An active power distribution network fault self-healing system and method thereof

Abstract

The application relates to the technical field of power distribution control, in particular to an active power distribution network fault self-healing system and a method thereof, which comprises a data sensing module, a fault positioning module, an isolation control module, a fault self-healing module and a power supply recovery module, and realizes intelligent processing of the whole process from fault detection to power supply recovery through cooperation of the multiple modules. The system can quickly position faults based on FTU layout structure and operation data, optimizes analysis efficiency in combination with a GPU computing power dynamic distribution mechanism, and realizes accurate isolation of a fault area, self-healing repair and smooth grid recovery by using intelligent strategy generation and digital twin technology. The application effectively improves the power supply reliability, system stability and operation intelligent level of the power distribution network.

Description

An active power distribution network fault self-healing system and method thereof Technical Field

[0001] This invention relates to the field of power distribution control technology, and in particular to an active power distribution network fault self-healing system and method thereof. Background Technology

[0002] With the high proportion of distributed energy access, active distribution network fault handling faces challenges: the speed and accuracy of fault location are difficult to balance due to limited edge computing power; traditional FTU (Feeder Terminal Unit) settings are rigid and cannot dynamically allocate computing power; when islanded systems are restored to grid connection, voltage and frequency fluctuations make smooth recovery difficult.

[0003] For example, Chinese Patent CN116388185B discloses a method and system for fault handling and rapid self-healing in active distribution networks, belonging to the field of power system control and protection. When a fault occurs in the active distribution network, causing the instantaneous voltage value to exceed the limit, the fault location device is immediately activated. Through a distribution terminal device equipped with a high-speed communication module, the electrical quantities, topology information, branch switch status, and tie switch status in the ring main unit are uploaded to the master station system. The master station system divides the active distribution network into islands based on fault information, independent power supply capacity, and communication capabilities, combining islands with shared loads. Using second-order cone programming, with load priority, power supply reliability, network loss, and minimum number of switching operations as objective functions, the active distribution system is divided and reorganized. The master station issues a fault network self-healing reconstruction command, disconnecting the faulty section and restoring power to the undervoltage ring main units in the non-faulty sections, thus achieving rapid self-healing of active distribution network faults. However, this solution still has problems such as insufficient fault location speed, conflict between its accuracy and edge computing power and computing speed, insufficient computing power due to non-dynamic FTU settings, and difficulty in controlling the smoothness of recovery after island operation. Summary of the Invention

[0004] To address these issues, the present invention provides an active power distribution network fault self-healing system and method to overcome the problems in the prior art, such as insufficient fault location speed, conflict between its accuracy and edge computing power and computing speed, inability of FTU to keep up with computing power due to non-dynamic settings, and difficulty in controlling the smoothness of recovery after islanded operation.

[0005] To achieve the above objectives, in one aspect, the present invention provides an active power distribution network fault self-healing system, comprising:

[0006] The data sensing module is used to collect power grid structure data;

[0007] The fault location module is used to analyze the fault location based on the FTU layout structure in the power grid structure data, obtain the fault location, perform fault calibration on the FTU layout structure based on the FTU operation data in the power grid structure data, and adjust the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data.

[0008] The isolation control module is used to generate a fault isolation strategy based on the fault location, and to isolate the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area.

[0009] The fault self-healing module is used to acquire fault operation data of the fault isolation area and generate a repair plan based on the fault operation data. The fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area.

[0010] The power restoration module is used to generate power restoration strategies, connect the repaired isolated area to the normal operation area according to the power restoration strategies, collect repair operation data, optimize the difficulty of the power restoration strategy according to the repair operation data, and perform computing power correction on the difficulty optimization process according to the GPU usage ratio in the power grid structure data.

[0011] Furthermore, the fault location module acquires FTU location data based on the FTU layout structure in the power grid structure data, and inputs the FTU location data into a pre-set fault location analysis model to obtain the fault location output by the fault location analysis model.

[0012] Furthermore, the fault location module inputs the FTU operating data from the power grid structure data into a pre-set FTU fault judgment model, obtains the FTU fault status output by the FTU fault judgment model, the FTU fault status includes FTU faults and FTU non-faults, and performs fault calibration on the FTU layout structure based on the FTU fault status, wherein:

[0013] When the FTU failure condition is that the FTU is not faulty, no fault calibration is performed on the FTU layout structure;

[0014] When the FTU failure condition is FTU failure, fault calibration is performed on the FTU layout structure, and the fault calibration includes:

[0015] Step A1: Select FTU devices with an FTU failure status as FTU optimization points;

[0016] Step A2: Remove the FTU optimization points from the FTU layout structure to obtain a new FTU layout structure;

[0017] Step A3: Reacquire the FTU location data according to the new FTU layout structure.

[0018] Furthermore, the fault location module adjusts the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data. Specifically, this includes: comparing the GPU usage ratio Z in the power grid structure data with the preset GPU (Graphics Processing Unit) usage ratio Z0; judging the computing power of edge nodes based on the comparison result; and adjusting the computing power of the fault calibration process based on the judgment result.

[0019] When Z≤Z0, the fault location module determines that the computing power of the edge node is sufficient and does not adjust the computing power during the fault calibration process;

[0020] When Z > Z0, the fault location module determines that the edge node's computing power is insufficient, and adjusts the computing power during the fault calibration process. The computing power adjustment includes:

[0021] Step B1: Analyze the edge nodes with insufficient computing power to identify the points where FTUs can be turned off;

[0022] Step B2: Add the points that can be turned off to the FTU optimization points.

