Fault self-healing control method for preventing override trip of intelligent power distribution network

Abstract

The invention relates to the technical field of intelligent power distribution networks, and provides a fault self-healing control method for preventing override trip of an intelligent power distribution network. The method comprises the following steps: S1, acquiring data of key nodes of a power distribution network in real time through multiple sensors, and performing fusion processing on edge computing nodes to realize anomaly detection and preliminary fault classification; and S2, performing fault trend prediction on the operation state of the power distribution network based on the deep learning model to obtain a fault risk score. By adopting the technical scheme of multi-sensor real-time data acquisition and edge computing node fusion processing, the technical effects of early fault detection and preliminary fault classification are realized; according to the method, the accuracy and response speed of fault detection are improved through multi-source data fusion and edge calculation, and the problems of inaccurate detection and untimely response caused by data noise and delay in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent distribution network, and in particular to a fault self-healing control method for preventing over-tripping of an intelligent distribution network. Background Art

[0002] As the scale of distribution networks continues to expand and their structures become increasingly complex, the risk of failures in distribution networks during operation increases. In particular, due to unstable power loads, network communication delays, and complex grid structures, over-tripping often occurs, leading to large-scale power outages and reduced power safety. Therefore, in order to improve the operational stability and fault resistance of distribution networks, fault self-healing control technology for smart distribution networks has become a hot topic of research. However, existing technologies still have deficiencies in many aspects.

[0003] Existing technologies usually rely on a single sensor to monitor key node data, lacking the integration of multi-dimensional and multi-source data. This single sensor monitoring method is easily affected by noise interference in the early stage of a fault, resulting in untimely fault detection or misjudgment. Existing fault warning technologies for smart distribution networks are often based on set rules and thresholds for judgment, lacking predictability. The rule and threshold method tends to ignore the trend information of fault development and cannot identify the potential risks of faults in advance, resulting in the system being difficult to respond in time when a fault occurs. Existing technologies have a slow response speed in fault isolation and topology reconstruction, and lack coordination when dealing with complex network structures, making it difficult to ensure power supply restoration in non-fault areas. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a fault self-healing control method for preventing over-tripping in an intelligent distribution network, which solves the problems of delayed fault detection response, inflexible protection devices, and slow fault isolation and power supply recovery in the intelligent distribution network.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A fault self-healing control method for preventing over-tripping in a smart distribution network comprises the following steps;

[0006] S1. Collect data from key nodes of the distribution network in real time through multiple sensors, and integrate and process them on edge computing nodes to achieve anomaly detection and preliminary fault classification;

[0007] S2. Predict the fault trend of the distribution network operation status based on the deep learning model and obtain the fault risk score;

[0008] S3: Dynamically adjust the action time and current threshold of the protection device according to the load conditions, and conduct adaptive protection coordination to ensure fault isolation within the minimum range and prevent over-tripping;

[0009] S4, quickly isolate the fault area through distributed fault isolation technology, and use graph algorithms to reconstruct the power grid topology and restore power supply to non-fault areas;

[0010] S5. Load reconstruction and distributed energy management based on a two-layer optimization model to ensure load balance in non-fault areas;

[0011] S6. Establish a multi-channel communication network between nodes to ensure low latency and high reliability of data transmission and achieve synchronization and coordinated control of fault information.

[0012] Preferably, the step S1 specifically includes:

[0013] S1.1. Install current sensors, voltage sensors, temperature sensors and vibration sensors at key nodes of the distribution network to collect current data, voltage data, temperature data and vibration data in real time;

[0014] S1.2. The collected data is transmitted to the edge computing node, and the multi-source data is fused using the Kalman filter algorithm to reduce data noise and improve data accuracy;

[0015] S1.3, perform anomaly detection on the fused data using the isolation forest algorithm, calculate the anomaly score, and determine whether there is abnormal data based on the score result;

[0016] S1.4. When an abnormality is detected, a preliminary fault classification is performed on the abnormal data based on the current, voltage, temperature and vibration characteristics to determine whether the fault type is a short circuit, overload or ground fault.

