A regulation control method for early warning and positioning of distribution network faults

Abstract

The present invention discloses a regulation control method for fault early warning and positioning of distribution network, which relates to the field of distribution network automation control technology, including: based on a multi-physical field data set, combining historical data and distribution network operating conditions to perform situational awareness analysis on the current state, and identify potential anomalies; for the identified potential anomalies, further calculate the specific location of the fault through fine-grained local change analysis, and generate fault location information; using the fault location information, call a deep reinforcement learning model, and generate a fault regulation control instruction according to the load distribution and equipment health status of the distribution network; transmit the fault regulation control instruction to the automation control equipment in the power grid, execute the isolation and load redistribution of the fault area, and monitor the power grid status in real time. The present invention realizes the rapid isolation and load redistribution of the fault area of ​​the power grid, enhances the operational stability and self-healing ability of the power grid, and thus reduces the scope and time of power outages caused by faults.

Description

Technical Field

[0001] The invention relates to the technical field of distribution network automation control, and in particular to a regulation control method for distribution network fault early warning and positioning. Background Art

[0002] With the development of smart grids, the complexity and scale of power systems continue to increase, and fault diagnosis and location of distribution networks have become a key link in ensuring the safe operation of power systems. Existing distribution network fault detection and location technologies usually rely on traditional single physical quantity monitoring methods, such as voltage, current, etc. This type of method mainly detects anomalies by setting thresholds. Once the preset limit is exceeded, it is considered that a fault has occurred. However, this fault detection method based on a single data source is not accurate enough when facing a complex and changeable power grid environment. In particular, when dealing with multi-point faults, complex fault modes, and noise interference, it is often difficult to quickly and accurately identify and locate faults. Therefore, how to effectively integrate data from multiple physical quantities to achieve more refined fault identification and location has become an important challenge in existing technologies.

[0003] The main shortcoming of the existing technology is that it cannot fully utilize a variety of physical field data for fault identification and location, resulting in low fault detection accuracy and positioning precision. Traditional technology usually lacks the means to integrate and analyze multimodal data of each node in the power grid, and it is difficult to extract effective fault features from multidimensional information. This not only affects the accuracy of fault warning, but also limits the precise determination of the fault location. Especially in terms of rapid detection and location of faults, the existing methods have a slow response speed and cannot take effective fault isolation and load redistribution measures in a timely manner. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a regulation and control method for early warning and positioning of distribution network faults to solve the problem that the detection and positioning of distribution network faults in the prior art are not accurate and timely enough.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a regulation and control method for early warning and positioning of distribution network faults, which includes collecting multiple physical field data through sensors arranged on key nodes and equipment, and preprocessing the multiple physical field data through a fusion algorithm to generate a multi-physical field data set;

[0008] Based on multi-physics data sets, historical data and distribution network operating conditions are combined to conduct situational awareness analysis of the current state and identify potential anomalies;

[0009] For the identified potential anomalies, the specific location of the fault is further calculated through fine-grained local change analysis to generate fault location information;

[0010] Using the fault location information, the deep reinforcement learning model is called to generate fault adjustment control instructions based on the load distribution of the distribution network and the health status of the equipment;

[0011] Transmit fault adjustment control instructions to the automation control equipment in the power grid, perform fault area isolation and load redistribution, and monitor the power grid status in real time.

[0012] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults of the present invention, wherein: the sensors arranged on key nodes and equipment are used to collect a variety of physical field data, and the specific steps are as follows:

[0013] Analyze the topological structure of the power grid, evaluate the importance of each node through network centrality, and select key nodes as the key nodes of the power grid based on the centrality score;

[0014] The multimodal sensor array is arranged at the key nodes of the power grid. Each sensor node uses a synchronous acquisition mechanism to collect multiple physical field data and transmits them through an ultra-wideband communication protocol.

[0015] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults described in the present invention, wherein: the multiple physical field data are preprocessed by a fusion algorithm to generate a multi-physical field data set, and the specific steps are as follows:

[0016] For the collected data of various physical fields, the signals are decomposed into multi-scale components through wavelet transform;

[0017] Apply soft thresholding to multi-scale components to remove high-frequency noise;

[0018] The dynamic time warping algorithm is used to align the time series of the denoised data, and a method combining Kalman filtering and Bayesian estimation is used to perform data fusion and output a multi-physics field data set.

[0019] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults described in the present invention, wherein: based on the multi-physical field data set, the current state is context-awarely analyzed in combination with historical data and distribution network operating conditions to identify potential anomalies, the specific steps are as follows:

[0020] The spatial-temporal graph convolutional network is introduced to capture the spatial-temporal relationship between nodes in the power grid;

[0021] Based on the spatiotemporal relationship between nodes in the power grid, combined with the predicted value in Kalman filtering and multiphysics data , the noise covariance matrix in Kalman filtering Perform adaptive adjustment, the expression is:

[0022] ;

[0023] in, represents the initial noise covariance matrix, Represents the adaptive adjustment coefficient;

[0024] Compare historical data with multi-physics data through adaptively adjusted Kalman filtering to predict future states and dynamically correct deviations in current data;