[0023] Furthermore, the isolation control module constructs an isolation strategy generation model using an isolation strategy generation model construction method, obtains the isolation strategy generation model by inputting the fault location into the isolation strategy generation model, acquires the fault isolation strategy output by the isolation strategy generation model, and performs fault isolation on the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area. The isolation strategy generation model construction method includes:

[0024] Step C1: Divide the isolation strategy dataset in the power grid structure data into a 70% strategy training set and a 30% strategy validation set;

[0025] Step C2: Select a recurrent neural network model as the neural network architecture for the isolation strategy generation model, and choose the Adam optimizer and cross-entropy loss function to train the recurrent neural network model.

[0026] Step C3: Load the policy training set into the recurrent neural network model, perform forward propagation through the recurrent neural network model, and calculate the output value of the recurrent neural network model;

[0027] Step C4: Calculate the loss function value based on the output value and the true value of the recurrent neural network model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the recurrent neural network model. Repeat the process of forward propagation, calculating the loss function value and backpropagation until the preset training rounds are reached.

[0028] Step C5: Verify the accuracy of the recurrent neural network model using the policy validation set, and output the recurrent neural network model with an accuracy of 90% or higher as the isolation policy generation model.

[0029] Furthermore, the fault self-healing module acquires fault operation data of the fault isolation area and inputs the fault operation data into a preset self-healing probability analysis model to obtain the self-healing probability G output by the self-healing probability analysis model. The self-healing probability G is compared with a preset self-healing probability G0, and the self-healing control status is judged based on the comparison result. A repair plan is generated based on the judgment result, wherein:

[0030] When G > G0, the fault self-healing module determines that the self-healing control is self-healable. The repair scheme is to send a self-healing control command to the active distribution network terminal and generate a temporary microgrid in the fault isolation area.

[0031] When G≤G0, the fault self-healing module determines that the self-healing control is not self-healing control. The repair plan is to send a manual repair alarm to the active distribution network terminal, and have a human expert repair the fault isolation area and generate a temporary microgrid in the fault isolation area.

[0032] The fault self-healing module repairs the fault isolation area according to the repair plan, resulting in a repaired isolation area.

[0033] Furthermore, the power restoration module acquires the repair operation data of the repaired isolation area, inputs the repair operation data into a preset power restoration strategy model to obtain the power restoration strategy output by the power restoration strategy model, and connects the repaired isolation area to the normal operation area according to the power restoration strategy.

[0034] Furthermore, the power restoration module calculates the smooth recovery difficulty coefficient P based on the voltage difference Vc, phase angle difference Jc, frequency difference Kc, first difficulty weight w1, second difficulty weight w2, and third difficulty weight w3 in the repair operation data, setting P = Vc×w1 + Jc×w2 + Kc×w3. It then compares the smooth recovery difficulty coefficient P with the preset smooth recovery difficulty coefficient P0, judges the ease of smooth recovery based on the comparison result, and optimizes the power restoration strategy based on the judgment result.

[0035] When P≤P0, the power restoration module determines that the smooth restoration is easy and does not perform any difficulty optimization on the power restoration strategy.

[0036] When P > P0, the power restoration module determines that smooth restoration is difficult and optimizes the power restoration strategy accordingly. This optimization includes:

[0037] Step D1: Input the repair operation data into the digital twin model set in the cloud;

[0038] Step D2: Obtain the power restoration optimization strategy output by the digital twin model;

[0039] Step D3: Replace the content of the power restoration strategy with the content of the power restoration optimization strategy, and connect the repaired isolated area to the normal operation area according to the power restoration strategy.

[0040] Furthermore, the power restoration module compares the GPU usage percentage Z in the power grid structure data with the preset GPU usage percentage Z0, judges the computing power of the edge nodes based on the comparison result, and corrects the computing power of the difficulty optimization process based on the judgment result, wherein:

[0041] When Z≤Z0, the power restoration module determines that the computing power of the edge node is sufficient and does not perform computing power correction during the difficulty optimization process;

[0042] When Z > Z0, the power restoration module determines that the edge node's computing power is insufficient and performs computing power correction on the difficulty optimization process. The computing power correction includes:

[0043] Step F1: Correct the preset smooth recovery difficulty coefficient P0 by the correction coefficient α to obtain the corrected preset smooth recovery difficulty coefficient P0j. Set α=1.15, P0j=α×P0;

[0044] Step F2: Replace the value of the preset smooth recovery difficulty coefficient P0 with the value of the corrected preset smooth recovery difficulty coefficient P0j.

[0045] Step F3: Compare the smooth recovery difficulty coefficient P with the preset smooth recovery difficulty coefficient P0 again.

[0046] On the other hand, the present invention also provides a method for an active power distribution network fault self-healing system, comprising:

[0047] Step S1: Collect power grid structure data.

[0048] Step S2: Analyze the fault location based on the FTU layout structure in the power grid structure data to obtain the fault location; perform fault calibration on the FTU layout structure based on the FTU operation data in the power grid structure data; and adjust the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data.

[0049] Step S3: Generate a fault isolation strategy based on the fault location, and perform fault isolation on the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area.

[0050] Step S4: A repair plan is generated based on the fault operation data. The fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area.