[0017] Preferably, the step S2 specifically includes:

[0018] S2.1. Standardize the historical and real-time collected current, voltage and load data;

[0019] S2.2, using the long short-term memory network (LSTM) model to predict the fault trend of the standardized time series data;

[0020] S2.3. Compare the prediction results of the LSTM model with the actual measured voltage and current values, and calculate the fault risk score R(t) = α·|V pred -V real |+β·|I pred -I real |;

[0021] S2.4. Set a fault risk score threshold and issue a fault warning signal when the threshold is exceeded, providing a warning basis for subsequent protection coordination.

[0022] Preferably, the action time and current threshold of the protection device in step S3 are dynamically adjusted according to the real-time load conditions, and the adjustment formula of the action time is:

[0023]

[0024] Among them, T0 is the initial action time, L m ax is the maximum load of the system, L(t) is the current load value, and γ is the adjustment coefficient.

[0025] Preferably, the step optimizes the action time and current threshold of the protection device through an adaptive protection algorithm. The adaptive protection algorithm in S3 performs reinforcement learning based on a Q-Learning algorithm, and the update formula of the Q function is:

[0026]

[0027] Among them, α is the learning rate, γ is the discount factor, and r t For immediate rewards.

[0028] Preferably, the distributed fault isolation technology in step S4 is implemented based on the adjacency tensor method, by generating an adjacency tensor And the connection edge weight of the fault area is set to infinity to disconnect the faulty node and isolate the faulty area.

[0029] Preferably, the topology reconstruction in step S4 adopts a graph-based shortest path algorithm to optimize the power supply path to restore power supply to the non-fault area. The optimization formula of the shortest path is:

[0030] d ij =min(d ij ,d ik +w kj )

[0031] Among them, d ij represents the shortest path distance from node i to node j, w k j is the weight between nodes k and j.

[0032] Preferably, the steps are based on the fault self-healing control method for preventing over-tripping in the intelligent distribution network according to claim 1, wherein load reconstruction and distributed energy management are implemented using a two-layer optimization model.

[0033] The optimization goal of the upper model is to improve the load balance of the distribution network by minimizing the power imbalance between load nodes.

[0034] The standard function is:

[0035]

[0036] Among them, P i,供电 represents the actual power supply of node i, D i ,demand represents the load demand power of node i, and N is the total number of load nodes in the distribution network;

[0037] The lower model is used to manage the output of distributed energy so that distributed energy can be optimally allocated according to load demand to improve the self-healing ability of the power grid. The optimization objective function of the lower model is:

[0038]

[0039] in, represents the output power of distributed energy node j, β j is the coefficient of distributed energy satisfying load demand, D j,需求 is the load demand power in the area where the distributed energy node j is located, and M is the total number of distributed energy nodes.

[0040] Preferably, in step S5, the load reconstruction and distributed energy management are solved by Lagrangian relaxation method, and the two-layer optimization model is converted into a single-layer optimization problem by relaxation method to achieve coordinated optimization of load reconstruction and distributed energy.

[0041] Preferably, the communication nodes of the multi-channel communication network in step S6 adopt a 5G and optical fiber dual-channel structure to ensure low latency of data transmission. The IEEE 1588 precision time protocol is used to achieve time synchronization, and the security and integrity of fault data during transmission are ensured based on blockchain technology.

[0042] The present invention provides a fault self-healing control method for preventing over-tripping in an intelligent distribution network. It has the following beneficial effects:

[0043] 1. The present invention achieves the technical effect of early fault detection and preliminary fault classification by adopting the technical solution of multi-sensor real-time data acquisition and edge computing node fusion processing. Compared with the single sensor data acquisition method in the prior art, the present invention improves the accuracy and response speed of fault detection through multi-source data fusion and edge computing, and solves the problems of inaccurate detection and untimely response caused by data noise and delay in the prior art.

[0044] 2. The present invention achieves the technical effect of quickly isolating the fault area and restoring power supply to the non-fault area by adopting the technical solution of distributed fault isolation and topology reconstruction. Compared with the fault isolation method that relies on centralized control in the prior art, the present invention realizes distributed rapid isolation and topology reconstruction through the adjacency tensor method and the shortest path algorithm, solving the problem of long power supply restoration time caused by centralized control delay or fault extension in the prior art.

[0045] 3. The present invention achieves the technical effect of dynamic load balancing in non-faulty areas by adopting a load reconstruction and distributed energy management technical solution based on a two-layer optimization model. Compared with the simple load reconstruction method in the prior art, the present invention optimizes the power distribution of load nodes and the output of distributed energy respectively through a two-layer optimization model, solving the problem of unstable power supply caused by unbalanced load distribution in the prior art.