[0025] By combining the deviation term between the current state and historical data and the time smoothness constraint term, the spatial state and time change of the power grid are comprehensively evaluated, and the expression is:

[0026] ;

[0027] in, represents the anomaly score of the context-aware analysis, An index variable representing the mode, Indicates the current time , modal Multiphysics data, represents the weights of different modes, Indicates the current time , modal The predicted value of Representing modality The measurement noise covariance matrix under represents the number of modes, represents the square term of the time rate of change, is the regulating factor;

[0028] Setting Adaptive Threshold , and adaptively adjust the threshold of anomaly detection through the Bayesian optimization algorithm. The Bayesian optimization expression is:

[0029] ;

[0030] in, represents the probability distribution of anomalies based on predicted values;

[0031] When situational awareness scores When a potential abnormality occurs, it will be judged that an abnormality exists, and the abnormal node will be identified through the spatiotemporal graph convolution network and Kalman filter detection, and the corresponding alarm signal will be output.

[0032] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults of the present invention, wherein: the identified potential abnormality is further calculated through fine-grained local change analysis to generate fault location information, and the specific steps are as follows:

[0033] Based on the network centrality and adjacency of abnormal nodes, the shortest path algorithm is used to divide the power grid into local areas;

[0034] After the local area is divided and locked, the multi-physics field data of the node is further analyzed, focusing on the local changes of the mode;

[0035] The local variation and spatiotemporal consistency of each node are integrated to calculate the local anomaly score and quantify the failure probability of each node;

[0036] The topological information and geographic location information of the power grid are used to map the location of the fault node to the actual geographic location to generate fault location information.

[0037] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults of the present invention, wherein: the fault location information is used to call the deep reinforcement learning model, and the fault regulation and control instructions are generated according to the load distribution of the distribution network and the health status of the equipment. The specific steps are:

[0038] Define the state space, action space, and reward mechanism in deep reinforcement learning models;

[0039] Combine state space with smart grid monitoring to obtain comprehensive operation status information of the power grid in real time

[0040] By introducing multi-task learning, the action space optimizes load distribution and protects the health status of equipment;

[0041] Design load balancing rewards, equipment protection rewards, and fault repair rewards, build a comprehensive reward function, and adaptively adjust the weights of each reward function based on changes in the power grid state;

[0042] Through repeated iterations of state space, action space, and reward mechanism, deep reinforcement learning model experience replay and Q-value update continuously optimize strategies;

[0043] In each iteration, the deep reinforcement learning model selects the best action under a given state and adjusts the strategy based on the reward after execution. After multiple iterations, the deep reinforcement learning model will generate fault adjustment control instructions based on the power grid status.

[0044] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults of the present invention, wherein: the state space includes the location information of the fault node, the load distribution of each node and the power supply status of each node;

[0045] The action space includes load redistribution, backup power switching, shedding of non-critical loads, and fault repair instructions generated and sent.

[0046] As a preferred solution of the regulation and control method for early warning and positioning of distribution network faults described in the present invention, wherein: the fault regulation and control instructions are transmitted to the automation control equipment in the power grid, the fault area is isolated and the load is redistributed, and the power grid status is monitored in real time. The specific steps are:

[0047] Standardize and encapsulate fault adjustment control instructions;

[0048] The standardized and packaged fault adjustment control instructions are transmitted to the automation control equipment through edge computing combined with 5G network;

[0049] The automation equipment determines the fault point and performs electrical isolation according to the fault isolation instruction, and operates the circuit breaker to cut off the power supply to the fault area;

[0050] After isolation, the automation equipment evaluates the load capacity and equipment health status of other nodes based on load redistribution instructions;

[0051] After load redistribution is completed, the operation status of the power grid is monitored in real time through smart sensors deployed at various nodes of the power grid.

[0052] In a second aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the regulation and control method for distribution network fault warning and positioning as described in the first aspect of the present invention is implemented.

[0053] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the regulation and control method for distribution network fault warning and positioning as described in the first aspect of the present invention is implemented.

[0054] The beneficial effects of the present invention are as follows: the present invention collects a variety of physical field data of key nodes and equipment through multimodal sensors, and adopts a fusion algorithm to pre-process the data, which can effectively improve the accuracy of fault detection; combined with situational awareness analysis and deep reinforcement learning models, it can accurately identify potential faults and precisely locate the fault location, significantly improving the accuracy of fault location; at the same time, using the generated fault adjustment control instructions, it can achieve rapid isolation and load redistribution of the power grid fault area, enhance the operating stability and self-healing ability of the power grid, and thus reduce the scope and duration of power outages caused by faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 This is a flow chart of the regulation and control method for distribution network fault early warning and positioning in Example 1.

[0057] Figure 2 This is a flow chart for collecting various physical field data in Example 1. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a regulation control method for early warning and positioning of distribution network faults, comprising the following steps:

[0062] S1. Collect various physical field data through sensors arranged on key nodes and equipment, and pre-process the various physical field data through fusion algorithms to generate multi-physical field data sets.