[0051] Step S5: Generate a power restoration strategy, connect the repaired isolated area to the normal operation area according to the power restoration strategy, optimize the difficulty of the power restoration strategy based on the repair operation data, and correct the computing power of the difficulty optimization process based on the GPU usage ratio in the power grid structure data.

[0052] Compared with existing technologies, the beneficial effects of this invention are as follows: the system collects power grid structure data in real time through the data sensing module, providing comprehensive and accurate data support for subsequent analysis; the system, through the fault location module, combines the FTU layout structure with real-time operating data to quickly and accurately analyze the fault location; and it innovatively introduces a dynamic GPU computing power adjustment mechanism, intelligently allocating computing resources based on the real-time GPU usage ratio, thereby significantly improving the analysis speed while ensuring location accuracy, effectively resolving the inherent conflict between location efficiency and computing power resources. The fault location module also performs dynamic fault calibration on the FTU layout structure; when a faulty FTU is identified, it can automatically remove it from the analysis model and update the topology data, ensuring the reliability of the data source used for fault analysis. This process, in conjunction with GPU computing power adjustment, constitutes dynamic sensing. With its adaptive computing power allocation system, the system overcomes the problem of computing power supply and demand imbalance caused by the rigidity of traditional FTU settings. Furthermore, the system automatically generates isolation strategies through an isolation control module, quickly dividing faulty and normal areas to prevent fault spread and ensure system safety. The system also uses a fault self-healing module to intelligently generate repair plans based on fault data, enabling automatic repair of faulty areas and reducing manual intervention and power outage time. The system calculates the smooth recovery difficulty coefficient through a power restoration module, and when the recovery difficulty is determined to be high, it automatically calls a cloud-based digital twin model to generate an optimization strategy, achieving precise control over voltage, frequency, and phase angle difference. Simultaneously, this module also integrates GPU computing power correction functions to ensure the efficiency of strategy generation in complex computing scenarios, thereby greatly improving the smoothness and stability of the islanded system's reconnection process and effectively preventing secondary impacts. Attached Figure Description

[0053] Figure 1 is a schematic diagram of the active power distribution network fault self-healing system in this embodiment;

[0054] Figure 2 is a schematic diagram of the fault calibration process in this embodiment;

[0055] Figure 3 is a schematic diagram of the difficulty optimization process in this embodiment;

[0056] Figure 4 is a flowchart illustrating the method of the active power distribution network fault self-healing system in this embodiment. Detailed Implementation

[0057] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0058] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0059] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0060] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0061] Please refer to Figures 1 to 3, which are schematic diagrams of the active power distribution network fault self-healing system of this embodiment. The system includes:

[0062] The data sensing module is used to collect power grid structure data;

[0063] The fault location module is used to analyze the fault location based on the FTU layout structure in the power grid structure data, obtain the fault location, perform fault calibration on the FTU layout structure based on the FTU operation data in the power grid structure data, and adjust the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data.

[0064] The isolation control module is used to generate a fault isolation strategy based on the fault location, and to isolate the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area.

[0065] The fault self-healing module is used to acquire fault operation data of the fault isolation area and generate a repair plan based on the fault operation data. The fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area.

[0066] The power restoration module is used to generate power restoration strategies, connect the repaired isolated area to the normal operation area according to the power restoration strategies, collect repair operation data, optimize the difficulty of the power restoration strategy according to the repair operation data, and perform computing power correction on the difficulty optimization process according to the GPU usage ratio in the power grid structure data.

[0067] Specifically, the active power distribution network fault self-healing system is applied in active power distribution network terminals. This system enables rapid fault location, precise isolation, and self-healing recovery, improving power supply reliability and intelligence, optimizing GPU computing power allocation, and enhancing system efficiency and stability. The system collects real-time power grid structure data through a data sensing module, providing comprehensive and accurate data support for subsequent analysis. The system, through the fault location module combined with FTU layout structure and real-time operating data, quickly and accurately analyzes fault locations and innovatively introduces a dynamic GPU computing power adjustment mechanism. This mechanism intelligently allocates computing resources based on real-time GPU usage, significantly improving analysis speed while ensuring location accuracy. This effectively resolves the inherent conflict between location efficiency and computing power resources. The fault location module also performs dynamic fault calibration on the FTU layout structure. When a faulty FTU is identified, it automatically removes it from the analysis model and updates the topology data, ensuring accurate fault location. The reliability of the analyzed data source is ensured through this process, which, in conjunction with GPU computing power adjustment, constitutes a dynamic perception and adaptive computing power allocation system. This overcomes the problem of computing power supply and demand imbalance caused by the rigidity of traditional FTU settings. The system also automatically generates isolation strategies through the isolation control module, quickly dividing fault areas into normal areas to prevent fault spread and ensure system safety. Furthermore, the system uses a fault self-healing module to intelligently generate repair plans based on fault data, enabling automatic repair of fault areas and reducing manual intervention and power outage time. The system calculates the smooth recovery difficulty coefficient through the power supply recovery module, and when the recovery difficulty is determined to be high, it automatically calls the cloud-based digital twin model to generate optimization strategies, achieving precise control over voltage, frequency, and phase angle difference. At the same time, the power supply recovery module also integrates GPU computing power correction functions to ensure the efficiency of strategy generation in complex computing scenarios, thereby greatly improving the smoothness and stability of the islanded system's reconnection process and effectively preventing secondary impacts.