[0046] 4. The present invention achieves the technical effect of real-time assessment of fault risk and early warning by adopting a fault trend prediction technical solution based on a deep learning model. Compared with the method of judging fault trends by relying on rules or thresholds in the prior art, the present invention utilizes the time series processing capability of the LSTM network to improve the accuracy and advance amount of fault trend prediction, and solves the problem of inability to perform effective preventive maintenance due to insufficient prediction in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Example:

[0050] Please see attached Figure 1 , an embodiment of the present invention provides a fault self-healing control method for preventing over-tripping in a smart distribution network, comprising the following steps;

[0051] S1. Collect data from key nodes of the distribution network in real time through multiple sensors, and integrate and process them on edge computing nodes to achieve anomaly detection and preliminary fault classification;

[0052] S2. Predict the fault trend of the distribution network operation status based on the deep learning model and obtain the fault risk score;

[0053] S3: Dynamically adjust the action time and current threshold of the protection device according to the load conditions, and conduct adaptive protection coordination to ensure fault isolation within the minimum range and prevent over-tripping;

[0054] S4, quickly isolate the fault area through distributed fault isolation technology, and use graph algorithms to reconstruct the power grid topology and restore power supply to non-fault areas;

[0055] S5. Load reconstruction and distributed energy management based on a two-layer optimization model to ensure load balance in non-fault areas;

[0056] S6. Establish a multi-channel communication network between nodes to ensure low latency and high reliability of data transmission and achieve synchronization and coordinated control of fault information.

[0057] In this embodiment, the real-time data of the power grid is monitored by installing various types of sensors at key nodes of the distribution network. The installed sensors include current sensors, voltage sensors, temperature sensors and vibration sensors, which are used to collect current, voltage, temperature and vibration data respectively. These data are collected in real time and transmitted to edge computing nodes through the Internet of Things technology. The edge computing nodes are close to the data collection source and have localized data processing capabilities. They can pre-process the sensor data in the first place and provide real-time data support for subsequent abnormality detection and fault classification.

[0058] Specifically, the collected multi-source data is fused and processed on the edge computing node through the Kalman filter algorithm. Kalman filter is a recursive estimation algorithm that can smooth noise interference during the data update process, making the fusion effect of different sensor data more accurate. The specific calculation process is as follows;

[0059] State prediction: predict the system state at the current moment, the calculation formula is as follows;

[0060] x(t|t-1)=Ax(t-1)+w(t-1)

[0061] Among them, x(t|t-1) represents the predicted state vector at time t, a is the state transfer matrix, which describes the dependency of the current state on the previous state, w(t-1) is the process noise vector, 2. Observation update: After obtaining the observation data, the state is corrected according to the difference between the predicted state and the observed data. The calculation formula for observation update is:

[0062] z(t)=Hx(t)+v(t)

[0063] Among them, z(t) is the observation data vector, H is the observation matrix, which describes the relationship between the observation data and the state, and v(t) is the observation noise. The sensor data is fused through Kalman filtering to remove noise and improve data accuracy, providing accurate data input for subsequent anomaly detection;

[0064] By fusing sensor data through Kalman filtering, noise can be removed and data accuracy can be improved, providing accurate data input for subsequent anomaly detection;

[0065] In this embodiment, the fused data will be further subjected to anomaly detection by the isolation forest algorithm. The isolation forest algorithm is a tree-structured anomaly detection algorithm that calculates anomaly scores by randomly dividing the data space and generating isolated trees. The calculation steps of the algorithm are as follows:

[0066] First, multiple isolation trees are generated for each data sample, and the path length of the sample is calculated in each tree;

[0067] Then, the average path length of the samples is calculated and the anomaly score is calculated according to the following formula;

[0068]

[0069] Among them, s(x) represents the anomaly score of sample x, E(h(x)) represents the average path length of the sample in the isolated tree, c(n) is the normalization factor, and n is the number of samples. The anomaly score is used to determine whether the sample is abnormal. If the anomaly score exceeds the set threshold, the sample is marked as abnormal data and fault classification is performed. In this embodiment, after the anomaly is detected, preliminary fault classification is performed based on multi-dimensional features such as current, voltage, temperature, and vibration to specifically determine whether the fault type is a short circuit, overload, or ground fault. Through this multi-dimensional data and classification logic, the accuracy of preliminary fault classification is improved;

[0070] In this embodiment, through multi-sensor data collection, edge computing node data fusion and anomaly detection, fault signals can be identified in real time and preliminarily classified at the early stage of the fault. The Kalman filter algorithm improves data accuracy by fusion processing of multi-source data, and the anomaly detection of the isolation forest algorithm can effectively identify the initial signs of fault occurrence.

