Active power distribution network self-healing and self-adaptive decision-making method, system, equipment and medium

Abstract

The invention relates to the technical field of active power distribution networks, and discloses a self-healing and self-adaptive decision-making method, system, equipment and medium for an active power distribution network, and the method comprises the steps: determining the fault type of the active power distribution network through the fault characteristics of the active power distribution network, constructing a deep reinforcement learning intelligent agent according to the operation state parameters of the active power distribution network, and carrying out the self-healing and self-adaptive decision-making of the active power distribution network; and inputting the current operation state parameter and the fault type of the active power distribution network when the fault occurs into the intelligent agent to obtain a self-healing action decision optimized by the intelligent agent, thereby realizing the self-adaptive decision of the fault of the active power distribution network. According to the method, the flexibility and adaptability of self-healing of the active power distribution network are improved, the reliability and efficiency of a self-healing system are ensured through continuous learning and updating mechanisms, and continuous and stable power supply of a power system is effectively guaranteed.

Description

Technical Field

[0001] This invention relates to the field of active power distribution network technology, and in particular to an active power distribution network self-healing adaptive decision-making method, system, device and medium. Background Technology

[0002] With the introduction of more distributed power sources into the distribution network, the network structure is becoming increasingly complex, with multiple branches and ring network operations. This presents numerous challenges to distribution network self-healing technology. Distributed energy resources are vast in number and widely distributed, but their individual capacities are relatively small, and their output is unpredictable and discontinuous. The large-scale, decentralized integration of this type of energy into the grid increases the difficulty of distribution network self-healing control analysis and decision-making. For example, it is difficult to accurately extract and perceive information such as the operating characteristics and power generation of distributed power sources, resulting in long self-healing recovery times and an inability to reduce the area of ​​power outages or shorten outage durations. This delay not only reduces the reliability of the distribution network but may also cause greater economic losses to users. Furthermore, current self-healing technologies lack sufficient integration with information technology, making it difficult to adjust control strategies according to real-time changes in network topology, thus affecting the overall performance of the self-healing system. Therefore, further research is needed on distribution network self-healing technology, optimizing self-healing algorithms and data processing to improve the self-healing capability of the distribution network and ensure a stable power supply.

[0003] In existing technologies, distribution network self-healing methods are prone to hindering power restoration schemes, have poor self-healing reliability, and lack flexibility and adaptability. Summary of the Invention

[0004] In view of this, the present invention provides an active distribution network self-healing adaptive decision-making method, system, device and medium, which solves the technical problems that the distribution network self-healing method is prone to obstruction of power restoration schemes, poor self-healing reliability, and low flexibility and adaptability.

[0005] The first aspect of this invention provides a self-healing and adaptive decision-making method for active distribution networks, comprising:

[0006] In response to a fault request from an active distribution network, the fault characteristics of the active distribution network are obtained, and the fault type of the active distribution network is determined based on the fault characteristics.

[0007] The operating status parameters of the active distribution network are obtained, and a deep reinforcement learning agent is constructed based on the operating status parameters of the active distribution network; the agent includes one or more of the following: a load transfer decision agent, a fault direction signal change decision agent, and a power direction acquisition and tracking change decision agent;

[0008] The current operating status parameters and fault type of the active power distribution network at the time of the fault are input into the intelligent agent, and the intelligent agent optimizes the current operating status parameters and the action parameters under the fault type to obtain the self-healing action decision optimized by the intelligent agent.

[0009] Optionally, acquiring the fault characteristics of the active distribution network and determining the fault type of the active distribution network based on the fault characteristics includes:

[0010] In the event of a fault in the active distribution network, acquire the current fault waveform data of the active distribution network;

[0011] Extract the fault features from the current fault recording data and input the fault features into a pre-trained fault classification prediction model to output the fault type corresponding to the current fault recording data of the active distribution network; wherein, the fault classification prediction model is obtained by training historical samples of fault features and historical samples of fault types based on machine learning.

[0012] Optionally, the method further includes: training the fault classification prediction model;

[0013] Training the fault classification prediction model includes:

[0014] Obtain historical fault recording data of the active power distribution network;

[0015] Extract multiple electrical quantity features and modal components from the fault waveform history data;

[0016] Each modal component is subjected to Hilbent transform, and the transformed modal components are filtered to obtain filtered waveform data;

[0017] A time-frequency matrix is ​​constructed based on the multiple filtered waveform data;

[0018] Singular value decomposition is performed on the time-frequency matrix, and useful fault recording data is obtained by determining the cumulative contribution rate of the filtered waveform data.

[0019] Based on the singular values ​​of the useful fault recording data, determine the statistical characteristics of the singular values;

[0020] Based on the outlier statistical characteristics and the electrical quantity characteristics, the fault characteristics are determined;

[0021] Based on the fault characteristics, a fault type corresponding to the fault characteristics is determined in a preset fault database; wherein, the preset fault database includes a mapping relationship between fault characteristics and fault types;

[0022] Based on the fault feature vector and the fault type, a training dataset and a test dataset are constructed;

[0023] The support vector machine is trained using the training dataset to obtain an initial fault classification prediction model;

[0024] The network parameters of the initial fault classification prediction model are optimized using the test dataset to obtain the fault classification prediction model.

[0025] Optionally, the intelligent agent is a load transfer decision-making intelligent agent;

[0026] The step of obtaining the operating state parameters of the active power distribution network and constructing a deep reinforcement learning agent based on the operating state parameters of the active power distribution network includes:

[0027] Obtain the operating status parameters of the active power distribution network;

[0028] Based on the operating status parameters, the action space and state space of the load transfer decision-making agent are determined; wherein, the action space includes the combined state of the tie switch, the output power adjustment of the distributed power source, and the load shedding amount, and the state space includes the load power, the output power of the distributed power source, the remaining capacity of the feeder, and the fault type.

[0029] Multiple evaluation indicators related to load transfer in the active distribution network are determined, and the reward function of the load transfer decision-making agent is determined by weighting the multiple evaluation indicators; wherein, the evaluation indicators include load balance, number of overloaded feeders, and time for critical loads to be restored to power supply.

[0030] Optionally, the intelligent agent is a fault direction signal change decision-making intelligent agent;

[0031] The step of obtaining the operating state parameters of the active power distribution network and constructing a deep reinforcement learning agent based on the operating state parameters of the active power distribution network includes:

[0032] Obtain the operating status parameters of the active power distribution network;

[0033] Based on the operating status parameters, the action space and state space of the fault direction signal change decision agent are determined; wherein, the action space includes the adjustment amount of the direction alarm threshold of the automatic switch and the adjustment amount of the fault judgment logic configuration of the protection device, and the state space includes the fault direction signal, network topology, current direction alarm parameters of each automatic switch and fault type;

[0034] Multiple evaluation indicators related to the fault direction signal change in the active distribution network are determined, and the multiple evaluation indicators are weighted and calculated to determine the reward function of the fault direction signal change decision agent; wherein, the evaluation indicators include the fault direction judgment accuracy and the number of protection device malfunctions.