[0063] Furthermore, the topological structure of the power grid is analyzed, the importance of each node is evaluated through network centrality, and key nodes are selected based on the centrality scores as the key nodes of the power grid;

[0064] Among them, high centrality nodes usually refer to those located in key positions of the network, such as trunk lines or important load nodes, which can quickly propagate or receive status information from other nodes.

[0065] It should be noted that the key nodes are selected according to the centrality scores as the key nodes of the power grid. The specific implementation steps are as follows:

[0066] First, the physical structure of the power grid is abstracted into a graph model, where:

[0067] Node represents the actual physical nodes in the power grid, such as substations, power plants, and load centers;

[0068] Edge represents the power transmission line between nodes;

[0069] The graph can be an undirected graph or a directed graph, depending on the directionality of the power flow. If the grid allows bidirectional power transfer, an undirected graph is used; if there is a clear power transfer direction in the grid, a directed graph is used.

[0070] According to the actual needs of the power grid, select appropriate centrality indicators to evaluate the importance of nodes. Commonly used centrality indicators are:

[0071] Degree Centrality: Degree centrality is defined as the number of connections of a node, indicating its direct connection relationship with other nodes. The expression is:

[0072] ;

[0073] in, Representation Node The degree of (i.e., the number of edges connected to it), Represents the total number of nodes in the power grid;

[0074] It should also be noted that degree centrality is suitable for evaluating nodes that directly carry multiple connections.

[0075] Closeness Centrality: Closeness Centrality measures the average distance between a node and all other nodes, indicating the transmission efficiency of the node to other nodes. The expression is:

[0076] ;

[0077] in, Is a node and The shortest path length between

[0078] It should also be noted that nodes with higher proximity centrality can transfer power (or information) with other nodes more quickly.

[0079] Calculate the centrality score for each node in the grid based on the selected centrality metric. The following algorithms or tools can be used to calculate centrality:

[0080] Algorithm: Use breadth-first search (BFS) or Dijkstra's algorithm to calculate the shortest path (applicable to weighted graphs), and then calculate the centrality score according to the formula.

[0081] Tools: Centrality can be automatically calculated using tools such as NetworkX (a network analysis library in Python) or Gephi.

[0082] All nodes are sorted according to the centrality score. The higher the centrality score, the greater the importance of the node. Different centrality indicators can be sorted separately, or according to actual needs, multiple centrality indicators can be weighted and averaged to form a comprehensive score. The weighted expression is as follows:

[0083] ;

[0084] in, , and are the weights of degree centrality, closeness centrality, and betweenness centrality, respectively. The selection of weights depends on the specific needs of the power grid (e.g., whether the number of connections, distance efficiency, or path transit capacity is more important in the power grid topology);

[0085] According to the sorted centrality scores, several nodes with the highest scores are selected as key nodes of the power grid. These nodes undertake important power transmission tasks in the power grid or play a vital role in the topological structure. The number of key nodes can be adjusted according to the scale and monitoring requirements of the power grid.

[0086] Finally, the selected key nodes are verified in combination with the actual operation data or historical fault data of the power grid. The status of these nodes is monitored in the power grid to ensure that the criticality of these nodes is consistent with the actual operation of the power grid. If necessary, the selection of key nodes can be dynamically adjusted according to the load changes or topology changes of the actual power grid operation.

[0087] The multimodal sensor array is deployed at the key nodes of the power grid, and each sensor node adopts a synchronous acquisition mechanism;

[0088] Among them, the multimodal sensors include:

[0089] Voltage sensor: Based on the principle of electromagnetic induction, it monitors voltage fluctuations in real time.

[0090] Current sensor: Uses Hall effect sensor to accurately capture current changes.

[0091] Temperature sensor: Non-contact infrared temperature sensor for equipment surface temperature monitoring.

[0092] Vibration sensor: Use MEMS accelerometers to capture tiny vibrations of equipment and identify mechanical failures.

[0093] The synchronous acquisition mechanism collects multiple physical field data, ensuring that different physical field data are collected in real time at the same timestamp, and transmits the data efficiently and with low power consumption through the ultra-wideband (UWB) communication protocol;

[0094] Among them, multiple physical field data refers to various data obtained from different physical phenomena or fields through sensors, which reflect the multi-dimensional operating status of a system or device. Specifically, physical field data usually includes the following categories:

[0095] Electric field data: electrical parameters such as voltage and current, reflecting the status and changes of the electrical part of the power system.

[0096] Thermal field data: such as temperature, reflects the heating status of the equipment and helps monitor equipment overheating or poor cooling.

[0097] Mechanical field data: such as vibration, displacement or acceleration, reflects the mechanical movement, vibration or structural changes of the equipment, and can detect mechanical failure or abnormal movement.

[0098] Acoustic field data: such as the detection of sound waves or noise, usually used to identify abnormal sounds or abnormal sound signals inside the equipment.

[0099] Magnetic field data: such as changes in magnetic field strength, reflecting changes in the magnetic field in electromagnetic equipment or power lines.

[0100] The core of multiple physical field data is to obtain multi-dimensional operating information of equipment or systems through different types of sensors. These data complement each other and can more comprehensively reflect the health status of the system and help achieve accurate fault diagnosis and prediction.