[0068] Specifically, the power grid structure data includes FTU layout structure, FTU operation data, and GPU usage ratio. The FTU layout structure refers to the geographical distribution and topological connection relationship of FTU equipment in the distribution network. The data sensing module collects the FTU layout structure through the power grid GIS (Geographic Information System). The FTU operation data refers to the real-time or historical operation status data collected by the FTU equipment during operation, such as current, voltage, power, and switch status (open / closed). The data sensing module collects the FTU operation data through the current transformer, voltage transformer, and switch monitoring circuit built into the FTU. The GPU usage ratio refers to the computing resource utilization rate of the graphics processor in the GPU system, that is, the load ratio of the GPU in the current task. The data sensing module queries the GPU usage ratio in real time through the nvidia-smi command tool (for NVIDIA GPUs).

[0069] Specifically, the fault location module acquires FTU location data based on the FTU layout structure in the power grid structure data, and inputs the FTU location data into a pre-set fault location analysis model to obtain the fault location output by the fault location analysis model.

[0070] Specifically, the pre-set fault location analysis model refers to a pre-set recurrent neural network model that takes FTU location data as input and outputs fault location. This embodiment does not limit the specific construction method of the fault location analysis model; those skilled in the art can set it according to actual conditions. For example, the fault analysis dataset can be divided into a 70% fault analysis training set, a 20% fault analysis validation set, and a 10% fault analysis test set. The fault analysis training set is input into the recurrent neural network for forward propagation to obtain the fault location analysis result. The cross-entropy function is used to calculate the loss value of the fault location analysis result. Backpropagation is then performed based on the loss value, and optimization is performed... The parameters of a device, such as Adam, are updated to obtain a trained recurrent neural network model. The fault analysis validation set is input into the trained recurrent neural network model. The trained recurrent neural network model with the lowest loss value on the fault analysis validation set is then forward-propagated again through the fault analysis test set to obtain the final fault location analysis result. The trained recurrent neural network model with a final fault location analysis result accuracy greater than 95% is output as the fault location analysis model. The fault analysis dataset refers to the training dataset used to construct the fault location analysis model. The fault analysis dataset includes historically acquired FTU location data and the fault locations corresponding to the historically acquired FTU location data.

[0071] Specifically, the fault location module inputs FTU operating data from the power grid structure data into a pre-set FTU fault judgment model, obtains the FTU fault status output by the FTU fault judgment model, which includes FTU faults and FTU non-faults, and performs fault calibration on the FTU layout structure based on the FTU fault status, wherein:

[0072] When the FTU failure condition is that the FTU is not faulty, no fault calibration is performed on the FTU layout structure;

[0073] When the FTU failure condition is FTU failure, fault calibration is performed on the FTU layout structure, and the fault calibration includes:

[0074] Step A1: Select FTU devices with an FTU failure status as FTU optimization points;

[0075] Step A2: Remove the FTU optimization points from the FTU layout structure to obtain a new FTU layout structure;

[0076] Step A3: Reacquire the FTU location data according to the new FTU layout structure.

[0077] Specifically, the pre-set FTU fault judgment model refers to a pre-set recurrent neural network model that takes FTU operating data as input and outputs FTU fault conditions. This embodiment does not limit the specific construction method of the FTU fault judgment model; those skilled in the art can set it according to actual conditions. For example, the fault judgment dataset can be divided into a 70% fault judgment training set, a 20% fault judgment validation set, and a 10% fault judgment test set. The fault judgment training set is input into the recurrent neural network for forward propagation to obtain the fault condition judgment result. The cross-entropy function is used to calculate the loss value of the fault condition judgment result. Backpropagation is then performed based on the loss value, and optimization is performed... The system updates parameters using a device such as Adam to obtain a trained recurrent neural network model. The fault judgment validation set is then input into the trained recurrent neural network model. The trained recurrent neural network model with the lowest loss value on the fault judgment validation set is then forward-propagated again through the fault judgment test set to obtain the final fault judgment result. The trained recurrent neural network model with a final fault judgment result accuracy greater than 95% is output as the FTU fault judgment model. The fault judgment dataset refers to the training dataset used to construct the FTU fault judgment model. The fault judgment dataset includes historically acquired FTU operating data and the FTU fault conditions corresponding to the historically acquired FTU operating data.

[0078] Specifically, the fault location module adjusts the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data. This includes comparing the GPU usage ratio Z in the power grid structure data with a preset GPU usage ratio Z0, judging the computing power of edge nodes based on the comparison result, and adjusting the computing power of the fault calibration process based on the judgment result.

[0079] When Z≤Z0, the fault location module determines that the computing power of the edge node is sufficient and does not adjust the computing power during the fault calibration process;

[0080] When Z > Z0, the fault location module determines that the edge node's computing power is insufficient, and adjusts the computing power during the fault calibration process. The computing power adjustment includes:

[0081] Step B1: Analyze the edge nodes with insufficient computing power to identify the points where FTUs can be turned off;

[0082] Step B2: Add the points that can be turned off to the FTU optimization points.

[0083] Specifically, the preset GPU usage percentage refers to a preset value used to judge the computing power of edge nodes. This embodiment does not limit the specific value of the preset GPU usage percentage. Those skilled in the art can set it according to the actual situation, such as setting the specific value of the preset GPU usage percentage to 70% according to the computing power redundancy requirements of the system.