[0071] In this embodiment, the current, voltage and load data collected historically and in real time from the distribution network are first standardized to eliminate the differences in different data dimensions and ensure that the model can be calculated within a unified numerical range. This step is an important part of data preprocessing. The formula for standardization is as follows:

[0072]

[0073] Among them, X is the original data, μ is the mean, σ is the standard deviation, X ′ For standardized data, the standardization process can set the mean of the data to zero and the standard deviation to one, thereby ensuring that different eigenvalues ​​have similar dimensions when input to the model, which helps to improve the convergence speed and prediction accuracy of the deep learning model.

[0074] In this embodiment, a long short-term memory (LSTM) network is used as the core model for fault trend prediction. The LSTM network is a deep learning model based on a recurrent neural network (RNN) structure. It has strong time series processing capabilities and is suitable for analyzing the time series characteristics contained in distribution network data. The LSTM network can memorize and associate historical data at different time steps through its internal forget gate, input gate, output gate and other mechanisms, thereby generating accurate predictions of future states. The core calculation steps of the LSTM network include:

[0075] Forget the Gate;

[0076] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0077] Among them, f t is the output value of the forget gate, indicating the proportion of information retained; W f is the weight matrix, h t-1 is the hidden state at the previous moment, x t is the current input data, b f is the bias term.

[0078] Input gate;

[0079] The input gate is used to update the memory information at the current moment by controlling the degree of fusion of the new input data and the state at the previous moment. The formula is:

[0080] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0081] Among them, i t is the output value of the input gate, indicating the weight ratio of the new information; W i is the weight matrix, b i is the bias term

[0082] Unit Status Update

[0083] The current system state is updated through the candidate memory unit, and the formula is as follows:

[0084]

[0085] in, is the candidate memory state, C t is the updated memory state, W C is the weight matrix, C t-1 is the unit state at the previous moment, b C is the bias term, and the unit state update mechanism updates the memory state by combining the information of the previous moment and the current input information;

[0086] Output gate;

[0087] The output gate selects part of the current memory state as output to predict the state at the next moment. The formula is as follows

[0088] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0089] h t =o t *tanh(C t )

[0090] Among them, t is the output value of the output gate, which controls the output ratio of the memory state; h t It is a hidden state, representing the output information at the current moment. After completing the above-mentioned time series data processing, the LSTM model can output the predicted current and voltage trend values. The predicted results are compared with the actual values ​​measured in real time to calculate the fault risk score. In this embodiment, the fault risk score is used to assess the risk level of the current distribution network state. By comparing the deviation between the predicted value and the actual value, it can be determined whether the power grid has potential fault hazards. The calculation formula for the fault risk score is as follows:

[0091] R(t)=α·|V pred -V real |+β·|I pred -I real |

[0092] Among them, V pred and I pred are the voltage and current values ​​predicted by the LSTM model, V real and I real is the actual measured value, α and β are weight factors used to balance the importance of voltage and current in risk scoring. According to the fault risk score, the system sets the risk threshold. When R(t) exceeds the threshold, the system will issue a fault warning signal;

[0093] In this embodiment, through the deep learning model fault trend prediction and risk score calculation, the system can identify potential risks before a fault occurs, thereby providing a reliable early warning basis for subsequent protection coordination. The LSTM network improves the accuracy of fault trend prediction based on the long and short-term memory characteristics of time series data, and the fault risk score provides a means for the system to quantify risks, enabling the system to achieve more flexible dynamic protection.