[0101] It should be noted that UWB has the advantages of high bandwidth, low power consumption, and strong anti-interference ability, and is suitable for the complex electromagnetic environment of the power grid. Through UWB, efficient and stable transmission of multimodal data to the data processing center is ensured.

[0102] Furthermore, multiple physical field data are preprocessed by fusion algorithm to generate multi-physical field data sets. The specific steps are as follows:

[0103] For the collected data of various physical fields, the signals are decomposed into multi-scale components through wavelet transform;

[0104] Apply soft thresholding to multi-scale components to remove high-frequency noise;

[0105] The threshold selection uses the VisuShrink algorithm, which automatically calculates the optimal denoising threshold based on the standard deviation of the noise, and then reconstructs the signal, retains the main features, and outputs the denoised signal sequence. This provides cleaner data input for the subsequent timing alignment.

[0106] Since different sensors have different sampling frequencies, the dynamic time warping (DTW) algorithm is needed to align the noise-reduced data in time series;

[0107] Preferably, DTW eliminates sampling frequency differences by calculating the optimal matching path of the time series, ensuring that all physical field data are synchronized at the same time point. The aligned multimodal data lays a unified time foundation for subsequent data fusion.

[0108] After time series alignment, a method combining Kalman filtering and Bayesian estimation is used to perform data fusion and output a multi-physics field data set.

[0109] Preferably, the Kalman filter estimates the optimal state of multimodal data in real time through state prediction and observation update; the Bayesian estimation modifies the prior distribution based on historical data to generate a posteriori estimate. This fusion process integrates multi-source sensor information and outputs a high-precision multi-physics field data set.

[0110] S2. Based on multi-physics field data sets, historical data and distribution network operating conditions are combined to conduct situational awareness analysis of the current state and identify potential anomalies.

[0111] Furthermore, the spatiotemporal graph convolutional network (ST-GCN) is introduced to capture the spatiotemporal relationship between various nodes in the power grid, such as transformers and distribution stations;

[0112] The calculation expression of spatiotemporal graph convolution is:

[0113] ;

[0114] in, No. The output of the layer graph convolutional network, is the adjacency matrix, which represents the connection relationship between power grid nodes. For the The weight matrix of the layer, represents the activation function;

[0115] Based on the spatiotemporal relationship between nodes in the power grid, combined with the predicted value in Kalman filtering and multiphysics data , the noise covariance matrix in Kalman filtering Perform adaptive adjustment, the expression is:

[0116] ;

[0117] in, represents the initial noise covariance matrix, Represents the adaptive adjustment coefficient;

[0118] Preferably, in the process of Kalman filtering, state estimation relies on the comparison between the summary of historical data and the current observation value. Through the spatiotemporal graph convolutional network (ST-GCN), a more accurate representation of historical data features can be obtained. , which will improve the prediction accuracy of Kalman filter.

[0119] Compare historical data with multi-physics data through adaptively adjusted Kalman filtering to predict future states and dynamically correct deviations in current data;

[0120] The state update expression of Kalman filter is as follows:

[0121] ;

[0122] in, is the Kalman gain, is the state estimate for the next moment, Represents the residual between the observed value and the predicted value;

[0123] It should be noted that in the power grid system, It can represent the state changes of multiple physical fields such as voltage and current at a node in the power grid at a future moment. By predicting and updating the state, the health of the power grid operation can be judged more accurately and potential anomalies can be identified in a timely manner; in Kalman filtering, this difference is used to update the estimated state. If there is a large difference between the predicted and actual observed values, the filter will make a large correction to the current state estimate based on this difference; if the difference is small, it means that the prediction model is more accurate and the correction will be smaller.

[0124] By combining the deviation term between the current state and historical data and the time smoothness constraint term, the spatial state and time change of the power grid are comprehensively evaluated, and the expression is:

[0125] ;

[0126] in, represents the anomaly score of the context-aware analysis, The index variable representing the mode, Indicates the current time , modal Multiphysics data, represents the weights of different modes, Indicates the current time , modal The predicted value of Representing modality The measurement noise covariance matrix under represents the number of modes, represents the square term of the time rate of change, is the regulating factor;

[0127] Setting Adaptive Threshold , and adaptively adjust the threshold of anomaly detection through the Bayesian optimization algorithm. The Bayesian optimization expression is:

[0128] ;

[0129] in, represents the probability distribution of anomalies based on predicted values;

[0130] When situational awareness scores When a potential abnormality exists, it will be judged, and the abnormal node will be identified through the spatiotemporal graph convolution network and Kalman filter detection, and the corresponding alarm signal will be output;

[0131] The output of anomaly detection not only includes the abnormal location of the node or area, but also provides the severity and development trend of the anomaly based on the size of the anomaly score and the time change trend, thereby helping operation and maintenance personnel make corresponding decisions.

[0132] S3. For the identified potential anomalies, the specific location of the fault is further calculated through fine-grained local change analysis to generate fault location information.