[0035] Optionally, the intelligent agent is a power direction acquisition, tracking, and decision-making intelligent agent;

[0036] The step of obtaining the operating state parameters of the active power distribution network and constructing a deep reinforcement learning agent based on the operating state parameters of the active power distribution network includes:

[0037] Obtain the operating status parameters of the active power distribution network;

[0038] Based on the operating status parameters, the action space and state space of the power direction acquisition and tracking change decision-making agent are determined; wherein, the action space includes the adjustment amount of the calculation threshold for positive and negative power data values ​​and the adjustment amount of the judgment threshold for positive and negative power data values, and the state space includes operating mode change parameters, real-time power direction, load curve change rate, and fault type;

[0039] Multiple evaluation indicators related to the power direction acquisition and tracking changes of the active distribution network are determined, and the reward function of the power direction acquisition and tracking change decision agent is determined by weighting the multiple evaluation indicators; wherein, the evaluation indicators include power direction tracking accuracy and voltage fluctuation.

[0040] Optionally, this method also includes:

[0041] Based on the current operating status parameters, fault type, and self-healing action decision of the active distribution network at the time of the fault, a strategy visualization analysis report is generated; the strategy visualization analysis report is used to record the execution status of the self-healing action decision and the operating status parameters after the self-healing action decision is executed.

[0042] Secondly, the present invention also provides an active distribution network self-healing adaptive decision-making system, comprising:

[0043] The fault identification module is used to respond to a fault request from the active distribution network, obtain the fault characteristics of the active distribution network, and determine the fault type of the active distribution network based on the fault characteristics.

[0044] The agent construction module is used to acquire the operating status parameters of the active power distribution network and construct a deep reinforcement learning agent based on the operating status parameters of the active power distribution network; the agent includes one or more of the following: load transfer decision agent, fault direction signal change decision agent, and power direction acquisition and tracking change decision agent;

[0045] The decision generation module is used to input the current operating status parameters and fault type of the active distribution network when a fault occurs into the intelligent agent, and to optimize the current operating status parameters and action parameters under the fault type according to the intelligent agent to obtain the self-healing action decision optimized by the intelligent agent.

[0046] Thirdly, the present invention also provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, the computer program being executed by the processor causing the processor to perform the steps of the active distribution network self-healing adaptive decision-making method as described in the first aspect.

[0047] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the active distribution network self-healing adaptive decision-making method as described in the first aspect.

[0048] As can be seen from the above technical solutions, this invention determines the fault type of an active distribution network based on its fault characteristics. Then, based on the operating state parameters of the active distribution network, it constructs a deep reinforcement learning agent. The current operating state parameters of the active distribution network at the time of the fault, along with the fault type, are input into the agent to obtain a self-healing action decision optimized by the agent, thus achieving adaptive decision-making for active distribution network faults. This method not only improves the flexibility and adaptability of active distribution network self-healing but also ensures the reliability and efficiency of the self-healing system through a continuous learning and updating mechanism, effectively guaranteeing the continuous and stable power supply of the power system. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or 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 of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is an application environment diagram of an active power distribution network self-healing and adaptive decision-making method provided in an embodiment of the present invention;

[0051] Figure 2 A flowchart of an active power distribution network self-healing adaptive decision-making method provided in an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of the fault feature extraction framework for an active power distribution network provided in an embodiment of the present invention;

[0053] Figure 4This is a schematic diagram of the DQN deep reinforcement learning algorithm flowchart provided in an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of the overall technical framework of the active power distribution network self-healing and adaptive decision-making method provided in the embodiments of the present invention;

[0055] Figure 6 This is a schematic diagram of the structure of an active power distribution network self-healing adaptive decision-making system provided in an embodiment of the present invention;

[0056] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0057] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The active distribution network self-healing adaptive decision-making method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be integrated onto server 102 or placed on a cloud or other network server. In response to a fault request from the active power distribution network, terminal 101 or server 102 acquires the fault characteristics of the active power distribution network and determines the fault type based on the fault characteristics; acquires the operating status parameters of the active power distribution network, and constructs a deep reinforcement learning agent based on these parameters; the agent includes one or more of the following: a load transfer decision agent, a fault direction signal change decision agent, and a power direction acquisition and tracking change decision agent; the current operating status parameters of the active power distribution network at the time of the fault and the fault type are input into the agent, and the agent optimizes the action parameters under the current operating status parameters and fault type to obtain the optimized self-healing action decision.

[0059] Deep reinforcement learning algorithms are machine learning methods that find optimal strategies through continuous trial and error and learning. They combine the perceptual capabilities of deep learning with the decision-making capabilities of reinforcement learning. In the active distribution network self-healing adaptive decision-making system of this invention, deep reinforcement learning algorithms are applied to construct an intelligent agent to achieve adaptive decision-making in response to active distribution network faults. The agent learns how to make optimal self-healing action decisions under different fault types and operating state parameters through interaction with the environment (i.e., the active distribution network).

[0060] The application of deep reinforcement learning algorithms in intelligent agents enables them to dynamically adjust their decision-making strategies based on the real-time operating status and fault characteristics of active power distribution networks. This dynamic adjustment capability improves the flexibility and adaptability of the self-healing system in active power distribution networks, allowing it to better cope with complex and ever-changing distribution network environments. Furthermore, due to the powerful learning capabilities of deep reinforcement learning algorithms, the agent can continuously accumulate experience from actual operation, further optimizing its decision-making strategies and improving the reliability and efficiency of self-healing.

[0061] Terminal 101 can be, but is not limited to, various personal computers, laptops, smartphones, and tablets.

[0062] Server 102 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0063] like Figure 2 As shown in the embodiments of this application, a self-healing adaptive decision-making method for active distribution networks is provided, which is applied to... Figure 1 Taking terminal 101 or server 102 as an example, the explanation includes the following steps S1 to S3. Wherein:

[0064] Step S1: In response to a fault request from the active distribution network, obtain the fault characteristics of the active distribution network and determine the fault type of the active distribution network based on the fault characteristics.

[0065] Among them, by collecting operating data such as current and voltage of active distribution networks when a fault occurs, the fault characteristics in the operating data can be extracted, which can reflect key information about the fault characteristics. Then, based on these fault characteristics, a pre-trained fault classification and prediction model can be used to accurately identify the fault type. Accurate identification of the fault type is a key step in realizing the self-healing of active distribution networks, which directly affects the formulation and implementation of subsequent self-healing strategies.

[0066] Step S2: Obtain the operating status parameters of the active distribution network, and construct a deep reinforcement learning agent based on the operating status parameters of the active distribution network; the agent includes one or more of the following: load transfer decision agent, fault direction signal change decision agent, and power direction acquisition and tracking change decision agent.

[0067] By monitoring and analyzing the real-time operating status parameters of active distribution networks, key information such as load power, distributed generation output power, and feeder remaining capacity can be obtained. These parameters reflect the current operating status and potential fault risks of the distribution network. Based on these operating status parameters, we can design corresponding action spaces and state spaces for different types of intelligent agents.