[0094] In this embodiment, the action time T of the protection device is first adjusted according to the real-time load data of the distribution network. trip and the current threshold I trip , to adapt to different load conditions. This dynamic adjustment process aims to optimize the response parameters of the protection device according to the current load level of the system, ensuring that the protection device can quickly and accurately isolate the fault when a fault occurs;

[0095] Dynamic adjustment of protection equipment parameters;

[0096] Action time of protection device T trip Dynamically adjust according to load conditions to reduce action delay when the load is large, or extend the action time when the load is small, thereby avoiding unnecessary frequent operations. The action time adjustment formula is:

[0097]

[0098] Among them, T0 is the initial setting action time, L max is the maximum load of the system, L(t) is the current real-time load value, and γ is the adjustment coefficient, which is usually determined according to the specific operating characteristics of the system. When the system load L(t) increases, the action time T trip The corresponding reduction ensures that the protection device can respond quickly under high load conditions, while under low load conditions, the action time is appropriately extended to prevent false operation and improve the stability of the system;

[0099] Adaptive protection algorithm based on Q-Learning

[0100] In this embodiment, in order to further optimize the parameters of the protection device, a reinforcement learning algorithm based on Q-Learning is used. Q-Learning is a model-free reinforcement learning algorithm that gradually optimizes its strategy through trial and error learning of the intelligent agent in the environment. The algorithm is based on the adjustment of the feedback results of the action time and current threshold of the protection device, so that the protection device can adaptively optimize the response behavior under different load conditions. State and action space: In the Q-Learning algorithm, the state space can be expressed as a combination of the current load condition and the fault risk score, and the action space is different combinations of protection device parameters (action time and current threshold). Update of Q function: The Q-Learning algorithm learns the optimal strategy by continuously updating the Q value. The update formula is:

[0101]

[0102] Among them, s t is the current state, a t is the current action, α is the learning rate, γ is the discount factor, r t is the current reward value Q(s t ,a t ) is the Q value of the state-action pair. By updating the Q value, the system can select the optimal action under different states, that is, the appropriate protection device parameter configuration. Reward mechanism: In the protection system, the reward mechanism is evaluated by the device response effect, that is, a positive reward is given when the fault is quickly isolated and no over-tripping occurs, and a negative reward is given if a false action occurs or the protection times out. Such a reward design ensures that the Q-Learning algorithm is optimized in the direction of minimizing the impact of faults and maximizing the accuracy of protection;

[0103] In this embodiment, the action time and current threshold of the protection device are dynamically adjusted in combination with the real-time load conditions, thereby effectively improving the response sensitivity and adaptability of the protection device. At the same time, the protection strategy is further optimized through the Q-Learning adaptive algorithm, so that the system can perform adaptive protection adjustment under different load conditions, thereby achieving the minimum range of fault isolation. The combination of dynamic adjustment and adaptive optimization strategy effectively prevents over-tripping and improves the self-healing ability and operation stability of the distribution network in fault conditions.

[0104] In this embodiment, the distributed fault isolation technology is first used to identify and quickly isolate the fault area. The fault isolation process is implemented by the adjacency tensor method. The adjacency tensor method constructs the connection relationship between each node in the power grid and describes the connectivity of each node in the form of a matrix. Specifically, the connection relationship between each node is represented by the adjacency tensor A\mathcal{A}A. When a fault occurs, the system automatically detects the faulty node, adjusts the adjacency tensor value of the node, and disconnects the edge connected to the faulty node to achieve isolation of the faulty area;

[0105] Specific;

[0106] Fault area identification;

[0107] When a fault occurs, the system first identifies the faulty node and its adjacent nodes and generates an adjacency tensor for the faulty area. The element a in ij Represents the connection weight between node i and node j. When a fault occurs, the connection edge weight of the faulty node is set to infinity to disconnect it from other nodes, thereby isolating the faulty area.

[0108] ai j＝ ∞(if node i or j belongs to the fault area)

[0109] Distributed isolation of fault areas;

[0110] During the isolation process, the system locally updates the adjacency tensor through a distributed algorithm, disconnects the faulty node, and ensures that the connection of the non-faulty area remains stable. The use of a distributed algorithm ensures that each node can process the isolation operation in parallel, reducing the response time and enabling fault isolation to be completed in a very short time. After the faulty area is isolated, this embodiment restores power to the non-faulty area through topology reconstruction based on a graph algorithm. The reconstruction of the power grid topology calculates the optimal power supply path from the non-faulty node to the main power source through the shortest path algorithm, thereby reallocating power resources to the non-faulty area and restoring power supply.