[0133] Furthermore, based on the network centrality and adjacency of abnormal nodes, the shortest path algorithm is used to divide the power grid into local areas to locate the preliminary areas affected by potential faults;

[0134] Preliminary area definition: taking the abnormal node as the core and combining the network distance of the neighboring nodes , at the set threshold Locking local area ;

[0135] After the local area is divided and locked, the multi-physics field data of the node is further analyzed, focusing on the local changes of modes such as voltage, current, temperature and vibration;

[0136] Among them, the voltage change of the local node needs to be calculated by the change amount and temperature changes , used to detect local anomalies; by introducing the spatiotemporal synergy coefficient , evaluate the consistency of changes between nodes in the area, and further lock possible faulty nodes.

[0137] The local variation and spatiotemporal consistency of each node are integrated to calculate the local anomaly score and quantify the failure probability of each node;

[0138] Among them, the local anomaly score specifically refers to combining the changes in voltage and temperature and the time-space coordination coefficient to obtain a comprehensive score for each node and determine the node with the highest possibility of failure.

[0139] Using the topological information and geographic location information (GIS) of the power grid, the location of the fault node is mapped to the actual geographical location, the fault location information is generated, and it is transmitted to the power grid operation and maintenance system;

[0140] It should also be noted that in order to further improve the accuracy of fault location, acoustic sensor arrays are deployed at key nodes of the power grid to assist in positioning through acoustic signals caused by faults. The time difference of arrival (TDOA) method is used to calculate the propagation path of the sound wave and accurately locate the fault source.

[0141] Acoustic positioning formula: Calculate the location of the fault source based on the sound wave arrival time and propagation distance.

[0142] Preferably, acoustic positioning provides cross-domain innovative support for multi-physics field data analysis, further enhancing the accuracy of fault location.

[0143] S4. Using the fault location information, call the deep reinforcement learning model to generate fault adjustment control instructions based on the load distribution of the distribution network and the health status of the equipment.

[0144] Furthermore, through fault detection and location information generation, the fault node or area has been determined. Next, based on this fault location information, the improved deep reinforcement learning (DRL) model will be used to generate fault adjustment control instructions in combination with the load distribution and equipment health status of the power grid.

[0145] Define the state space, action space, and reward mechanism in deep reinforcement learning models;

[0146] Combine the state space with smart grid monitoring to obtain comprehensive operation status information of the power grid in real time, including fault location, load distribution, equipment health status and power supply. This information together constitutes the global state of the power grid at each moment and is used to support the decision-making process of the deep reinforcement learning model.

[0147] It should be noted that when the state space is input into the model, the deep reinforcement learning (DRL) model uses the strategy learned by its Q network to decide which control measures to take in the current state (i.e., the mapping from state to action). In this process, the multi-task learning module simultaneously considers load balancing, equipment protection, and fault repair efficiency to ensure that the model can comprehensively evaluate all state variables.

[0148] In this phase, the model selects appropriate actions in the action space, for example:

[0149] If the load on some nodes is too high, the model may choose to redistribute the load and transfer it to other nodes;

[0150] If some devices are in poor health, the model may choose load shedding to reduce the load on these nodes;

[0151] If insufficient power supply is detected, the model can select backup power switchover to enable backup power supply.

[0152] By introducing multi-task learning, the action space optimizes load distribution and protects the health status of equipment, improving the multi-dimensional optimization capability of control instructions;

[0153] Design load balancing rewards, equipment protection rewards, and fault repair rewards, build a comprehensive reward function, and adaptively adjust the weights of each reward function based on changes in the power grid state;

[0154] Through repeated iterations of state space, action space, and reward mechanism, deep reinforcement learning model experience replay and Q-value update continuously optimize strategies;

[0155] In each iteration, the deep reinforcement learning model selects the best action under a given state and adjusts the strategy based on the reward after execution. After multiple iterations, the deep reinforcement learning model will generate the optimal fault adjustment control instructions based on the power grid status (fault location, load distribution, equipment health, etc.).

[0156] The fault adjustment control instructions specifically include:

[0157] Load redistribution instructions: redistribute the load in the power grid to avoid overloading certain nodes;

[0158] Enable backup power command: automatically switch to backup power when main power failure is detected;

[0159] Non-critical load reduction instructions: In an emergency, reduce the load of non-critical nodes to ensure power supply to important areas;

[0160] Fault repair instructions: Explicitly instruct maintenance personnel or automatic systems to repair faulty nodes.

[0161] S5. Transmit the fault adjustment control instruction to the automation control equipment in the power grid, perform the isolation of the fault area and the load redistribution, and monitor the power grid status in real time.

[0162] Furthermore, the fault adjustment control instructions are standardized and encapsulated, and the control instructions are converted into a communication protocol that meets the grid automation standards;

[0163] Preferably, the communication protocol is widely used in smart grid communications and can ensure interoperability of commands between different devices.

[0164] The standardized and packaged fault adjustment control instructions are transmitted to the automation control equipment through edge computing combined with 5G network to ensure low latency and high reliability;

[0165] Preferably, the edge computing node can quickly process the instructions from the DRL model and reduce transmission delays. After the device confirms that it has received the instruction, it feeds back a confirmation signal to ensure correct execution.