[0084] Specifically, the isolation control module constructs an isolation strategy generation model using an isolation strategy generation model construction method, obtains the isolation strategy generation model by inputting the fault location into the isolation strategy generation model, acquires the fault isolation strategy output by the isolation strategy generation model, and performs fault isolation on the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area. The isolation strategy generation model construction method includes:

[0085] Step C1: Divide the isolation strategy dataset in the power grid structure data into a 70% strategy training set and a 30% strategy validation set;

[0086] Step C2: Select a recurrent neural network model as the neural network architecture for the isolation strategy generation model, and choose the Adam optimizer and cross-entropy loss function to train the recurrent neural network model.

[0087] Step C3: Load the policy training set into the recurrent neural network model, perform forward propagation through the recurrent neural network model, and calculate the output value of the recurrent neural network model;

[0088] Step C4: Calculate the loss function value based on the output value and the true value of the recurrent neural network model, calculate the gradient through the backpropagation algorithm, and update the weights and biases of the recurrent neural network model. Repeat the process of forward propagation, calculating the loss function value and backpropagation until the preset training rounds are reached.

[0089] Step C5: Verify the accuracy of the recurrent neural network model using the policy validation set, and output the recurrent neural network model with an accuracy of 90% or higher as the isolation policy generation model.

[0090] Specifically, the fault isolation strategy refers to the strategy of isolating the fault location, the power grid area refers to the area covered by the entire power supply power grid, the fault isolation refers to isolating the fault location, the fault isolation area refers to the area formed by the isolated fault location, and the normal operation area refers to the area other than the fault isolation area.

[0091] Specifically, the fault self-healing module acquires fault operation data of the fault isolation area and inputs the fault operation data into a preset self-healing probability analysis model to obtain the self-healing probability G output by the self-healing probability analysis model. The self-healing probability G is compared with a preset self-healing probability G0, and the self-healing control status is judged based on the comparison result. A repair plan is generated based on the judgment result, wherein:

[0092] When G > G0, the fault self-healing module determines that the self-healing control is self-healable. The repair scheme is to send a self-healing control command to the active distribution network terminal and generate a temporary microgrid in the fault isolation area.

[0093] When G≤G0, the fault self-healing module determines that the self-healing control is not self-healing control. The repair plan is to send a manual repair alarm to the active distribution network terminal, and have a human expert repair the fault isolation area and generate a temporary microgrid in the fault isolation area.

[0094] The fault self-healing module repairs the fault isolation area according to the repair plan, resulting in a repaired isolation area.

[0095] Specifically, this embodiment does not limit the method of acquiring fault operation data in the fault isolation area. Those skilled in the art can set it according to actual conditions. For example, the fault operation voltage in the fault operation data can be collected using a voltmeter. The pre-set self-healing probability analysis model refers to a pre-set machine learning model that takes fault operation data as input and outputs self-healing probability. This embodiment does not limit the construction method of the self-healing probability analysis model. Those skilled in the art can set it according to actual conditions. For example, the machine learning model can be trained using a self-healing analysis dataset to obtain the self-healing probability analysis model. The self-healing analysis dataset refers to the training dataset used to construct the self-healing probability analysis model. The self-healing analysis dataset includes historically acquired fault operation data and... The fault probabilities corresponding to historical fault operation data are cleaned and labeled in the self-healing analysis dataset to clarify the binary result of whether self-healing is successful or not for each data point. This serves as the target for model learning. Based on this, in-depth feature engineering is carried out to extract key features related to self-healing capability from the fault operation data, such as fault nature, network topology connectivity, and automated equipment status. An appropriate machine learning model is selected for training, here a gradient boosting tree is chosen, to learn the complex mapping relationship from features to self-healing results. Probability calibration technology is used to ensure that the output values ​​have true probabilistic meaning. Finally, the trained and optimized machine learning model and its processing flow are solidified and pre-installed into the fault self-healing module as a self-healing probability analysis model.

[0096] Specifically, the power restoration module acquires the repair operation data of the isolated area after repair, inputs the repair operation data into a preset power restoration strategy model, obtains the power restoration strategy output by the power restoration strategy model, and connects the isolated area after repair to the normal operation area according to the power restoration strategy.

[0097] Specifically, the repair operation data refers to the real-time operation data of the isolated area after repair. This data includes frequency difference, phase angle difference, and voltage difference. This embodiment does not limit the method of acquiring the repair operation data; those skilled in the art can set it according to actual conditions, such as collecting voltage difference data using a voltmeter. The pre-set power restoration strategy model refers to a pre-set decision tree model that takes the repair operation data as input and outputs a power restoration strategy. This embodiment does not limit the construction method of the power restoration strategy model; those skilled in the art can set it according to actual conditions, such as training the decision tree model using a decision analysis dataset. The decision analysis dataset refers to the dataset used to construct the power restoration strategy model. The training dataset, specifically the decision analysis dataset, includes historically acquired repair operation data and corresponding power restoration strategies. The input repair operation data undergoes deep cleaning and structuring to extract key features. Simultaneously, experts, based on historical experience and safety procedures, abstract and categorize effective operational schemes corresponding to the repair operation data into standardized power restoration strategies, such as "switching operations to restore the main line," "activating standby generators for temporary power supply," and "transferring load to adjacent feeders," which serve as the output target for model learning. A large number of historical or simulated repair operation data samples and their corresponding optimal power restoration strategies, determined manually or by experts, are used as the training set. CART (Classification and Applied Logic) is employed. The And Regression Tree (AR) algorithm recursively selects the most discriminative features, such as "Is there a mains fault?", to partition the data. The goal is to make the policy categories within the child nodes as pure as possible. This process continues until a stopping condition is met, which is that the number of samples in a node is less than 20. Finally, a tree structure consisting of a root node, internal decision nodes, and leaf nodes is generated. Each leaf node is associated with a specific power restoration strategy. This embodiment does not limit the specific implementation method of connecting the repaired isolated area to the normal operation area according to the power restoration strategy. Those skilled in the art can set it according to the actual situation, such as using a smooth recovery method to connect the repaired isolated area to the normal operation area according to the power restoration strategy. The smooth recovery refers to the entire process of the power grid reconnecting the faulted area to the normal operation area and gradually bringing it back to the load after the faulted area is repaired.