[0068] The self-healing adaptive decision-making mechanism for active distribution networks is formulated based on the fault types of the active distribution network. It analyzes three aspects: load transfer strategy, adaptive adjustment to fault direction signal changes, and adaptive adjustment to power direction acquisition and tracking changes. Then, it integrates these analyses using a deep reinforcement learning algorithm (Deep Q-Network, DQN) to optimize the self-healing strategy, thereby improving the operational stability and reliability of the distribution network. This allows for rapid response and adjustment in the event of a fault, minimizing the impact of the fault on the power grid and ensuring continuous and stable power supply to the power system.

[0069] After identifying the fault types in the active distribution network fault type analysis, each link of the load transfer strategy needs to be adjusted accordingly. An intelligent load transfer decision-making agent is formulated based on an analysis of the total transferred load, the complex grid transfer paths with multiple grid connection points, the impact of low-voltage distributed power generation on grid connection and disconnection, and the available capacity of feeders. This agent continuously optimizes its decision-making strategy through deep reinforcement learning algorithms to achieve optimal control over load transfer actions. In actual operation, the intelligent load transfer decision-making agent can comprehensively consider various factors such as load power, distributed power generation output power, and remaining feeder capacity to formulate a load transfer scheme that ensures safe grid operation while minimizing the impact of power outages. Simultaneously, it can dynamically adjust the load transfer strategy according to changes in grid operating conditions, ensuring that the self-healing system can adapt to complex and ever-changing distribution network environments, improving the grid's self-healing capability and power supply reliability.

[0070] The affected grid connection points and lines are determined based on the fault type. When calculating the optimal power transfer path, the intelligent algorithm sets the relevant grid connection points in the fault area as unavailable or reduces their weight, and prioritizes paths far away from the fault area to reduce the interference of the fault on the power transfer process. If the fault occurs in a specific area with a high density of grid connection points, a power transfer path bypassing that area is quickly planned to improve power transfer efficiency and stability.

[0071] If the fault type is related to low-voltage distributed power sources, such as power equipment failure leading to grid disconnection, more accurate predictions of load changes after disconnection are needed. By analyzing historical data on power disconnection under this type of fault and combining it with the current system operating status, power transfer strategies can be adjusted in advance to maintain system power balance and ensure stable system operation.

[0072] Assess the available capacity of feeders by considering the impact of fault types on them. For feeders whose capacity may decrease due to faults, enhanced monitoring and real-time adjustments to load distribution are necessary. For example, after a line short-circuit fault is repaired, its carrying capacity may temporarily decrease. In this case, the load transfer of the feeder should be adjusted appropriately to prevent overload.

[0073] The fault direction signal change decision agent is a key component in power systems, its function being to determine which side of the protection device a fault occurs on. Accurate fault direction signals are crucial for triggering the correct protection actions.

[0074] Automatic adjustment of directional setting parameters: Due to the different propagation characteristics of different fault types in the power grid, the abnormal locations of fault direction signals also vary. Based on the fault type, the system makes targeted adjustments to the directional alarm parameters of distribution feeder switches in areas where faults may occur. For example, for common phase-to-phase short-circuit faults, the directional alarm threshold of the switches is appropriately reduced in prone areas to improve the detection sensitivity of the fault direction, ensuring that the protection device can quickly and accurately determine the fault direction and initiate corresponding protection actions.

[0075] Automatic setting parameter verification and download: Based on common fault modes summarized in fault type analysis, the verification rules are optimized. For specific fault types, changes in relevant parameters are carefully checked. When the operating mode changes or protection settings are modified, the system automatically performs more targeted setting parameter verification to prevent protection malfunctions caused by improper parameter settings. In addition, a manual download function is retained to allow operators to intervene in special circumstances.

[0076] In a power direction acquisition and tracking change decision-making agent, power direction is a key factor determining whether protection devices such as power direction relays will operate. Based on active distribution network fault type analysis, an adaptive adjustment strategy for power direction acquisition and tracking changes is optimized. The analysis is conducted from two aspects:

[0077] Power adjustment after changes in operating mode: Different fault types cause different changes in the operating mode of the distribution network, and the changes in power direction also vary. When a fault type causes an abnormal power output on a certain line, the system can quickly identify the change in power direction and, based on the power change pattern corresponding to that fault type, accurately adjust the positive and negative values ​​of the power data to ensure the correct operation of the power direction relay.

[0078] Load curve and mode linkage: By combining the impact of different faults on load changes in fault type analysis, the relationship between load changes and operating mode changes is more accurately displayed. In planned maintenance mode, for load fluctuations that may occur due to specific fault types, it helps maintenance personnel to develop more reasonable maintenance plans and improves the distribution network's ability to cope with different faults.

[0079] Step S3: Input the current operating status parameters and fault type of the active distribution network when a fault occurs into the intelligent agent, and optimize the action parameters under the current operating status parameters and fault type according to the intelligent agent to obtain the self-healing action decision optimized by the intelligent agent.

[0080] In this system, the agent utilizes deep reinforcement learning algorithms to update its strategy based on the rewards or penalties received after receiving current operating state parameters and fault types. The reward mechanism is designed based on the effectiveness of self-healing action decisions. If the self-healing action can quickly restore the normal operation of the power distribution network and reduce the time and scope of power outages, the agent will receive a higher reward; conversely, if the self-healing action is ineffective and may lead to more serious consequences, the agent will be penalized.

[0081] After multiple iterations, the agent gradually converges to an optimal strategy, which can make optimal self-healing action decisions under different fault types and operating state parameters. This optimal self-healing action decision is used to guide the self-healing process of the active distribution network, including load transfer decisions, fault direction signal change decisions, and power direction acquisition and tracking change decisions, ensuring that the power grid can quickly and accurately restore power supply when a fault occurs.

[0082] It should be noted that this embodiment determines the fault type of the active distribution network based on its fault characteristics, and constructs a deep reinforcement learning agent based on the operating state parameters of the active distribution network. The current operating state parameters of the active distribution network at the time of the fault and the fault type are input into the agent to obtain a self-healing action decision optimized by the agent, thus achieving adaptive decision-making for active distribution network faults. This method not only improves the flexibility and adaptability of the active distribution network's self-healing but also ensures the reliability and efficiency of the self-healing system through a continuous learning and updating mechanism, effectively guaranteeing the continuous and stable power supply of the power system.

[0083] Among these features, fault type analysis not only enables the rational allocation of maintenance resources, ensuring priority attention and maintenance of key areas and critical equipment, but also identifies potential equipment problems, enabling preventative maintenance and avoiding large-scale failures. Furthermore, by reducing unnecessary repairs, it can significantly lower operation and maintenance costs.

[0084] In some embodiments, acquiring fault characteristics of an active distribution network and determining the fault type of the active distribution network based on the fault characteristics includes:

[0085] Step S101: In the event of a fault in the active distribution network, obtain the current fault waveform data of the active distribution network.

[0086] For example, after a fault occurs in an active power distribution network system, a waveform recording device installed at the low-voltage side outlet of the main transformer is used to record the three-phase voltage and current waveform data of one cycle before and after the fault as the current fault waveform recording data.

[0087] Step S102: Extract the fault features from the current fault recording data and input the fault features into the pre-trained fault classification prediction model to output the fault type corresponding to the current fault recording data of the active distribution network; wherein, the fault classification prediction model is obtained by training historical samples of fault features and historical samples of fault types based on machine learning.