[0166] The automation equipment determines the fault point and performs electrical isolation according to the fault isolation instruction, and operates the circuit breaker to cut off the power supply to the fault area. It is used to determine the fault location and perform corresponding electrical isolation, such as cutting off the power supply to the fault area by operating the circuit breaker or switch to prevent further spread;

[0167] After isolation, the automation equipment evaluates the load capacity and health status of other nodes based on the load redistribution instructions to decide which nodes to transfer the load of the fault area to. This process needs to consider the current load, maximum load capacity and health status of the equipment;

[0168] After the load redistribution is completed, the intelligent sensors deployed at each node of the power grid monitor the operation status of the power grid in real time, including current, voltage, power, load and equipment health status, to ensure the effectiveness of load adjustment and prevent new faults or load imbalances;

[0169] Preferably, through the deployment of intelligent sensors in the power grid, the system monitors voltage, current, load status and equipment health in real time.

[0170] This embodiment also provides a computer device, which is suitable for the case of a regulation and control method for distribution network fault warning and positioning, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the regulation and control method for distribution network fault warning and positioning proposed in the above embodiment.

[0171] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0172] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the regulation and control method for early warning and positioning of distribution network faults proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0173] In summary, the present invention collects multiple physical field data of key nodes and equipment through multimodal sensors, and adopts fusion algorithm to pre-process the data, which can effectively improve the accuracy of fault detection; combined with situational awareness analysis and deep reinforcement learning model, it can accurately identify potential faults and precisely locate the fault location, significantly improving the accuracy of fault location; at the same time, using the generated fault adjustment control instructions, it can achieve rapid isolation and load redistribution of the fault area of ​​the power grid, enhance the operating stability and self-healing ability of the power grid, and thus reduce the scope and duration of power outages caused by faults.

[0174] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of the regulation and control method for early warning and positioning of distribution network faults are provided.

[0175] The experimental object is a medium-voltage distribution network in a city, where a variety of sensors are deployed to collect various physical field data such as electrical parameters, temperature, vibration, and power flow. The sensors are deployed at key nodes in the network, which are selected through the centrality evaluation method and cover substations, switch stations, and terminal load points. The sensors at each node use ultra-wideband communication protocols to transmit the collected data to the central processing system.

[0176] First, the topology of the power grid was analyzed, and 10 key nodes with high centrality were selected, and multimodal sensors were installed on each node. These sensors can synchronously collect multiple physical field data such as current, voltage, temperature, and vibration. The data is initially processed by the edge computing node and then uploaded to the central control system. In the experiment, the collected multi-physical field data was decomposed using wavelet transform to extract the multi-scale components of each physical quantity. At the same time, the soft threshold method was used to remove high-frequency noise to ensure the accuracy of the data.

[0177] Secondly, the fusion processing of multi-physics field data was carried out. The method of combining Kalman filtering and Bayesian estimation was used to align and fuse the noise-reduced data in time series to generate a multi-physics field data set. Based on this data set, combined with historical data and the operating conditions of the distribution network, the system conducted a situational awareness analysis of the current state of the distribution network, captured the spatiotemporal relationship between nodes through the spatiotemporal graph convolutional network, and identified potential abnormal states.

[0178] Furthermore, after the abnormal state is identified, the system further performs fine-grained local change analysis, divides the power grid into local areas based on the network centrality and adjacency of abnormal nodes, and finally locates the fault location using the shortest path algorithm.

[0179] Finally, after the fault location was completed, the system called the deep reinforcement learning model, which generated fault adjustment control instructions based on the load distribution of the distribution network and the health status of the equipment. The adjustment control instructions were transmitted to the automation control equipment through the 5G network, and the fault area was isolated and the load was redistributed. The entire process monitored the operation status of the power grid in real time to ensure that fault isolation and load redistribution could be carried out smoothly.

[0180] The details are shown in Table 1 below:

[0181] Table 1 Experimental comparison table

[0182] Parameter name Existing technical solutions Solution 1 of the present invention Solution 2 of the present invention Solution 3 of the present invention Fault detection time (seconds) 12.5 8.2 7.9 7.5 Fault location error (meter) 30 15 12 10 Fault isolation time (seconds) 9.5 5.5 5.2 5.0 Load redistribution time (seconds) 15.0 10.2 9.8 9.5 System recovery stability (minutes) 7.2 4.8 4.5 4.3

[0183] Through the analysis of experimental data, it is obvious that the method of the present invention is superior to the prior art in multiple key performance indicators. First, the fault detection time is significantly reduced in multiple implementation schemes of the present invention. The fault detection time of the prior art is 12.5 seconds on average, while Scheme 1, Scheme 2 and Scheme 3 of the present invention reduce the fault detection time to 8.2 seconds, 7.9 seconds and 7.5 seconds respectively. This significant improvement stems from the fact that the present invention can identify potential anomalies more quickly and accurately through multi-physics field data fusion and situational awareness analysis.