[0098] Specifically, the power restoration module calculates the smooth recovery difficulty coefficient P based on the voltage difference Vc, phase angle difference Jc, frequency difference Kc, first difficulty weight w1, second difficulty weight w2, and third difficulty weight w3 in the repair operation data, setting P = Vc×w1 + Jc×w2 + Kc×w3. It then compares the smooth recovery difficulty coefficient P with a preset smooth recovery difficulty coefficient P0, judges the ease of smooth recovery based on the comparison result, and optimizes the power restoration strategy based on the judgment result.

[0099] When P≤P0, the power restoration module determines that the smooth restoration is easy and does not perform any difficulty optimization on the power restoration strategy.

[0100] When P > P0, the power restoration module determines that smooth restoration is difficult and optimizes the power restoration strategy accordingly. This optimization includes:

[0101] Step D1: Input the repair operation data into the digital twin model set in the cloud;

[0102] Step D2: Obtain the power restoration optimization strategy output by the digital twin model;

[0103] Step D3: Replace the content of the power restoration strategy with the content of the power restoration optimization strategy, and connect the repaired isolated area to the normal operation area according to the power restoration strategy.

[0104] Specifically, the voltage difference refers to the difference in voltage amplitude between the repaired isolation region and the normal operating region; the phase angle difference refers to the difference in phase angle between the repaired isolation region and the normal operating region; the frequency difference refers to the difference in frequency between the repaired isolation region and the normal operating region; the first difficulty weight refers to the weighting coefficient corresponding to the voltage difference in the calculation of the smooth recovery difficulty coefficient; the second difficulty weight refers to the weighting coefficient corresponding to the phase angle difference in the calculation of the smooth recovery difficulty coefficient; the third difficulty weight refers to the weighting coefficient corresponding to the frequency difference in the calculation of the smooth recovery difficulty coefficient; and the preset smooth recovery difficulty coefficient refers to a preset value for judging the ease or difficulty of smooth recovery. This embodiment does not limit the specific value of the preset smooth recovery difficulty coefficient; those skilled in the art can set it according to actual conditions. For example, through multiple experiments, it was found that when the smooth recovery difficulty coefficient is greater than 0.75, the smooth recovery difficulty is considered difficult to perform. Therefore, P0 is set to 0.75. The ease or difficulty of smooth recovery refers to the ease or difficulty of performing smooth recovery based on the smooth recovery difficulty coefficient and the preset smooth recovery difficulty coefficient. The ease or difficulty of smooth recovery includes difficult smooth recovery and easy smooth recovery. The cloud refers to computing resources accessed via the network. The digital twin model refers to a simulation model that takes repair operation data as input and power supply recovery optimization strategy as output. This embodiment does not limit the construction method of the digital twin model. Those skilled in the art can set it according to the actual situation, such as training the simulation model with a power supply optimization dataset to obtain a digital twin model. The power supply optimization dataset refers to the training dataset used to construct the digital twin model. The power supply optimization dataset includes historically acquired repair operation data and power supply recovery optimization strategies corresponding to the historically acquired repair operation data. The derivation process of the formula P=Vc×w1+Jc×w2+Kc×w3 is as follows:

[0105] According to the core technical requirements for power system grid connection operation, successful grid connection must meet the following three conditions for "quasi-synchronization":

[0106] Equal voltage: The voltage amplitude difference Vc between the two systems must be very small. If the voltage difference is too large, it will cause a huge inrush current, causing the protection equipment to operate or damaging the equipment.

[0107] Phase angle consistency: The voltage phase angle difference Jc between the two systems must be very small. The phase angle difference directly determines the power surge at the moment of closing, which is a key factor causing transient instability of the system.

[0108] Same frequency: The frequency difference Kc between the two systems must be very small. The frequency difference will cause the phase angle to change continuously, making it impossible to continuously meet the grid connection conditions, and even causing power oscillation.

[0109] Therefore, Vc, Jc, and Kc are the three parameters that directly affect the success or failure of smooth recovery and are the most core and intuitive physical quantities. The weighted summation method is used to combine the three factors into a smooth recovery difficulty coefficient. Relying on the knowledge of power system experts, the phase angle difference Jc has the most direct and severe impact on the transient stability of the system. Therefore, its weight w2 is usually set to the highest, with w2=0.6. The frequency difference Kc affects the long-term synchronization maintenance capability, with the second highest weight, with w3=0.3. The impact of the voltage difference Vc is relatively easy to compensate for by adjusting the transformer tap, with the lowest weight w1, with w1=0.1.

