporal features from the dynamic diagram structure data of the distribution network and output fault judgment results. The fault judgment results include fault section location information and / or fault propagation path evolution information. A reinforcement learning agent is constructed, comprising a state space, an action space, and a reward function. The state space is used to characterize the running state of the TodyNet base model and the state of the power distribution network environment. The action space is used to define candidate operations for dynamically adjusting the TodyNet base model. The reward function is used to evaluate the performance of the model after executing the candidate operations. The reinforcement learning agent collects the operating status of the power distribution network and the operating status of the TodyNet basic model online in real time, generates a current state vector, and adjusts its actions according to the target output based on the current state vector. Based on the target adjustment action, the model parameters, model structure, or feature processing strategy of the TodyNet base model are adaptively adjusted to obtain the optimized TodyNet model; The optimized TodyNet model is used to perform real-time fault assessment of the power distribution network and output the final fault location results.

[0010] Preferably, the method further includes the following steps: The reward value is calculated based on the fault assessment result output by the optimized TodyNet model, and the calculated reward value is fed back to the reinforcement learning agent to update the decision-making strategy, forming an online closed-loop adaptive optimization.

[0011] Preferably, the state space specifically includes the following state information: The graph topology information of the current TodyNet model; The number of layers and / or weight parameters of the graph isomorphic network in the TodyNet model; The kernel size of the temporal convolution module; The false alarm rate, false negative rate, or fault location error of the current batch of data; Environmental change indicators for the power distribution network, including topology change frequency and / or load fluctuation amplitude.

[0012] Preferably, the action space specifically includes the following adjustment operations: Adjust the number of layers or nodes in the graph isomorphic network; Adjust the sparsification threshold of the graph adjacency matrix; Adjust the combination of convolution kernel sizes in the temporal convolution module; Switch the fusion method of node features; Trigger online fine-tuning, retraining, or version rollback of the TodyNet model for the power distribution network; Adjust the logical threshold for fault determination.

[0013] Preferably, the reward function is a multi-objective weighted function that simultaneously considers fault location accuracy, inference latency, and topology change response timeliness. It positively rewards high accuracy and low latency, and negatively penalizes failure to adapt to topology changes in a timely manner. It is constructed based on the following evaluation metrics: Accuracy of fault detection and classification; The actual geographical error value for fault location; Response latency of TodyNet model for single fault assessment; The computational resource overhead consumed by model updates; Receive external feedback ratings from maintenance personnel.

[0014] Preferably, the reinforcement learning agent employs a proximal policy optimization algorithm, a deep deterministic policy gradient algorithm, or a soft actor-critic algorithm to fine-tune the parameters of the TodyNet model in a continuous action space.

[0015] Preferably, the step of adaptively adjusting the model parameters, model structure, or feature processing strategy of the TodyNet base model according to the target adjustment action to obtain the optimized TodyNet model specifically includes the following steps: Based on the target adjustment action, the sparsity of the dynamic adjacency matrix or the dimension of the node embedding vector is dynamically adjusted to adapt to the real-time topology changes of the distribution network. Based on the current fault type, different sizes of temporal convolutional kernels or different depths of graph isomorphic network layers are dynamically selected to optimize the extraction capability of spatiotemporal features.

[0016] This specification provides one or more embodiments of a distribution network TodyNet model optimization and adaptive adjustment device, including: The model building module is used to build the TodyNet basic model. The TodyNet basic model is used to extract spatiotemporal features from the dynamic diagram structure data of the distribution network and output fault judgment results. The fault judgment results include fault section location information and / or fault propagation path evolution information. The agent construction module is used to construct a reinforcement learning agent. The reinforcement learning agent includes a state space, an action space, and a reward function. The state space is used to represent the running state of the TodyNet base model and the state of the power distribution network environment. The action space is used to define candidate operations for dynamically adjusting the TodyNet base model. The reward function is used to evaluate the performance of the model after executing the candidate operations. The state awareness module is used by the reinforcement learning agent to collect the operating status of the power distribution network and the operating status of the TodyNet basic model online in real time, generate the current state vector, and adjust the action according to the target output based on the current state vector. The dynamic adjustment module is used to adaptively adjust the model parameters, model structure, or feature processing strategy of the TodyNet base model according to the target adjustment action, so as to obtain an optimized TodyNet model. The fault location module is used to perform real-time fault assessment of the distribution network based on the optimized TodyNet model and output the final fault location result.