[0111] Topology reconstruction and shortest path optimization;

[0112] In this embodiment, after the fault isolation is completed, the system uses the shortest path algorithm to calculate the best power supply path from the non-fault area to the power supply. The shortest path algorithm is based on the principle of graph theory and finds the power supply path by calculating the minimum distance between each node to ensure efficient distribution of power. The specific calculation formula is:

[0113] d ij =min(d ij ,d ik +w kj )

[0114] Among them, d ij represents the shortest path distance from node i to node j, d i k and wk j Respectively represent the connection distance and weight between node i and node j. Through the shortest path algorithm, the system can quickly find the power supply path without increasing additional power loss, so that the load in the non-fault area can be restored as soon as possible.

[0115] Dynamic adjustment of topology;

[0116] In this embodiment, after fault isolation and topology reconstruction, the system monitors the stability of the grid load and topology in real time. If load fluctuations or new faults are detected, the system will automatically re-run the shortest path algorithm to adjust the topology to ensure the power supply stability of the grid;

[0117] In this embodiment, the rapid self-healing of the power grid is achieved through distributed fault isolation technology and topological reconstruction of the graph algorithm. Distributed fault isolation technology can quickly identify and isolate the fault area after a fault occurs, reducing the impact of fault expansion; topological reconstruction optimizes the power supply path through the shortest path algorithm, so that the non-fault area can quickly restore power supply, thereby greatly reducing the impact of the fault on the stability of the distribution network. Through the dynamic adjustment of the topological structure, the present invention effectively improves the self-healing ability and reliability of the distribution network.

[0118] In this embodiment, load reconstruction and distributed energy management are designed using a two-layer optimization model to ensure load balance in non-fault areas. The two-layer optimization model is divided into an upper-layer load reconstruction optimization model and a lower-layer distributed energy management optimization model, which respectively handle the supply and demand balance of load nodes and the optimal output of distributed energy, thereby achieving coordinated management of power supply stability and load regulation.

[0119] Specific;

[0120] Load reconfiguration upper layer optimization model;

[0121] The goal of the upper model is to minimize the power imbalance of each load node in the non-fault area to ensure reasonable load distribution. The model optimizes the power difference between nodes to make the power supply match the required power as much as possible. Its objective function is defined as

[0122]

[0123] Among them, P i,供电 represents the actual power supply of load node i, D i ,Demand represents the load demand power of node i, and N is the total number of load nodes in the non-fault area.,This optimization model adjusts the power of each node to minimize the power imbalance between nodes, thereby achieving a dynamic balance of load distribution.

[0124] Distributed energy management lower-level optimization model;

[0125] The lower-level model is used to coordinate the output of distributed energy resources (DER) to meet the power demand in the non-fault area. The goal of the lower-level optimization is to optimize the power supply quality and ensure that the load demand is effectively met by adjusting the output of distributed energy resources. The optimization objective function of the lower-level model is:

[0126]

[0127] in, represents the actual output power of distributed energy node j, β j is the coefficient of distributed energy satisfying load demand, D j,需求 represents the load demand power in the area where distributed energy node j is located, and M is the total number of distributed energy nodes. This objective function optimizes the output of distributed energy to better meet the current load demand and ensure the efficient management of distributed energy;

[0128] Lagrangian relaxation method to solve bi-level optimization problems;

[0129] In this embodiment, in order to coordinate the optimization process of upper-layer load reconstruction and lower-layer distributed energy management, the Lagrangian relaxation method is used to transform the two-layer optimization model into a single-layer optimization problem. By introducing Lagrangian multipliers, the objective functions of the two-layer model are combined and transformed into a joint optimization problem. The objective function of the joint optimization is as follows:

[0130]

[0131] Among them, L is the Lagrange multiplier, which is used to balance the demand of upper-layer load reconstruction and lower-layer distributed energy output. Through the iterative solution of the Lagrangian relaxation method, the optimal solution that meets the supply and demand balance can be found to achieve the coordination of the upper and lower optimization objectives.

[0132] Application and feedback of optimization results;

[0133] After completing the optimization solution, the system applies the results of upper-layer load reconstruction and lower-layer distributed energy management to the power allocation in non-fault areas to ensure load balance. The system monitors the changes in load nodes in real time, makes feedback adjustments based on real-time data, and further optimizes load distribution and the output power of distributed energy to cope with possible load fluctuations or new fault events;

[0134] In this embodiment, through the two-layer optimization model, the present invention can quickly adjust the load distribution in the non-fault area after a fault occurs to ensure the dynamic balance of the system load. The upper optimization model effectively reduces the power imbalance between load nodes, while the lower distributed energy management model achieves effective satisfaction of load demand by optimizing the output of distributed energy. The application of Lagrangian relaxation method further improves the coordination of load reconstruction and distributed energy management, and ultimately improves the stability and self-healing ability of the distribution network.