[0110] Specifically, the power restoration module compares the GPU usage percentage Z in the power grid structure data with the preset GPU usage percentage Z0, judges the computing power of edge nodes based on the comparison result, and corrects the computing power of the difficulty optimization process based on the judgment result, wherein:

[0111] When Z≤Z0, the power restoration module determines that the computing power of the edge node is sufficient and does not perform computing power correction during the difficulty optimization process;

[0112] When Z > Z0, the power restoration module determines that the edge node's computing power is insufficient and performs computing power correction on the difficulty optimization process. The computing power correction includes:

[0113] Step F1: Correct the preset smooth recovery difficulty coefficient P0 by the correction coefficient α to obtain the corrected preset smooth recovery difficulty coefficient P0j. Set α=1.15, P0j=α×P0;

[0114] Step F2: Replace the value of the preset smooth recovery difficulty coefficient P0 with the value of the corrected preset smooth recovery difficulty coefficient P0j.

[0115] Step F3: Compare the smooth recovery difficulty coefficient P with the preset smooth recovery difficulty coefficient P0 again.

[0116] Specifically, the correction coefficient refers to a preset value used to correct the preset smooth recovery difficulty coefficient. Its value is set to 1.15. When the computing power of the edge node is insufficient, the preset smooth recovery difficulty coefficient is increased by 1.15 times, making it more difficult to determine the smooth recovery difficulty when the computing power is insufficient. Therefore, the preset smooth recovery difficulty coefficient P0 is multiplied by the correction coefficient α to obtain the corrected preset smooth recovery difficulty coefficient increased by 1.15 times.

[0117] Please refer to Figure 4, which is a flowchart illustrating the method of the active power distribution network fault self-healing system in this embodiment. The method includes:

[0118] Step S1: Collect power grid structure data;

[0119] Step S2: Analyze the fault location based on the FTU layout structure in the power grid structure data to obtain the fault location; perform fault calibration on the FTU layout structure based on the FTU operation data in the power grid structure data; and adjust the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data.

[0120] Step S3: Generate a fault isolation strategy based on the fault location, and perform fault isolation on the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area.

[0121] Step S4: A repair plan is generated based on the fault operation data. The fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area.

[0122] Step S5: Generate a power restoration strategy, connect the repaired isolated area to the normal operation area according to the power restoration strategy, optimize the difficulty of the power restoration strategy based on the repair operation data, and correct the computing power of the difficulty optimization process based on the GPU usage ratio in the power grid structure data.

[0123] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An active power distribution network fault self-healing system, characterized in that, include: The data sensing module is used to collect power grid structure data; The fault location module is used to analyze the fault location based on the FTU layout structure in the power grid structure data, obtain the fault location, perform fault calibration on the FTU layout structure based on the FTU operation data in the power grid structure data, and adjust the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data. The isolation control module is used to generate a fault isolation strategy based on the fault location, and to isolate the power grid area according to the fault isolation strategy to obtain the fault isolation area and the normal operation area. The fault self-healing module is used to acquire fault operation data of the fault isolation area and generate a repair plan based on the fault operation data. The fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area. The power restoration module is used to generate power restoration strategies, connect the repaired isolated area to the normal operation area according to the power restoration strategies, collect repair operation data, optimize the difficulty of the power restoration strategy according to the repair operation data, and perform computing power correction on the difficulty optimization process according to the GPU usage ratio in the power grid structure data. The fault location module inputs the FTU operating data from the power grid structure data into a pre-set FTU fault judgment model, obtains the FTU fault status output by the FTU fault judgment model, and the FTU fault status includes FTU fault and FTU not fault. Based on the FTU fault status, the module performs fault calibration on the FTU layout structure. Specifically: when the FTU fault status is FTU not fault, no fault calibration is performed on the FTU layout structure; when the FTU fault status is FTU fault, fault calibration is performed on the FTU layout structure. The fault calibration includes: Step A1, designating the FTU equipment with the FTU fault status as an FTU optimization point; Step A2, removing the FTU optimization point from the FTU layout structure to obtain a new FTU layout structure; Step A3, re-acquiring the FTU location data based on the new FTU layout structure. The fault location module adjusts the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data. Specifically, it compares the GPU usage ratio Z in the power grid structure data with a preset GPU usage ratio Z0, judges the computing power of the edge nodes based on the comparison result, and adjusts the computing power of the fault calibration process based on the judgment result. Specifically, when Z≤Z0, the fault location module determines that the computing power of the edge nodes is sufficient and does not adjust the computing power of the fault calibration process; when Z>Z0, the fault location module determines that the computing power of the edge nodes is insufficient and adjusts the computing power of the fault calibration process. The computing power adjustment includes: step B1, analyzing the edge nodes with insufficient computing power to obtain FTU points that can be shut down; step B2, adding the FTU points that can be shut down to the FTU optimization points.

2. The active power distribution network fault self-healing system according to claim 1, characterized in that, The fault location module acquires FTU location data based on the FTU layout structure in the power grid structure data, and inputs the FTU location data into a pre-set fault location analysis model to obtain the fault location output by the fault location analysis model.