[0017] This specification provides one or more embodiments of a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described distribution network TodyNet model optimization and adaptive adjustment method.

[0018] This specification provides one or more embodiments of a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described distribution network TodyNet model optimization and adaptive adjustment method.

[0019] This disclosure provides a method, apparatus, device, and medium for optimizing and adaptively adjusting a TodyNet model in a distribution network. Its advantages lie in that, by constructing a basic TodyNet model, it enables deep spatiotemporal feature extraction of the dynamic graph structure data of the distribution network, thereby initially achieving accurate location of fault sections and evolutionary analysis of fault propagation paths. This provides a reliable benchmark model and initial judgment capability for subsequent adaptive optimization of the model, laying the foundation for the entire method. A reinforcement learning agent comprising a state space, action space, and reward function is constructed, establishing a systematic decision-making mechanism for the dynamic adjustment of the TodyNet model. The state space comprehensively represents the operating state of the model and the distribution network, the action space clearly defines the feasible adjustment operations, and the reward function quantifies the impact of adjustment actions on model performance, thus providing clear optimization objectives and constraints for the agent to learn the optimal adjustment strategy. The reinforcement learning agent collects data from the distribution network and the TodyNet model online in real time. The system operates by generating a current state vector and then outputting an adaptive target adjustment action based on this vector. This ensures that the agent can respond promptly to changes in the environment and model, enabling dynamic and adaptive decision-making and giving the adjustment action a high degree of real-time performance and specificity. Based on the target adjustment action output by the reinforcement learning agent, the parameters, structure, or feature processing strategies of the TodyNet base model are adaptively adjusted to obtain an optimized TodyNet model. This effectively improves the model's adaptability to the dynamic characteristics of the distribution network, enhances its accuracy and robustness in fault assessment under complex operating conditions, and enables continuous model evolution and optimization. Real-time fault assessment of the distribution network is performed based on the optimized TodyNet model, and the final fault location result is output. Applying the adaptively adjusted high-performance model to actual fault detection significantly improves the accuracy of fault segment location and the reliability of fault propagation path prediction, providing strong technical support for the safe and stable operation of the distribution network. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a method for optimizing and adaptively adjusting a TodyNet model for a power distribution network, provided for one or more embodiments of this specification; Figure 2 A schematic diagram of the TodyNet basic model provided for one or more embodiments of this specification; Figure 3 This is a schematic diagram of the structure of dynamic graph transformation provided in one or more embodiments of this specification; Figure 4 A schematic diagram of a TodyNet model optimization and adaptive adjustment device for a power distribution network, provided for one or more embodiments of this specification; Figure 5 This is a schematic diagram of the structure of a computer device provided for one or more embodiments of this specification. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this invention.

[0023] The present invention will now be described in detail with reference to specific embodiments and accompanying drawings.

[0024] Method Implementation Examples According to embodiments of the present invention, a method for optimizing and adaptively adjusting a TodyNet model for a power distribution network is provided, such as... Figure 1 The diagram shown illustrates the flowchart of the TodyNet model optimization and adaptive adjustment method for distribution networks provided in this embodiment. This method employs a two-layer architecture: The bottom layer is a TodyNet dynamic graph neural network, responsible for fault feature extraction, propagation path modeling, and section location in the distribution network. The upper layer is a reinforcement learning adaptive optimization agent, responsible for real-time adjustment of model parameters, structure, and strategies. The two layers form a perception-decision-execution-feedback closed loop, achieving model self-optimization. The TodyNet model optimization and adaptive adjustment method for distribution networks according to this embodiment includes the following steps: S110. Construct the TodyNet basic model. The TodyNet basic model is used to extract spatiotemporal features from the dynamic diagram structure data of the distribution network and output fault assessment results. The fault assessment results include fault section location information and / or fault propagation path evolution information, such as... Figure 2The diagram shown illustrates the TodyNet basic model provided in this embodiment. The TodyNet basic model includes the following core modules: To establish initial relationships between different dimensions, the graph construction and learning module generates a set of graph adjacency matrices, each corresponding to a different time slot. The elements of the graph adjacency matrix can be learned through an iterative process. The k dynamic graph neural network modules consist of dynamic graph transformation and dynamic graph isomorphism networks. Dynamic graph transformation utilizes the dynamic correlations between different time graphs, and its result is implicit in the new adjacency matrix. The dynamic graph isomorphism network processes time series data through k temporal convolutional modules, capturing the spatiotemporal dependencies of multivariate time series.