[0135] In this embodiment, in order to achieve efficient communication between nodes in the distribution network, a multi-channel communication network is constructed using a 5G and optical fiber dual-channel structure. 5G and optical fiber networks are used as wireless and wired communication channels respectively. According to the communication load and delay requirements of the system, the appropriate communication path is automatically selected to ensure low latency and high reliability of data transmission.

[0136] Dual-channel communication network architecture;

[0137] In this embodiment, each key node (such as distribution station, transformer, protection device, etc.) is equipped with a 5G module and a fiber optic module, forming a dual-channel communication network composed of 5G and fiber optic. The system dynamically selects the appropriate communication channel based on factors such as transmission delay, communication load, and real-time performance:

[0138] 5G channel: When the system requires fast response and low-latency communication, the 5G channel is preferred. It is suitable for short-distance transmission and real-time data monitoring scenarios;

[0139] Fiber Channel: Fiber Channel is preferred in scenarios that require high data throughput and stability, and is suitable for remote transmission and non-real-time monitoring scenarios;

[0140] The dual-channel structure can effectively improve the communication flexibility and transmission reliability of the system, ensuring stable data transmission under different communication loads and network conditions.

[0141] Time synchronization and precise control;

[0142] In this embodiment, in order to ensure the consistency of data transmission and synchronization of fault response of each node, the IEEE1588 Precision Time Protocol (PTP) is used to achieve time synchronization between nodes. The IEEE 1588 protocol distributes time information between nodes, enabling all nodes in the system to achieve millisecond-level or even higher precision synchronization, thereby ensuring the consistency of timestamps of fault events. The specific synchronization process is:

[0143] Time server setting: Set up a time server in the system to synchronize time to each node through the PTP protocol;

[0144] Node time adjustment: Each node receives the time information sent by the time server and adjusts the local time in real time according to the time error;

[0145] Through time synchronization, it is possible to ensure the synchronous response of multiple nodes and the time accuracy of data transmission when a fault event occurs, thereby achieving coordinated control of the system;

[0146] Data encryption and transmission security;

[0147] To ensure the security and integrity of data transmission, blockchain technology is used in this embodiment to encrypt and store fault data. Blockchain technology can effectively prevent data from being tampered with or lost during transmission through hash algorithms and tamper-proof chain structures. The specific steps are as follows:

[0148] Data encryption: Before data transmission, the data is encrypted using a hash algorithm to generate a unique hash value for the data;

[0149] Chain storage: Store data hash values ​​and fault information in the blockchain to achieve chain recording and encryption protection of data;

[0150] Data verification: When receiving data, the integrity and authenticity of the data are verified by comparing the hash values ​​before and after transmission;

[0151] The application of blockchain technology prevents fault information from being tampered with or lost during transmission, ensuring data security and improving the anti-attack capability of the distribution network.

[0152] In this embodiment, the dual-channel structure of the multi-channel communication network, the precise time synchronization mechanism and the encryption protection of blockchain technology have achieved high efficiency, accuracy and security of data transmission in the distribution network. The dual-channel structure improves the flexibility and stability of data transmission in the system, ensuring the communication efficiency of the system under high load; the IEEE 1588 time synchronization mechanism ensures the coordination and consistency of fault information in the system; and the application of blockchain technology improves the security and integrity of data transmission.

[0153] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fault self-healing control method for preventing over-tripping in an intelligent distribution network, characterized in that: The steps include: S1. Collect data from key nodes of the distribution network in real time through multiple sensors, and integrate and process them on edge computing nodes to achieve anomaly detection and preliminary fault classification; S2. Predict the fault trend of the distribution network operation status based on the deep learning model and obtain the fault risk score; S3: Dynamically adjust the action time and current threshold of the protection device according to the load conditions, and conduct adaptive protection coordination to ensure fault isolation within the minimum range and prevent over-tripping; S4, quickly isolate the fault area through distributed fault isolation technology, and use graph algorithms to reconstruct the power grid topology and restore power supply to non-fault areas; S5. Load reconstruction and distributed energy management based on a two-layer optimization model to ensure load balance in non-fault areas; S6. Establish a multi-channel communication network between nodes to ensure low latency and high reliability of data transmission and achieve synchronization and coordinated control of fault information.