[0088] Among them, the fault characteristics are determined based on the singular value statistical index and related electrical quantity characteristics of the fault waveform data. The fault characteristics include singular value statistical characteristics and electrical quantity characteristics. By inputting the fault characteristics of the current fault waveform data into the pre-trained fault classification prediction model, the fault type corresponding to the current fault waveform data of the active distribution network is output.

[0089] The fault types include, but are not limited to, short circuit faults (three-phase short circuit, two-phase short circuit, single-phase ground fault), open circuit faults, distributed power source related faults (abnormal power fluctuation, abnormal grid connection) and abnormal load faults.

[0090] In some embodiments, a fault classification prediction model is trained;

[0091] Training the fault classification and prediction model includes:

[0092] Step S111: Obtain historical fault recording data of the active distribution network.

[0093] Step S112: Extract multiple electrical quantity features and modal components from the historical fault recording data.

[0094] Among them, such as Figure 3 The feature extraction process shown in the fault waveform history data uses Karrenbauer phase mode transformation to calculate the zero-sequence voltage and zero-sequence current of the fault waveform history data for subsequent fault type feature analysis.

[0095] The formula for calculating the zero-sequence voltage is as follows:

[0096]

[0097] In the formula, It is the zero-sequence voltage. These are the instantaneous voltage values ​​of phases a, b, and c in a three-phase circuit, respectively, during normal symmetrical operation of an active distribution network. It is 0; when an asymmetrical fault occurs in an active distribution network, it will have a non-zero value.

[0098] The formula for calculating zero-sequence current is:

[0099]

[0100] In the formula, It is the zero-sequence current. These are the instantaneous current values ​​of phases a, b, and c in a three-phase circuit, during normal operation of an active distribution network. The value is 0, but a non-zero value will appear when an asymmetrical fault occurs in an active distribution network.

[0101] The Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) method is used to perform empirical mode decomposition on historical fault recording data to obtain a series of modal components of intrinsic mode functions. These modal components help to reveal potential fault characteristics in the data.

[0102] Step S113: Perform Hilbent transform on each modal component, and filter the transformed modal components to obtain filtered waveform data.

[0103] In this process, the Hilbent transform converts each modal component into a representation of its instantaneous frequency and amplitude. Subsequently, the transformed data is filtered using a bandpass filtering algorithm according to the set bandwidth (300Hz) and number of bands (10) to obtain filtered waveform data.

[0104] Step S114: Construct a time-frequency matrix based on multiple filtered waveform data.

[0105] The filtered waveform data are arranged in rows to form a time-frequency matrix, where the number of columns equals the number of sampling points per cycle and the number of rows equals the number of frequency bands.

[0106] Step S115: Perform singular value decomposition on the time-frequency matrix and obtain useful fault recording data by determining the cumulative contribution rate of the filtered waveform data.

[0107] Among them, singular value decomposition is performed on the time-frequency matrix. By retaining the first K order singular values ​​that make the cumulative contribution rate greater than 90%, useful fault signals can be extracted from the noise, which are the useful fault waveform data.

[0108] Step S116: Determine the statistical characteristics of the singular values ​​based on the singular values ​​of the useful fault recording data.

[0109] This involves calculating statistical indicators such as singular spectral kurtosis, entropy, mean, and peak factor of the singular values ​​in useful fault waveform data. These statistical indicators help to further quantify fault characteristics. To eliminate the influence of different waveform data dimensions, the singular value statistical indicators are normalized for subsequent fault analysis and classification.

[0110] The statistical characteristics of singular values ​​are as follows: singular spectrum kurtosis corresponds to the waveform changes of electrical quantities (such as voltage and current); singular spectrum entropy reflects the amount of information and complexity contained in the signal; singular spectrum mean reflects the average energy level of the fault signal; and singular spectrum peak factor reflects the peak characteristics of the signal. These singular values ​​are classified into different fault types, thus yielding different fault characteristics for different fault types.

[0111] Step S117: Determine the fault characteristics based on the statistical characteristics of singular values ​​and the characteristics of electrical quantities.

[0112] The fault characteristics include outlier statistical characteristics and electrical quantity characteristics.

[0113] Step S118: Based on the fault characteristics, determine the fault type corresponding to the fault characteristics in the preset fault database; wherein, the preset fault database includes the mapping relationship between fault characteristics and fault types.

[0114] This application classifies fault types into: short-circuit faults (three-phase short circuit, two-phase short circuit, single-phase ground fault), open-circuit faults, distributed power source-related faults (abnormal power fluctuation, abnormal grid connection), and abnormal load faults. The fault database is determined by the mapping relationship between fault characteristics and fault types, as shown in Table 1.

[0115] Table 1

[0116]

[0117] Step S119: Construct training and test datasets based on fault feature vectors and fault types.

[0118] All fault feature vectors and fault types are divided into a training dataset (70%) and a test dataset (30%). The training dataset is used to train a Support Vector Machine (SVM) model, allowing the model to learn the feature patterns corresponding to different fault types; the test dataset is used to evaluate the performance of the trained model and test its accuracy in classifying unknown fault samples.

[0119] Step S1110: Train the support vector machine using the training dataset to obtain the initial fault classification prediction model.

[0120] Support Vector Machines (SVMs) are a supervised learning algorithm primarily used for classification and regression analysis. They are very popular in machine learning, especially in pattern recognition and data mining. In classification problems, the goal of an SVM is to find a hyperplane (a line in two-dimensional space, a plane in three-dimensional space, and so on) that best separates data points from different classes. Ideally, this hyperplane completely separates the two classes and is as far away from both classes as possible, maximizing the margin between them. Support vectors are the data points closest to this hyperplane; they are the key factors determining the hyperplane's position.

[0121] Choosing the SVM classifier effectively addresses the potentially complex nonlinear relationship between fault characteristics and fault types in active distribution networks. Its expression is:

[0122]

[0123] in, This is a sign function used to determine the classification result; The number of support vectors; For Lagrange multipliers, To select the radial basis function (RBF), b is a bias term. For support vectors The corresponding category labels represent different fault types.

[0124] The radial basis kernel function is:

[0125]

[0126] In the formula, x is the input vector to be classified. These are nuclear parameters.

[0127] The training dataset is input into the SVM model. During the training process, the SVM enables the feature vectors of different fault types to be separated to the greatest extent and classified into different fault types.

[0128] Step S1111: Optimize the network parameters of the initial fault classification prediction model using the test dataset to obtain the fault classification prediction model.

[0129] The test dataset is input into the trained SVM model, which classifies each sample according to the learned classification rules and determines which type of fault it belongs to.

[0130] In some embodiments, the agent is a load transfer decision-making agent;

[0131] Obtain the operating state parameters of the active power distribution network, and construct a deep reinforcement learning agent based on these parameters, including:

[0132] Step S201: Obtain the operating status parameters of the active power distribution network.

[0133] The operating status parameters of the active distribution network include information such as real-time load power, distributed generation output power, and feeder remaining capacity.