[0025] The graph building module is used to generate a learnable adjacency matrix for each time slot, which is computed and sparsified by embedding vectors.

[0026] The Dynamic Graph Transform (DGT) module is used to establish directed connections between adjacent time slot graphs, capturing the temporal relationships of fault propagation, such as... Figure 3 The diagram shown is a schematic diagram of the dynamic graph transformation provided in this embodiment.

[0027] The dynamic graph isomorphic network module is used to propagate node information at each layer, aggregate neighbor features, and consider edge weights.

[0028] The temporal convolution module is used to extract the temporal dependency features of each node using multi-scale convolution kernels, and outputs a node representation that integrates spatiotemporal information.

[0029] The output layer is used to output the failure probability distribution, failure type, or failure node probability based on the task.

[0030] S120. Construct a reinforcement learning agent, which includes a state space, an action space, and a reward function.

[0031] The state space is used to characterize the operating state of the TodyNet basic model and the state of the distribution network environment. Specifically, based on the graph topology information of the current TodyNet model, the number of layers and / or weight parameters of the graph isomorphic network in the TodyNet model, the kernel size of the temporal convolution module, the false alarm rate, false negative rate or location error of the fault detection in the current batch of data, the environmental change indicators of the distribution network, including the frequency of topology changes and / or the amplitude of load fluctuations, and the current state output of the TodyNet model, such as the correlation features of adjacent time slot graphs generated by the dynamic graph transformation module (DGT), a real-time state representation for reinforcement learning is constructed.

[0032] The action space is used to define candidate operations for dynamically adjusting the TodyNet base model, and the reward function is used to evaluate the model's performance after executing the candidate operations. Specifically, the actions include, but are not limited to, the following adjustment operations: Adjust the number of layers or nodes in the graph isomorphic network; Adjust the sparsification threshold of the graph adjacency matrix; Adjust the combination of convolution kernel sizes in the temporal convolution module; Switch the fusion method of node features; Trigger online fine-tuning, retraining, or version rollback of the TodyNet model for the power distribution network; Based on real-time data changes, adjust the optimal feature input metrics of node electrical characteristics (voltage, current, etc.) in the multi-head attention mechanism of Graph Attention Network (GAT); Adjust the logical thresholds for fault determination and alarm strategies.

[0033] The reward function evaluates the model's performance after executing candidate operations. It is a multi-objective weighted function that considers fault location accuracy, inference latency, and the timeliness of topology change response. Positive rewards are given for high accuracy and low latency, such as improved fault assessment accuracy and fault location accuracy, to reduce deviations from GIS mapping. Negative rewards are given for untimely adaptation to topology changes, such as increased false positive / false negative rates and excessive model inference latency, to ensure real-time performance. Specifically, the reward function is constructed based on the following evaluation metrics: accuracy of fault detection and classification. The actual geographical error value for fault location; Response latency of TodyNet model for single fault assessment; The computational resource overhead consumed by model updates; Receive external feedback ratings from maintenance personnel.

[0034] Training reinforcement learning agents: By generating a large number of dynamic scenarios using historical fault data and simulation environments, a training set is constructed. At each time step, the agent observes the current state and selects an action according to the current policy. After the action is executed, the structure or parameters of the TodyNet model change. Under the new configuration, subsequent data is used to perform fault assessment, calculate rewards, and store the experience in an experience pool. Algorithms such as deep Q-networks, proximal policy optimization, or soft actor-critic algorithms are used to update the agent network, enabling it to learn the optimal action policy, maximize long-term cumulative rewards, and achieve fine-tuning of the TodyNet model parameters.