[0184] Secondly, the accuracy of fault location is greatly improved. The average fault location error of the prior art is 30 meters, while the location errors of solutions 1, 2 and 3 of the present invention are 15 meters, 12 meters and 10 meters respectively. This is due to the combined application of Kalman filtering, spatiotemporal graph convolutional network and local change analysis, which greatly improves the accuracy of fault location, thereby reducing power grid maintenance time and fault repair costs.

[0185] In terms of fault isolation time, the present invention also shows significant advantages. The fault isolation time of the prior art is 9.5 seconds, while the solution of the present invention shortens it to 5.5 seconds, 5.2 seconds and 5.0 seconds. This is because the fault adjustment control instructions generated by the deep reinforcement learning model of the present invention are more accurate, and can execute isolation actions faster, avoiding unnecessary delays.

[0186] The comparison of load redistribution time also shows the advantages of the present invention. The load redistribution time of the prior art is 15.0 seconds, while the various solutions of the present invention shorten this time to 10.2 seconds, 9.8 seconds and 9.5 seconds. This is because the present invention makes the load redistribution process more efficient and reduces the risk of secondary imbalance of the power system through real-time monitoring and optimization adjustment of the distribution network load.

[0187] Finally, in terms of system recovery stability, the solution of the present invention is significantly better than the prior art. The system recovery time of the prior art is 7.2 minutes, while the solution of the present invention shortens the system recovery time to 4.8 minutes, 4.5 minutes, and 4.3 minutes. This shows that the present invention can not only respond quickly when a fault occurs, but also restore the normal operation of the system more quickly. This advantage is attributed to the combined use of fusion processing of multi-physics field data, situational awareness analysis, and deep reinforcement learning, which greatly improves the system's response and recovery capabilities to faults.

[0188] It should also be noted that in traditional distribution network fault detection and location technologies, the main fault detection methods used are those based on single physical field data, such as current and voltage abnormal fluctuation detection. These technologies usually rely on simple threshold analysis or frequency domain analysis methods such as Fourier transform to identify the characteristics of fault signals. However, this method has the following limitations:

[0189] Limitations of a single data source: Existing technologies usually rely on only a single type of physical field data (such as current or voltage). In a complex distribution network, a single data source is difficult to fully reflect the overall state of the system and is prone to miss some early abnormal signals, resulting in delayed fault detection.

[0190] Large noise interference: Due to the complex operating environment of the power grid, electromagnetic interference, environmental noise and other factors often affect the accuracy of sensor data. Existing technologies are relatively limited in processing noise, and usually can only remove more obvious noise through simple filtering methods, and cannot effectively deal with high-frequency noise or complex background noise.

[0191] Low detection and positioning accuracy: Existing fault location technologies mostly rely on simple distance protection, differential protection and other methods. These methods can usually only provide relatively rough fault location. Especially in distributed power grids or complex network structures, the positioning accuracy is low, and the fault location error usually reaches tens of meters or even higher.

[0192] Slow fault handling: After a fault occurs, existing technologies mostly use manual intervention or traditional automated control equipment to isolate the fault and redistribute the load. The operation cycle is long and usually requires multiple steps to gradually isolate the fault area and manually adjust the load distribution. This processing method is not only time-consuming, but also prone to human operation errors or delayed system response.

[0193] Lack of intelligent adjustment means: Existing technologies mostly use preset rules or fixed logic control in fault adjustment control, which makes it difficult to make dynamic adjustments based on the real-time status of the power grid and load distribution. The real-time changes in equipment health status and load distribution are not fully considered, resulting in insufficient flexibility and adaptability in fault handling.

[0194] In summary, although the existing technical solutions can achieve basic fault detection and fault area isolation, they have obvious deficiencies in detection accuracy, positioning accuracy, fault processing speed and intelligent adjustment capabilities. The present invention significantly improves the performance of fault detection and processing through technical means such as multi-physical field data fusion, situational awareness analysis, fine-grained local change analysis, and deep reinforcement learning. It can more efficiently deal with faults in complex power grids and achieve rapid recovery of the power grid after a fault through intelligent adjustment means.

[0195] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A regulation and control method for early warning and positioning of distribution network faults, characterized in that: include, Collect multiple physical field data through sensors placed on key nodes and equipment, and pre-process the multiple physical field data through fusion algorithms to generate multi-physical field data sets; Based on multi-physics data sets, historical data and distribution network operating conditions are combined to conduct situational awareness analysis of the current state and identify potential anomalies; For the identified potential anomalies, the specific location of the fault is further calculated through fine-grained local change analysis to generate fault location information; Using the fault location information, the deep reinforcement learning model is called to generate fault adjustment control instructions based on the load distribution of the distribution network and the health status of the equipment; Transmit fault adjustment control instructions to the automation control equipment in the power grid, perform fault area isolation and load redistribution, and monitor the power grid status in real time; The method is based on a multi-physics field data set, combines historical data and distribution network operation conditions to perform situational awareness analysis on the current state and identify potential anomalies. The specific steps are: The spatial-temporal graph convolutional network is introduced to capture the spatial-temporal relationship between nodes in the power grid; Based on the spatiotemporal relationship between nodes in the power grid, combined with the predicted value in Kalman filtering and multiphysics data , the noise covariance matrix in Kalman filtering Perform adaptive adjustment, the expression is: ； in, represents the initial noise covariance matrix, Represents the adaptive adjustment coefficient; Compare historical data with multi-physics data through adaptively adjusted Kalman filtering to predict future states and dynamically correct deviations in current data; By combining the deviation term between the current state and historical data and the time smoothness constraint term, the spatial state and time change of the power grid are comprehensively evaluated, and the expression is: ； in, represents the anomaly score of the context-aware analysis, An index variable representing the mode, Indicates the current time , modal Multiphysics data, represents the weights of different modes, Indicates the current time , modal The predicted value of Representing modality The measurement noise covariance matrix under represents the number of modes, represents the square term of the time rate of change, is the regulating factor; Setting Adaptive Threshold , and adaptively adjust the threshold of anomaly detection through the Bayesian optimization algorithm. The Bayesian optimization expression is: ； in, represents the probability distribution of anomalies based on predicted values; When situational awareness scores When a potential abnormality occurs, it will be judged that an abnormality exists, and the abnormal node will be identified through the spatiotemporal graph convolution network and Kalman filter detection, and the corresponding alarm signal will be output.