[0134] Step S202: Determine the action space and state space of the load transfer decision-making agent based on the operating status parameters; wherein, the action space includes the combined state of the tie switch, the output power adjustment of the distributed power source, and the load shedding amount, and the state space includes the load power, the output power of the distributed power source, the remaining capacity of the feeder, and the fault type.

[0135] In the design action space, the power supply path is switched. The combined state of the tie switch is Sw. The combined state of the tie switch is a series of binary variables. Each variable represents the state of a tie switch, with 0 indicating open and 1 indicating closed.

[0136] The output power adjustment of a distributed power source is expressed as: is a continuous variable that represents the increase or decrease in the output power of the distributed power source.

[0137] Load shearing volume is expressed as It is also a continuous variable, representing the amount of load removed. Action space Recorded as:

[0138] .

[0139] The state space design considers the real-time load power Li(t), distributed generation output power Gj(t), feeder remaining capacity Ck(t), and fault type F. Load power P and distributed generation output power PG reflect the real-time supply and demand relationship of the distribution network, feeder remaining capacity C determines the distribution network's transfer capacity, and fault type F affects the formulation of self-healing decisions. The state space S(t) is represented as a vector:

[0140]

[0141] Here, F represents a defined fault type code. The corresponding value in the action space is changed based on the fault type F. For example, if F represents a short-circuit fault on a certain line, the corresponding switch combination Sw can be changed in the action space. The state space can be modified differently for different fault types F based on subsequent agent training.

[0142] The goal of a load transfer decision-making agent is to enable the distribution network to quickly restore power supply after a fault occurs by adjusting the state of tie switches, the output power of distributed generation sources, and disconnecting some loads, while ensuring the safe and stable operation of the grid. To achieve this goal, the agent needs to select the optimal action strategy based on the current state information and action space. Deep reinforcement learning algorithms play a crucial role in this process. By learning from historical data, it continuously optimizes the agent's strategy, enabling it to make optimal self-healing action decisions under different fault types and operating state parameters.

[0143] Step S203: Determine multiple evaluation indicators related to load transfer in the active distribution network, perform weighted calculations on the multiple evaluation indicators, and determine the reward function of the load transfer decision-making agent; among them, the evaluation indicators include load balance, number of overloaded feeders, and time for critical loads to restore power supply.

[0144] In the design of the reward function, the evaluation indicators are first determined. The load balance index E can be determined by the load rate of each feeder, where the feeder load rate is:

[0145]

[0146] In the formula, Let be the load power on feeder k at time t. Let be the remaining capacity of feeder k at time t. The standard deviation used to measure the load-bearing capacity of feeder k at time t is calculated as follows:

[0147]

[0148] In the formula, = , Let the average load rate of all feeders at time t be used as an intermediate parameter for calculating the standard deviation. The load balance index E is then calculated as follows:

[0149]

[0150] The larger the value of the load balance index E, the more balanced the load is.

[0151] If the feeder load rate >1 indicates overload. Let the number of overloaded feeders be Nover. The number of overloaded feeders is used to measure the severity of the overload in the power grid at time t. Eliminating the overload will improve this indicator.

[0152] The critical load restoration time (Trec) is one of the key indicators for measuring the effectiveness of load transfer, and it is used to represent the time it takes for critical loads to go from power outage to power restoration.

[0153] By weighting multiple evaluation indicators, the reward function of the load transfer decision-making agent is determined, resulting in:

[0154]

[0155] In the formula, For reward values, α, β, is a weighting coefficient, typically ranging from [0,1], used to adjust the relative importance of different evaluation indicators in the reward function, and adjusted according to actual power grid operation needs and objectives.

[0156] In some embodiments, the agent is a decision-making agent for changes in fault direction signals;

[0157] Obtain the operating state parameters of the active power distribution network, and construct a deep reinforcement learning agent based on these parameters, including:

[0158] Step S211: Obtain the operating status parameters of the active power distribution network.

[0159] The operating status parameters of the active distribution network include the real-time fault direction signal Fd(t), the network topology T, and the current direction alarm parameters Ps(t) of each switch.

[0160] Step S212: Determine the action space and state space of the fault direction signal change decision agent based on the operating status parameters; wherein, the action space includes the adjustment amount of the direction alarm threshold of the automatic switch and the adjustment amount of the fault judgment logic configuration of the protection device, and the state space includes the fault direction signal, network topology, current direction alarm parameters of each automatic switch and fault type.

[0161] In the action space design, the adjustment amount ΔPs(t) of the directional alarm threshold of the automatic switch is adjusted, where s represents the switch number, and ΔPs(t) is the adjustment amount of the directional alarm threshold of switch s at time t. The fault judgment logic Lr of the protection device is reconfigured, where Lr represents different fault judgment logic configuration schemes, each with a unique code. The action space... Recorded as:

[0162] .

[0163] Step S213: Determine multiple evaluation indicators related to the change in fault direction signal of active distribution network, perform weighted calculation on multiple evaluation indicators, and determine the reward function of the decision-making agent for the change in fault direction signal; wherein, the evaluation indicators include the accuracy of fault direction judgment and the number of times the protection device malfunctions.

[0164] In the reward design, the evaluation indicators are first determined, among which the accuracy of fault direction judgment is included. This is obtained by statistically analyzing the ratio of the number of times the fault direction was correctly identified to the total number of identifications over a period of time.

[0165]

[0166] In the formula, Ncorrect represents the number of times the fault direction was correctly determined, and Ntotal represents the total number of determinations.

[0167] The number of malfunctions of the protection device, Nmis, is the number of times the protection device malfunctions within a certain period of time, reflecting the impact of the strategy on the reliability of the protection device.

[0168] By weighting multiple evaluation indicators, the reward function of the decision-making agent for fault direction signal change is determined, resulting in:

[0169]

[0170] In the formula, α and β are weighting coefficients. α is used to adjust the importance of the accuracy of fault direction judgment in the reward, and β is used to adjust the importance of the number of malfunctions of protection devices in the reward. They can be adjusted according to the actual power grid's requirements for reliability and accuracy.

[0171] In some embodiments, the agent is a power direction acquisition and tracking change decision-making agent;

[0172] Obtain the operating state parameters of the active power distribution network, and construct a deep reinforcement learning agent based on these parameters, including:

[0173] Step S221: Obtain the operating status parameters of the active power distribution network.

[0174] The operating status parameters of an active distribution network include operating mode change parameters, real-time power direction, load curve change rate, and fault type. Operating mode change parameters describe the operating mode of the distribution network under different time periods or operating conditions, such as adjustments to the network topology and changes in power source access points, including normal operation, maintenance, and emergency operation modes. Real-time power direction reflects the power transmission between power sources and loads, serving as a crucial basis for judging the power flow distribution of the grid. The load curve change rate reflects the load's trend over time, helping to predict future load demand and providing decision support for grid dispatching and operation.

[0175] Step S222: Based on the operating status parameters, determine the action space and state space of the power direction acquisition and tracking change decision-making agent; wherein, the action space includes the adjustment amount of the calculation threshold for positive and negative power data values ​​and the adjustment amount of the judgment threshold for positive and negative power data values, and the state space includes the operating mode change parameters, real-time power direction, load curve change rate and fault type.