[0035] S130. Deploy the trained reinforcement learning agent and TodyNet model in the power distribution network monitoring system. The reinforcement learning agent collects the operating status of the power distribution network and the operating status of the TodyNet basic model online in real time, dynamically updates the state space, generates the current state vector, and adjusts the action according to the target output based on the current state vector every certain period (such as every 5 minutes) or when the state change exceeds the threshold.

[0036] S140. Adjust actions according to the target, adaptively adjust the model parameters, model structure or feature processing strategy of the TodyNet base model to obtain the optimized TodyNet model. It also includes real-time monitoring of the performance indicators of the TodyNet model. When the performance indicators are detected to be lower than the preset threshold, the online learning and update module is automatically triggered to perform model fine-tuning or retraining.

[0037] S150 performs real-time fault assessment of the distribution network based on the optimized TodyNet model, outputs the fault probability distribution, and, combined with a GIS geographic information system, maps fault nodes to their actual geographical locations, outputting the final fault location results for visualized location. Based on the fault propagation path probability matrix, it generates the fault impact range and repair suggestions. It is suitable for application scenarios involving dynamic changes in distribution network topology, load fluctuations, and various types of low-current grounding faults.

[0038] The method provided in this embodiment, by constructing a TodyNet basic model, can perform deep spatiotemporal feature extraction on the dynamic graph structure data of the distribution network, thereby initially realizing accurate location of fault sections and evolutionary analysis of fault propagation paths. This provides a reliable benchmark model and initial judgment capability for subsequent adaptive optimization of the model, laying the foundation for the entire method. A reinforcement learning agent containing a state space, action space, and reward function is constructed, establishing a systematic decision-making mechanism for the dynamic adjustment of the TodyNet model. The state space comprehensively represents the operating state of the model and the distribution network, the action space clearly defines the feasible adjustment operations, and the reward function quantifies the impact of adjustment actions on model performance, thus providing clear optimization objectives and constraints for the agent to learn the optimal adjustment strategy. The reinforcement learning agent collects the operating state of the distribution network and the TodyNet model online in real time, generates the current state vector, and then... The target adjustment action is adapted to the output of this vector, ensuring that the agent can respond promptly to changes in the environment and model, achieving dynamic and adaptive decision-making, and making the adjustment action highly real-time and targeted. Based on the target adjustment action output by the reinforcement learning agent, the parameters, structure, or feature processing strategy of the TodyNet base model are adaptively adjusted to obtain an optimized TodyNet model. This effectively improves the model's adaptability to the dynamic characteristics of the distribution network, enhances its accuracy and robustness in fault judgment under complex operating conditions, and enables continuous evolution and optimization of the model. Based on the optimized TodyNet model, real-time fault judgment of the distribution network is performed, and the final fault location result is output. Applying the adaptively adjusted high-performance model to actual fault detection significantly improves the accuracy of fault section location and the reliability of fault propagation path prediction, providing strong technical support for the safe and stable operation of the distribution network.

[0039] In one embodiment, the TodyNet base model's parameters, structure, or feature processing strategy are adaptively adjusted based on the target adjustment action to obtain an optimized TodyNet model. This includes the following steps: Adjust actions based on objectives, dynamically adjust the sparsity of the dynamic adjacency matrix or the dimension of the node embedding vector to adapt to real-time topology changes in the distribution network.

[0040] Based on the current fault type, different sizes of temporal convolutional kernels or different depths of graph isomorphic network layers are dynamically selected to optimize the extraction capability of spatiotemporal features.

[0041] Device Examples According to embodiments of the present invention, a device for optimizing and adaptively adjusting a TodyNet model of a power distribution network is provided, such as... Figure 4The diagram shown is a structural schematic of the distribution network TodyNet model optimization and adaptive adjustment device provided in this embodiment. According to this embodiment of the invention, the distribution network TodyNet model optimization and adaptive adjustment device includes: Model building module 41 is used to build the TodyNet basic model. The TodyNet basic model is used to extract spatiotemporal features from the dynamic diagram structure data of the distribution network and output fault judgment results. The fault judgment results include fault section location information and / or fault propagation path evolution information.