2. The regulation and control method for early warning and positioning of distribution network faults according to claim 1, characterized in that: The specific steps of collecting various physical field data by sensors arranged on key nodes and equipment are as follows: Analyze the topological structure of the power grid, evaluate the importance of each node through network centrality, and select key nodes as the key nodes of the power grid based on the centrality score; The multimodal sensor array is arranged at the key nodes of the power grid. Each sensor node uses a synchronous acquisition mechanism to collect multiple physical field data and transmits them through an ultra-wideband communication protocol.

3. The regulation and control method for early warning and positioning of distribution network faults according to claim 2, characterized in that: The multiple physical field data are preprocessed by a fusion algorithm to generate a multi-physical field data set. The specific steps are: For the collected data of various physical fields, the signals are decomposed into multi-scale components through wavelet transform; Apply soft thresholding to multi-scale components to remove high-frequency noise; The dynamic time warping algorithm is used to align the time series of the denoised data, and a method combining Kalman filtering and Bayesian estimation is used to perform data fusion and output a multi-physics field data set.

4. The regulation and control method for early warning and positioning of distribution network faults according to claim 3, characterized in that: The identified potential anomaly is further calculated through fine-grained local change analysis to generate fault location information. The specific steps are: Based on the network centrality and adjacency of abnormal nodes, the shortest path algorithm is used to divide the power grid into local areas; After the local area is divided and locked, the multi-physics field data of the node is further analyzed, focusing on the local changes of the mode; The local variation and spatiotemporal consistency of each node are integrated to calculate the local anomaly score and quantify the failure probability of each node; The topological information and geographic location information of the power grid are used to map the location of the fault node to the actual geographic location to generate fault location information.

5. The regulation and control method for early warning and positioning of distribution network faults according to claim 4, characterized in that: The method utilizes the fault location information to call the deep reinforcement learning model and generates fault adjustment control instructions according to the load distribution of the distribution network and the health status of the equipment. The specific steps are: Define the state space, action space, and reward mechanism in deep reinforcement learning models; Combine state space with smart grid monitoring to obtain comprehensive operation status information of the power grid in real time By introducing multi-task learning, the action space optimizes load distribution and protects the health status of equipment; Design load balancing rewards, equipment protection rewards, and fault repair rewards, build a comprehensive reward function, and adaptively adjust the weights of each reward function based on changes in the power grid state; Through repeated iterations of state space, action space, and reward mechanism, deep reinforcement learning model experience replay and Q-value update continuously optimize strategies; In each iteration, the deep reinforcement learning model selects the best action under a given state and adjusts the strategy based on the reward after execution. After multiple iterations, the deep reinforcement learning model will generate fault adjustment control instructions based on the power grid status.

6. The regulation and control method for early warning and positioning of distribution network faults according to claim 5, characterized in that: The state space includes location information of the faulty node, load distribution of each node and power supply status of each node; The action space This includes load redistribution, backup power switching, shedding of non-critical loads, and generation and sending of fault repair instructions.

7. The regulation and control method for early warning and positioning of distribution network faults according to claim 6, characterized in that: The fault adjustment control instruction is transmitted to the automatic control device in the power grid, the fault area is isolated and the load is redistributed, and the power grid status is monitored in real time. The specific steps are: Standardize and encapsulate fault adjustment control instructions; The standardized and packaged fault adjustment control instructions are transmitted to the automation control equipment through edge computing combined with 5G network; The automation equipment determines the fault point and performs electrical isolation according to the fault isolation instruction, and operates the circuit breaker to cut off the power supply to the fault area; After isolation, the automation equipment evaluates the load capacity and equipment health status of other nodes based on load redistribution instructions; After load redistribution is completed, the operation status of the power grid is monitored in real time through smart sensors deployed at various nodes of the power grid.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the regulation and control method for distribution network fault early warning and positioning as described in any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the regulation and control method for distribution network fault early warning and positioning as described in any one of claims 1 to 7 are implemented.

Images