[0176] In the motion space design, the adjustment amount ΔTp(t) for calculating the positive and negative values ​​of power data is the adjustment amount for the threshold of judging the positive and negative values ​​of power data at time t, used to change the sensitivity of power direction judgment. The adjustment amount Ap for the threshold of judging the positive and negative values ​​of power data is configured with different algorithm parameters, such as filter coefficients and sampling period. Each configuration has a specific code, and different parameters will affect the accuracy and response speed of power direction tracking. Motion Space Recorded as:

[0177]

[0178] In state-space design, the state space is determined based on the operating mode variation parameters, real-time power direction, load curve change rate, and fault type:

[0179] S(t) = [M(t), Pd(t), Rl(t), F]

[0180] In the formula, M(t) is the operating mode change parameter at time t, Pd(t) is the real-time power direction at time t, and Rl(t) is the load curve change rate at time t.

[0181] Step S223: Determine multiple evaluation indicators related to the power direction acquisition and tracking changes in the active distribution network, perform weighted calculations on the multiple evaluation indicators, and determine the reward function of the power direction acquisition and tracking change decision-making agent; wherein, the evaluation indicators include power direction tracking accuracy and voltage fluctuation.

[0182] In the reward function design, the evaluation metric is determined as follows: Power direction tracking accuracy Apt: This is obtained by statistically analyzing the ratio of the number of times the power direction was correctly tracked within a certain period to the total number of tracking attempts.

[0183]

[0184] In the formula, It is the number of times the power direction is correctly tracked. This represents the total number of tracking attempts.

[0185] Voltage fluctuation Ss is used to characterize system stability. It quantifies the proportion of time that voltage fluctuations occur within a specified range. The closer the value is to 1, the more stable the system is.

[0186] The reward function is quantized as follows:

[0187]

[0188] In the formula, α and β are weighting coefficients (the values ​​are usually between [0,1]. α is used to adjust the importance of power direction tracking accuracy in the reward, and β is used to adjust the importance of voltage fluctuation in the reward. The values ​​are adjusted according to the actual power grid requirements for power tracking and voltage fluctuation).

[0189] Among the aforementioned agents, a unified DQN agent can be designed, and the optimal self-healing action decision can be obtained through training and optimization, including:

[0190] DQN agent construction and training:

[0191] Neural network construction: A deep neural network is used. The input layer receives the state S(t), and the output layer outputs the Q value corresponding to each action Ai(t), i.e., Q(S(t), Ai(t)).

[0192] Experience replay: Store the quadruple (S(t), A(t), R(t), S(t+1)) from each interaction into the experience replay buffer D. Randomly sample a batch of samples {(Si, Ai, Ri, Si+1)} from D for training.

[0193] Target network update: Let the Q-value of the target network output be Qtarget, and the output of the online network be Qonline. The target Q-value is calculated as follows: ,in, Discount factor ( The closer to 1, the more emphasis is placed on future rewards. A′ is one of all possible actions that can be taken at the next time step (i.e., time t+1) (the optimal action value at the next time step is considered when calculating the target Q value).

[0194] Parameter update: by minimizing the loss function as follows:

[0195]

[0196] Update the online network parameters using gradient descent.

[0197] Strategy generation and execution:

[0198] Strategy selection: based on Greedy strategy, The probability of randomly selecting an action, in order to The probability of choosing the action with the highest Q value for:

[0199]

[0200] In the formula, for Exploration probability in a greedy strategy The larger the value, the higher the likelihood of the agent taking random exploratory actions, which helps in discovering new and effective strategies; The smaller the value, the more likely the agent is to choose what it currently considers the optimal action, utilizing existing experience.

[0201] Strategy execution: Execute the selected action. Observe changes in the power grid status.

[0202] Strategy Evaluation and Optimization: The effectiveness of the strategy is evaluated based on the new state S(t+1) and reward R(t). In the load transfer strategy, evaluation is based on indicators such as load balance E, overload elimination (measured by the number of overloaded feeders, Nore), and critical load recovery time, Trec. In the adaptive adjustment of fault direction signals, evaluation is based on the fault direction judgment accuracy Af and the number of protection device malfunctions, Nmis. In the adaptive adjustment of power direction acquisition and tracking changes, evaluation is based on the power direction tracking accuracy Apt and voltage fluctuation Ss. The agent is continuously trained, and its parameters and reward function weights are adjusted based on the evaluation results to continuously optimize its strategies and improve its operational effectiveness and performance in different scenarios.

[0203] like Figure 4 As shown, by applying the DQN deep reinforcement learning algorithm, the self-healing strategy is adaptively adjusted and generated, thereby improving the intelligence level and operation performance of the distribution network.

[0204] In some embodiments, the method further includes:

[0205] Based on the current operating status parameters, fault type, and self-healing action decisions of the active distribution network at the time of a fault, a strategy visualization analysis report is generated. The strategy visualization analysis report is used to record the execution status of the self-healing action decisions and the operating status parameters after the self-healing action decisions are executed.

[0206] It's worth noting that in the intelligent operation and maintenance system of active power distribution networks, cloud-edge computing technology is closely integrated with strategy visualization and analysis, providing comprehensive and intuitive decision support for operation and maintenance personnel. The cloud-edge computing platform leverages the real-time monitoring capabilities of edge devices to collect various data from the power distribution network operation, covering key information such as real-time load power, distributed power output, feeder remaining capacity, fault direction signals, power direction, and network topology. This data is uploaded to the cloud in real time, where it undergoes in-depth analysis and integration using its powerful data processing capabilities.

[0207] Strategy visualization is the core function of this stage. Through professional visualization tools, self-healing strategies are presented to operations and maintenance personnel in an intuitive graphical way.

[0208] For example, taking load transfer strategies as an example, the transfer path can be accurately displayed on a map, with different colors and lines distinguishing normal lines from transfer lines. Real-time load data and power flow direction are also marked on the lines, allowing maintenance personnel to clearly see the load transfer in the power grid and determine whether there is an overload risk. For fault direction signal adaptive adjustment and power direction acquisition and tracking change adaptive adjustment strategies, the visualization interface will display the adjustment process of the switch direction alarm threshold, the changes in the fault judgment logic of the protection device, and the impact of changes in power direction tracking algorithm parameters on power data processing in the form of dynamic charts, helping maintenance personnel to deeply understand the execution details of the strategy.

[0209] Meanwhile, the cloud-edge computing platform automatically generates a strategy visualization analysis report. The report not only records in detail the execution status of the self-healing strategy at each stage but also provides a quantitative evaluation of the strategy's effectiveness. In the evaluation of the load transfer strategy's effectiveness, the report uses charts to display the changing trends of the load balance index, compares the changes in the number of overloaded feeders before and after strategy implementation, and accurately calculates the power restoration time for critical loads. For the fault direction signal adaptive adjustment strategy, the report uses data analysis to illustrate the improvement in fault direction judgment accuracy and the reduction in the number of malfunctions of protection devices. Regarding the power direction acquisition and tracking change adaptive adjustment strategy, the report focuses on analyzing the degree of improvement in power direction tracking accuracy and the enhancement of system stability.