[0042] The agent construction module 42 is used to construct a reinforcement learning agent. The reinforcement learning agent includes a state space, an action space, and a reward function. The state space is used to represent the running state of the TodyNet base model and the state of the power distribution network environment. The action space is used to define candidate operations for dynamically adjusting the TodyNet base model. The reward function is used to evaluate the performance of the model after executing the candidate operations.

[0043] The state perception module 43 is used for the reinforcement learning agent to collect the operating status of the power distribution network and the operating status of the TodyNet basic model online in real time, generate the current state vector, and adjust the action according to the target output based on the current state vector.

[0044] The dynamic adjustment module 44 is used to adaptively adjust the model parameters, model structure, or feature processing strategy of the TodyNet base model according to the target adjustment action, so as to obtain the optimized TodyNet model.

[0045] The fault location module 45 is used to perform real-time fault analysis on the distribution network based on the optimized TodyNet model and output the final fault location result.

[0046] The device provided in this embodiment constructs a TodyNet basic model through the model building module 41, which can perform deep spatiotemporal feature extraction on the dynamic graph structure data of the distribution network. This enables the initial accurate location of fault sections and evolutionary analysis of fault propagation paths, providing a reliable benchmark model and initial judgment capability for subsequent adaptive optimization of the model, thus laying the foundation for the entire method. The agent construction module 42 constructs a reinforcement learning agent containing a state space, action space, and reward function, establishing a systematic decision-making mechanism for the dynamic adjustment of the TodyNet model. The state space comprehensively represents the operating state of the model and the distribution network, the action space clearly defines the feasible adjustment operations, and the reward function quantifies the impact of adjustment actions on model performance, thereby providing clear optimization goals and constraints for the agent to learn the optimal adjustment strategy. The state perception module 43 collects the operating state of the distribution network and the TodyNet model online in real time and generates the current state vector. The dynamic adjustment module 44, based on the target adjustment action output by the reinforcement learning agent, adaptively adjusts the parameters, structure, or feature processing strategy of the TodyNet basic model according to the target adjustment action output by the reinforcement learning agent, thereby obtaining an optimized TodyNet model. This effectively improves the model's adaptability to the dynamic characteristics of the distribution network, enhances its accuracy and robustness in fault judgment under complex working conditions, and enables continuous evolution and optimization of the model. The fault location module 45 performs real-time fault judgment on the distribution network based on the optimized TodyNet model and outputs the final fault location result. Applying the adaptively adjusted high-performance model to actual fault detection significantly improves the accuracy of fault section location and the reliability of fault propagation path prediction, providing strong technical support for the safe and stable operation of the distribution network.

[0047] The embodiments of the present invention are device embodiments corresponding to the above method embodiments. The specific operations of each module processing step can be understood with reference to the description of the method embodiments, and will not be repeated here.

[0048] like Figure 5 As shown, the present invention also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the distribution network TodyNet model optimization and adaptive adjustment method in the above embodiments, or when the computer program is executed by a processor, it implements the distribution network TodyNet model optimization and adaptive adjustment method in the above embodiments.

[0049] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0050] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and the contents not described in detail in the specification of the present invention are known to those skilled in the art.

Claims

1. A method for optimizing and adaptively adjusting a TodyNet model for power distribution networks, characterized in that, Includes the following steps: A TodyNet basic model is constructed. The TodyNet basic model is used to extract spatiotemporal features from the dynamic diagram structure data of the distribution network and output fault judgment results. The fault judgment results include fault section location information and / or fault propagation path evolution information. A reinforcement learning agent is constructed, comprising a state space, an action space, and a reward function. The state space is used to characterize the running state of the TodyNet base model and the state of the power distribution network environment. The action space is used to define candidate operations for dynamically adjusting the TodyNet base model. The reward function is used to evaluate the performance of the model after executing the candidate operations. The reinforcement learning agent collects the operating status of the power distribution network and the operating status of the TodyNet basic model online in real time, generates a current state vector, and adjusts its actions according to the target output based on the current state vector. Based on the target adjustment action, the model parameters, model structure, or feature processing strategy of the TodyNet base model are adaptively adjusted to obtain the optimized TodyNet model; The optimized TodyNet model is used to perform real-time fault assessment of the power distribution network and output the final fault location results.