3. The active power distribution network fault self-healing system according to claim 1, characterized in that, The isolation control module constructs an isolation strategy generation model using an isolation strategy generation model construction method. The fault location is input into the isolation strategy generation model to obtain the fault isolation strategy output by the model. Based on the fault isolation strategy, the module performs fault isolation on the power grid area, resulting in a fault isolation area and a normal operating area. The isolation strategy generation model construction method includes: Step C1, dividing the isolation strategy dataset in the power grid structure data into a 70% strategy training set and a 30% strategy validation set; Step C2, selecting a recurrent neural network model as the neural network architecture for the isolation strategy generation model, and selecting the Adam optimizer and cross-entropy. Step C3: The policy training set is loaded into the recurrent neural network model, and forward propagation is performed through the recurrent neural network model to calculate the output value of the recurrent neural network model. Step C4: The loss function value is calculated based on the output value and the true value of the recurrent neural network model. The gradient is calculated through the backpropagation algorithm, and the weights and biases of the recurrent neural network model are updated. The process of forward propagation, calculating the loss function value, and backpropagation is repeated until the preset training rounds are reached. Step C5: The accuracy of the recurrent neural network model is verified through the policy validation set. The recurrent neural network model with an accuracy of 90% or higher is used as the isolated policy generation model for output.

4. The active power distribution network fault self-healing system according to claim 1, characterized in that, The fault self-healing module acquires fault operation data of the fault isolation area and inputs the fault operation data into a preset self-healing probability analysis model to obtain the self-healing probability G output by the self-healing probability analysis model. The self-healing probability G is compared with a preset self-healing probability G0. Based on the comparison result, the self-healing control status is judged, and a repair plan is generated based on the judgment result. Specifically: when G > G0, the fault self-healing module determines that the self-healing control status is self-healable, and the repair plan is: sending a self-healing control command to the active distribution network terminal and generating a temporary microgrid in the fault isolation area; when G ≤ G0, the fault self-healing module determines that the self-healing control status is not self-healable, and the repair plan is: sending a manual repair alarm to the active distribution network terminal, whereby a human expert repairs the fault isolation area and generates a temporary microgrid in the fault isolation area; the fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area.

5. The active power distribution network fault self-healing system according to claim 1, characterized in that, The power restoration module acquires the repair operation data of the isolated area after repair, inputs the repair operation data into a preset power restoration strategy model, obtains the power restoration strategy output by the power restoration strategy model, and connects the isolated area after repair to the normal operation area according to the power restoration strategy.

6. The active power distribution network fault self-healing system according to claim 1, characterized in that, The power restoration module calculates the smooth recovery difficulty coefficient P based on the voltage difference Vc, phase angle difference Jc, frequency difference Kc, first difficulty weight w1, second difficulty weight w2, and third difficulty weight w3 in the repair operation data, setting P = Vc×w1 + Jc×w2 + Kc×w3. It then compares the smooth recovery difficulty coefficient P with a preset smooth recovery difficulty coefficient P0, judging the ease of smooth recovery based on the comparison result, and optimizing the power restoration strategy based on the judgment result. Specifically, when P ≤ P0, the power restoration module determines that smooth recovery is difficult. In cases where smooth recovery is easy, no difficulty optimization is performed on the power restoration strategy. When P > P0, the power restoration module determines that smooth recovery is difficult and performs difficulty optimization on the power restoration strategy. The difficulty optimization includes: step D1, inputting the repair operation data into the digital twin model set in the cloud; step D2, obtaining the power restoration optimization strategy output by the digital twin model; step D3, replacing the content of the power restoration strategy with the content of the power restoration optimization strategy, and connecting the repaired isolated area to the normal operation area according to the power restoration strategy.

7. The active power distribution network fault self-healing system according to claim 1, characterized in that, The power restoration module compares the GPU usage ratio Z in the power grid structure data with the preset GPU usage ratio Z0. Based on the comparison result, it judges the computing power of the edge nodes and performs computing power correction on the difficulty optimization process. Specifically: when Z≤Z0, the power restoration module determines that the computing power of the edge nodes is sufficient and does not perform computing power correction on the difficulty optimization process; when Z>Z0, the power restoration module determines that the computing power of the edge nodes is insufficient and performs computing power correction on the difficulty optimization process. The computing power correction includes: step F1, correcting the preset smooth recovery difficulty coefficient P0 with a correction coefficient α to obtain the corrected preset smooth recovery difficulty coefficient P0j, setting α=1.15, P0j=α×P0; step F2, replacing the value of the preset smooth recovery difficulty coefficient P0 with the value of the corrected preset smooth recovery difficulty coefficient P0j; step F3, comparing the smooth recovery difficulty coefficient P again with the preset smooth recovery difficulty coefficient P0.

8. A method applied to an active distribution network fault self-healing system as described in any one of claims 1-7, characterized in that, Step S1: Collect power grid structure data. Step S2: Analyze the fault location based on the FTU layout structure in the power grid structure data to obtain the fault location; perform fault calibration on the FTU layout structure based on the FTU operation data in the power grid structure data; and adjust the computing power of the fault calibration process based on the GPU usage ratio in the power grid structure data. Step S3: Generate a fault isolation strategy based on the fault location, and perform fault isolation on the power grid area based on the fault isolation strategy to obtain the fault isolation area and the normal operation area. Step S4: A repair plan is generated based on the fault operation data. The fault self-healing module repairs the fault isolation area according to the repair plan to obtain the repaired isolation area. Step S5: Generate a power restoration strategy, connect the repaired isolated area to the normal operation area according to the power restoration strategy, optimize the difficulty of the power restoration strategy based on the repair operation data, and correct the computing power of the difficulty optimization process based on the GPU usage ratio in the power grid structure data.

Images