[0210] Through cloud-edge computing visualization and strategy visualization analysis report generation, operation and maintenance personnel can fully grasp the execution process and effects of self-healing strategies, promptly identify potential problems, and provide strong support for subsequent strategy optimization. This not only improves the intelligent operation and maintenance level of active power distribution networks but also further enhances the self-healing capabilities of the power grid, ensuring the reliability and stability of power supply.

[0211] In summary, as Figure 5 The overall technical framework shown in this application, through a series of closely linked steps including active distribution network fault type analysis, self-healing strategy adaptive adjustment analysis, self-healing strategy adaptive adjustment generation, and cloud-edge computing visualization and strategy visualization analysis report generation, constructs a complete and efficient intelligent operation and maintenance and self-healing system for active distribution networks. Each step cooperates and operates collaboratively, from fault feature extraction and classification, to fault type-based strategy adjustment and optimization, to the generation and execution of self-healing strategies using advanced algorithms, and finally, to providing decision support for operation and maintenance personnel through visualization technology. This effectively improves the fault response speed and fault handling accuracy of active distribution networks, enhances the self-healing capability of the power grid, and provides a solid guarantee for the stable and reliable operation of the power system.

[0212] Based on the same inventive concept, this application also provides an active distribution network self-healing adaptive decision system for implementing the active distribution network self-healing adaptive decision method described above.

[0213] The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more active distribution network self-healing adaptive decision system embodiments provided below can be found in the limitations of the active distribution network self-healing adaptive decision method described above, and will not be repeated here.

[0214] like Figure 6 As shown in the figure, this application provides an active distribution network self-healing adaptive decision-making system, including:

[0215] The fault identification module 100 is used to respond to a fault request from the active distribution network, obtain the fault characteristics of the active distribution network, and determine the fault type of the active distribution network based on the fault characteristics.

[0216] The intelligent agent construction module 200 is used to acquire the operating status parameters of the active distribution network and construct a deep reinforcement learning intelligent agent based on the operating status parameters of the active distribution network; the intelligent agent includes one or more of the following: load transfer decision intelligent agent, fault direction signal change decision intelligent agent, and power direction acquisition and tracking change decision intelligent agent;

[0217] The decision generation module 300 is used to input the current operating status parameters and fault type of the active distribution network when a fault occurs into the intelligent agent, and to optimize the action parameters under the current operating status parameters and fault type based on the intelligent agent to obtain the self-healing action decision optimized by the intelligent agent.

[0218] In some embodiments, the fault identification module 100 is used for:

[0219] In the event of a fault in the active distribution network, acquire the current fault waveform data of the active distribution network;

[0220] Extract the fault features from the current fault recording data and input the fault features into a pre-trained fault classification prediction model to output the fault type corresponding to the current fault recording data of the active distribution network. The fault classification prediction model is trained on historical samples of fault features and fault types based on machine learning.

[0221] In some embodiments, the system further includes: a model training module for training a fault classification and prediction model;

[0222] Training the fault classification and prediction model includes:

[0223] Acquire historical fault recording data of active power distribution networks;

[0224] Extract multiple electrical quantity features and modal components from historical fault recording data;

[0225] Perform Hilbent transform on each modal component, and then filter the transformed modal components to obtain filtered waveform data;

[0226] A time-frequency matrix is ​​constructed based on multiple filtered waveform data;

[0227] Singular value decomposition is performed on the time-frequency matrix, and useful fault recording data is obtained by determining the cumulative contribution rate of the filtered waveform data.

[0228] Determine the statistical characteristics of singular values ​​based on the singular values ​​of useful fault recording data;

[0229] Based on the statistical characteristics of singular values ​​and the characteristics of electrical quantities, the fault characteristics are determined;

[0230] Based on the fault characteristics, the fault type corresponding to the fault characteristics is determined from a preset fault database; wherein, the preset fault database includes the mapping relationship between fault characteristics and fault types;

[0231] Based on the fault feature vectors and fault types, construct training and test datasets;

[0232] The support vector machine is trained using the training dataset to obtain the initial fault classification prediction model;

[0233] The network parameters of the initial fault classification prediction model were optimized using the test dataset to obtain the fault classification prediction model.

[0234] In some embodiments, the agent is a load transfer decision-making agent;

[0235] Intelligent agent building module 200, used for:

[0236] Obtain the operating status parameters of the active power distribution network;

[0237] Based on the operating status parameters, the action space and state space of the load transfer decision-making agent are determined; the action space includes the combined state of the tie switch, the output power adjustment of the distributed power source, and the load shedding amount, while the state space includes the load power, the output power of the distributed power source, the remaining capacity of the feeder, and the fault type.

[0238] Several evaluation indicators related to load transfer in active distribution networks are identified, and a weighted calculation is performed on these indicators to determine the reward function for the load transfer decision-making agent. The evaluation indicators include load balance, number of overloaded feeders, and time for critical loads to be restored to power.

[0239] In some embodiments, the agent is a decision-making agent for changes in fault direction signals;

[0240] Intelligent agent building module 200, used for:

[0241] Obtain the operating status parameters of the active power distribution network;

[0242] Based on the operating status parameters, the action space and state space of the decision-making agent for the change of fault direction signal are determined; the action space includes the adjustment amount of the direction alarm threshold of the automatic switch and the adjustment amount of the fault judgment logic configuration of the protection device, and the state space includes the fault direction signal, network topology, current direction alarm parameters of each automatic switch and fault type.

[0243] Multiple evaluation indicators related to the fault direction signal change in the active distribution network are identified, and the reward function of the fault direction signal change decision agent is determined by weighted calculation of the multiple evaluation indicators. Among them, the evaluation indicators include the fault direction judgment accuracy and the number of protection device malfunctions.

[0244] In some embodiments, the agent is a power direction acquisition and tracking change decision-making agent;

[0245] Intelligent agent building module 200, used for:

[0246] Obtain the operating status parameters of the active power distribution network;

[0247] Based on the operating status parameters, the action space and state space of the power direction acquisition and tracking change decision-making agent are determined. The action space includes the adjustment amount of the calculation threshold for positive and negative power data and the adjustment amount of the judgment threshold for positive and negative power data. The state space includes the operating mode change parameters, real-time power direction, load curve change rate, and fault type.

[0248] Several evaluation indicators related to power direction acquisition and tracking changes in active distribution networks are identified, and the reward function of the power direction acquisition and tracking change decision agent is determined by weighting these indicators. The evaluation indicators include power direction tracking accuracy and voltage fluctuation.

[0249] In some embodiments, the system further includes: a report generation module, used to generate a strategy visualization analysis report based on the current operating status parameters, fault type, and self-healing action decision of the active distribution network when a fault occurs; the strategy visualization analysis report is used to record the execution status of the self-healing action decision and the operating status parameters after the execution of the self-healing action decision.

[0250] like Figure 7As shown, this application provides an electronic device 10, which includes a memory 20 and a processor 30. The memory 20 stores a computer program. When the computer program is executed by the processor 30, the processor 30 performs the steps of the active power distribution network self-healing adaptive decision-making method as described in the above embodiment.