2. The method for optimizing and adaptively adjusting the TodyNet model of a distribution network as described in claim 1, characterized in that, It also includes the following steps: The reward value is calculated based on the fault assessment result output by the optimized TodyNet model, and the calculated reward value is fed back to the reinforcement learning agent to update the decision-making strategy, forming an online closed-loop adaptive optimization.

3. The method for optimizing and adaptively adjusting the TodyNet model of a distribution network as described in claim 1, characterized in that, The state space specifically includes the following state information: The graph topology information of the current TodyNet model; The number of layers and / or weight parameters of the graph isomorphic network in the TodyNet model; The kernel size of the temporal convolution module; The false alarm rate, false negative rate, or fault location error of the current batch of data; Environmental change indicators for the power distribution network, including topology change frequency and / or load fluctuation amplitude.

4. The method for optimizing and adaptively adjusting the TodyNet model of a distribution network as described in claim 1, characterized in that, The action space specifically includes the following adjustment operations: Adjust the number of layers or nodes in the graph isomorphic network; Adjust the sparsification threshold of the graph adjacency matrix; Adjust the combination of convolution kernel sizes in the temporal convolution module; Switch the fusion method of node features; Trigger online fine-tuning, retraining, or version rollback of the TodyNet model for the power distribution network; Adjust the logical threshold for fault determination.

5. The method for optimizing and adaptively adjusting the TodyNet model of a distribution network as described in claim 1, characterized in that, The reward function is a multi-objective weighted function that simultaneously considers fault location accuracy, inference latency, and topology change response timeliness. Positive rewards are given for high accuracy and low latency, while negative rewards are given for failure to adapt to topology changes in a timely manner. It is constructed based on the following evaluation metrics: Accuracy of fault detection and classification; The actual geographical error value for fault location; Response latency of TodyNet model for single fault assessment; The computational resource overhead consumed by model updates; Receive external feedback ratings from maintenance personnel.

6. The method for optimizing and adaptively adjusting the TodyNet model of a distribution network as described in claim 1, characterized in that, The reinforcement learning agent employs a proximal policy optimization algorithm, a deep deterministic policy gradient algorithm, or a soft actor-critic algorithm to fine-tune the parameters of the TodyNet model in a continuous action space.

7. The method for optimizing and adaptively adjusting the TodyNet model of a distribution network as described in claim 1, characterized in that, The step of adaptively adjusting the model parameters, model structure, or feature processing strategy of the TodyNet base model according to the target adjustment action to obtain the optimized TodyNet model specifically includes the following steps: Based on the target adjustment action, the sparsity of the dynamic adjacency matrix or the dimension of the node embedding vector is dynamically adjusted to adapt to the real-time topology changes of the distribution network. Based on the current fault type, different sizes of temporal convolutional kernels or different depths of graph isomorphic network layers are dynamically selected to optimize the extraction capability of spatiotemporal features.

8. A device for optimizing and adaptively adjusting a TodyNet model of a power distribution network, characterized in that, include: The model building module is used to build the TodyNet basic model. The TodyNet basic model is used to extract spatiotemporal features from the dynamic diagram structure data of the distribution network and output fault judgment results. The fault judgment results include fault section location information and / or fault propagation path evolution information. The agent construction module is used to construct a reinforcement learning agent. The reinforcement learning agent includes a state space, an action space, and a reward function. The state space is used to represent the running state of the TodyNet base model and the state of the power distribution network environment. The action space is used to define candidate operations for dynamically adjusting the TodyNet base model. The reward function is used to evaluate the performance of the model after executing the candidate operations. The state awareness module is used by the reinforcement learning agent to collect the operating status of the power distribution network and the operating status of the TodyNet basic model online in real time, generate the current state vector, and adjust the action according to the target output based on the current state vector. The dynamic adjustment module is used to adaptively adjust the model parameters, model structure, or feature processing strategy of the TodyNet base model according to the target adjustment action, so as to obtain an optimized TodyNet model. The fault location module is used to perform real-time fault assessment of the distribution network based on the optimized TodyNet model and output the final fault location result.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the distribution network TodyNet model optimization and adaptive adjustment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the distribution network TodyNet model optimization and adaptive adjustment method as described in any one of claims 1 to 7.

Images