2. The fault self-healing control method for preventing over-tripping in the intelligent distribution network according to claim 1 is characterized in that: The step S1 specifically includes: S1.

1. Install current sensors, voltage sensors, temperature sensors and vibration sensors at key nodes of the distribution network to collect current data, voltage data, temperature data and vibration data in real time; S1.

2. The collected data is transmitted to the edge computing node, and the multi-source data is fused using the Kalman filter algorithm to reduce data noise and improve data accuracy; S1.3, perform anomaly detection on the fused data using the isolation forest algorithm, calculate the anomaly score, and determine whether there is abnormal data based on the score result; S1.

4. When an abnormality is detected, a preliminary fault classification is performed on the abnormal data based on the current, voltage, temperature and vibration characteristics to determine whether the fault type is a short circuit, overload or ground fault.

3. The fault self-healing control method for preventing over-tripping in the intelligent distribution network according to claim 1 is characterized in that: The step S2 specifically includes: S2.

1. Standardize the historical and real-time collected current, voltage and load data; S2.2, using the long short-term memory network (LSTM) model to predict the fault trend of the standardized time series data; S2.

3. Compare the prediction results of the LSTM model with the actual measured voltage and current values, and calculate the fault risk score R(t) = α·|V pred -V real |+β·|I pred -I real |; S2.

4. Set a fault risk score threshold and issue a fault warning signal when the threshold is exceeded, providing a warning basis for subsequent protection coordination.

4. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: In step S3, the action time and current threshold of the protection device are dynamically adjusted according to the real-time load conditions, and the adjustment formula of the action time is: Among them, T0 is the initial action time, L m ax is the maximum load of the system, L(t) is the current load value, and γ is the adjustment coefficient.

5. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: The step optimizes the action time and current threshold of the protection device through an adaptive protection algorithm. The adaptive protection algorithm in S3 performs reinforcement learning based on the Q-Learning algorithm, and the update formula of the Q function is: Among them, α is the learning rate, γ is the discount factor, and r t For immediate rewards.

6. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: The distributed fault isolation technology in step S4 is implemented based on the adjacency tensor method, by generating an adjacency tensor And the connection edge weight of the fault area is set to infinity to disconnect the faulty node and isolate the faulty area.

7. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: The topology reconstruction in step S4 uses a graph-based shortest path algorithm to optimize the power supply path to restore power supply to the non-fault area. The optimization formula for the shortest path is: d ij =min(d ij ,d ik +w kj ) Among them, d ij represents the shortest path distance from node i to node j, w kj is the weight between nodes k and j.

8. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: The steps are as follows: according to the fault self-healing control method for preventing over-tripping in the intelligent distribution network according to claim 1, wherein load reconstruction and distributed energy management are implemented using a two-layer optimization model. The optimization goal of the upper model is to improve the load balance of the distribution network by minimizing the power imbalance between load nodes. The standard function is: Among them, P i,供电 represents the actual power supply of node i, D i ,demand represents the load demand power of node i, and N is the total number of load nodes in the distribution network; The lower model is used to manage the output of distributed energy so that distributed energy can be optimally allocated according to load demand to improve the self-healing ability of the power grid. The optimization objective function of the lower model is: in, represents the output power of distributed energy node j, β j is the coefficient of distributed energy satisfying load demand, D j,需求 is the load demand power in the area where the distributed energy node j is located, and M is the total number of distributed energy nodes.

9. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: In step S5, the load reconstruction and distributed energy management are solved by the Lagrangian relaxation method, and the two-layer optimization model is converted into a single-layer optimization problem by the relaxation method to achieve coordinated optimization of load reconstruction and distributed energy.

10. The fault self-healing control method for preventing over-tripping in a smart distribution network according to claim 1, characterized in that: The communication nodes of the multi-channel communication network in step S6 adopt a 5G and optical fiber dual-channel structure to ensure low latency of data transmission. The IEEE 1588 precision time protocol is used to achieve time synchronization, and the security and integrity of fault data during transmission are ensured based on blockchain technology.

Images