[0251] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed, implements the steps of the active distribution network self-healing adaptive decision-making method as described in the above embodiments.

[0252] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, electronic devices, and computer storage media described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0253] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0254] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0255] In the embodiments provided by this invention, it should be understood that the disclosed systems, electronic devices, computer storage media, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0256] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0257] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0258] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods described in the various embodiments of the present invention through a computer device (which may be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0259] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A self-healing adaptive decision-making method for active distribution networks, characterized in that, include: In response to a fault request from an active distribution network, the fault characteristics of the active distribution network are obtained, and the fault type of the active distribution network is determined based on the fault characteristics. The operating status parameters of the active distribution network are obtained, and a deep reinforcement learning agent is constructed based on the operating status parameters of the active distribution network; the agent includes one or more of the following: a load transfer decision agent, a fault direction signal change decision agent, and a power direction acquisition and tracking change decision agent; The current operating status parameters and fault type of the active power distribution network at the time of the fault are input into the intelligent agent, and the intelligent agent optimizes the current operating status parameters and the action parameters under the fault type to obtain the self-healing action decision optimized by the intelligent agent.

2. The active distribution network self-healing adaptive decision-making method according to claim 1, characterized in that, The step of acquiring the fault characteristics of the active distribution network and determining the fault type of the active distribution network based on the fault characteristics includes: In the event of a fault in the active distribution network, acquire the current fault waveform data of the active distribution network; Extract the fault features from the current fault recording data and input the fault features into a pre-trained fault classification prediction model to output the fault type corresponding to the current fault recording data of the active distribution network; wherein, the fault classification prediction model is obtained by training historical samples of fault features and historical samples of fault types based on machine learning.

3. The active distribution network self-healing adaptive decision-making method according to claim 2, characterized in that, Also includes: The fault classification and prediction model is trained. Training the fault classification prediction model includes: Obtain historical fault recording data of the active power distribution network; Extract multiple electrical quantity features and modal components from the fault waveform history data; Each modal component is subjected to Hilbent transform, and the transformed modal components are filtered to obtain filtered waveform data; A time-frequency matrix is ​​constructed based on the multiple filtered waveform data; Singular value decomposition is performed on the time-frequency matrix, and useful fault recording data is obtained by determining the cumulative contribution rate of the filtered waveform data. Based on the singular values ​​of the useful fault recording data, determine the statistical characteristics of the singular values; Based on the outlier statistical characteristics and the electrical quantity characteristics, the fault characteristics are determined; Based on the fault characteristics, a fault type corresponding to the fault characteristics is determined in a preset fault database; wherein, the preset fault database includes a mapping relationship between fault characteristics and fault types; Based on the fault feature vector and the fault type, a training dataset and a test dataset are constructed; The support vector machine is trained using the training dataset to obtain an initial fault classification prediction model; The network parameters of the initial fault classification prediction model are optimized using the test dataset to obtain the fault classification prediction model.

4. The active distribution network self-healing adaptive decision-making method according to claim 1, characterized in that, The intelligent agent is a load transfer decision-making intelligent agent; The step of obtaining the operating state parameters of the active power distribution network and constructing a deep reinforcement learning agent based on the operating state parameters of the active power distribution network includes: Obtain the operating status parameters of the active power distribution network; Based on the operating status parameters, the action space and state space of the load transfer decision-making agent are determined; wherein, the action space includes the combined state of the tie switch, the output power adjustment of the distributed power source, and the load shedding amount, and the state space includes the load power, the output power of the distributed power source, the remaining capacity of the feeder, and the fault type. Multiple evaluation indicators related to load transfer in the active distribution network are determined, and the reward function of the load transfer decision-making agent is determined by weighting the multiple evaluation indicators; wherein, the evaluation indicators include load balance, number of overloaded feeders, and time for critical loads to be restored to power supply.

5. The active distribution network self-healing adaptive decision-making method according to claim 1, characterized in that, The intelligent agent is a decision-making intelligent agent for changes in fault direction signals; The step of obtaining the operating state parameters of the active power distribution network and constructing a deep reinforcement learning agent based on the operating state parameters of the active power distribution network includes: Obtain the operating status parameters of the active power distribution network; Based on the operating status parameters, the action space and state space of the fault direction signal change decision agent are determined; wherein, the action space includes the adjustment amount of the direction alarm threshold of the automatic switch and the adjustment amount of the fault judgment logic configuration of the protection device, and the state space includes the fault direction signal, network topology, current direction alarm parameters of each automatic switch and fault type; Multiple evaluation indicators related to the fault direction signal change in the active distribution network are determined, and the multiple evaluation indicators are weighted and calculated to determine the reward function of the fault direction signal change decision agent; wherein, the evaluation indicators include the fault direction judgment accuracy and the number of protection device malfunctions.

6. The active distribution network self-healing adaptive decision-making method according to claim 1, characterized in that, The intelligent agent is a power direction acquisition, tracking, and decision-making intelligent agent. The step of obtaining the operating state parameters of the active power distribution network and constructing a deep reinforcement learning agent based on the operating state parameters of the active power distribution network includes: Obtain the operating status parameters of the active power distribution network; Based on the operating status parameters, the action space and state space of the power direction acquisition and tracking change decision-making agent are determined; wherein, the action space includes the adjustment amount of the calculation threshold for positive and negative power data values ​​and the adjustment amount of the judgment threshold for positive and negative power data values, and the state space includes operating mode change parameters, real-time power direction, load curve change rate, and fault type; Multiple evaluation indicators related to the power direction acquisition and tracking changes of the active distribution network are determined, and the reward function of the power direction acquisition and tracking change decision agent is determined by weighting the multiple evaluation indicators; wherein, the evaluation indicators include power direction tracking accuracy and voltage fluctuation.

7. The active distribution network self-healing adaptive decision-making method according to claim 1, characterized in that, Also includes: Based on the current operating status parameters, fault type, and self-healing action decisions of the active distribution network at the time of a fault, a strategy visualization analysis report is generated. The strategy visualization analysis report is used to record the execution status of the self-healing action decision and the running status parameters after the self-healing action decision is executed.

8. An active power distribution network self-healing adaptive decision-making system, characterized in that, include: The fault identification module is used to respond to a fault request from the active distribution network, obtain the fault characteristics of the active distribution network, and determine the fault type of the active distribution network based on the fault characteristics. The agent construction module is used to acquire the operating status parameters of the active power distribution network and construct a deep reinforcement learning agent based on the operating status parameters of the active power distribution network; the agent includes one or more of the following: load transfer decision agent, fault direction signal change decision agent, and power direction acquisition and tracking change decision agent; The decision generation module is used to input the current operating status parameters and fault type of the active distribution network when a fault occurs into the intelligent agent, and to optimize the current operating status parameters and action parameters under the fault type according to the intelligent agent to obtain the self-healing action decision optimized by the intelligent agent.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the active distribution network self-healing adaptive decision-making method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the steps of the active distribution network self-healing adaptive decision-making method as described in any one of claims 1-7.

Images