Power distribution network switch cabinet adaptive control method and system based on machine learning

Abstract

This invention relates to the field of distribution network control technology, and discloses a machine learning-based adaptive control method and system for distribution network switchgear. The method includes: acquiring real-time data of the distribution network; analyzing the data to determine changes in the dynamic power flow direction, constructing the network topology and performing power flow calculations to obtain preliminary voltage and power results; analyzing voltage stability based on the preliminary results and generating alarm information for local overvoltage; constructing a switch state matrix based on the alarm information to determine a temporary load allocation scheme for equipment; analyzing operational constraints and overload risks to obtain an optimized combination of load transfer paths; simulating power dispatch, judging the probability of failure, and obtaining adjusted dispatch instructions; and executing real-time load transfer operations according to the dispatch instructions, monitoring node voltage changes, and determining stability improvement paths. This method can adapt to changes in the power grid, handle local overvoltage, optimize load transfer, and improve power grid stability.

Description

Technical Field

[0001] This invention relates to the field of power distribution network control technology, and in particular to an adaptive control method and system for power distribution network switchgear based on machine learning. Background Technology

[0002] Currently, in modern power systems, the distribution network, as a crucial link connecting generation and consumption, directly impacts the safety and quality of electricity supply for countless households. With the ever-increasing demand for electricity, ensuring the reliable operation of the distribution network in complex environments has become a critical research area. The control and optimization of the distribution network not only affects the continuity of power supply but also relates to energy utilization efficiency and the system's ability to cope with sudden faults.

[0003] In existing technologies, some methods attempt to control distribution networks using fixed rules or static models. For example, some schemes utilize simple machine learning algorithms to analyze historical load data to generate a predictive dispatching scheme. However, this approach largely relies on preset rules to execute subsequent switching controls, making it difficult to adapt to real-time changes in power demand and network conditions. Especially when multiple devices are working collaboratively, these methods lack a comprehensive consideration of real-time changes in network topology and dynamic power flow distribution. This limitation leads to difficulties in determining an optimal sequence of operations that balances load and avoids the risk of local overload when load transfer or network adjustments are required, due to the inability to accurately grasp the interaction between topology and power flow in real time.

[0004] Therefore, existing technologies suffer from low load transfer efficiency and stability risks. Summary of the Invention

[0005] This invention provides an adaptive control method and system for distribution network switchgear based on machine learning, in order to solve the technical problems of low load transfer efficiency and stability risks.

[0006] Firstly, to address the aforementioned technical problems, this invention provides a machine learning-based adaptive control method for power distribution network switchgear, comprising:

[0007] Acquire real-time data of the power distribution network and analyze the real-time data of the power distribution network to determine the change information of the dynamic power flow direction;

[0008] Based on the information on the change in the dynamic power flow direction, the current network topology is constructed, and power flow calculation is performed on the current network topology in combination with preset branch impedance parameters to obtain preliminary results of node voltage and power distribution.

[0009] Based on the preliminary results of the node voltage and power distribution, the node power balance state is calculated to obtain the voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, a local overvoltage alarm message is generated.

[0010] Based on the alarm information of the local overvoltage, a switch state matrix is ​​constructed to determine a temporary scheme for equipment load distribution;

[0011] Based on the temporary load distribution scheme of the equipment, the operation sequence constraints are analyzed and the overload risk is assessed to obtain an optimal combination of load transfer paths;

[0012] Based on the optimized combination of the load transfer paths, power dispatch is simulated, and the possibility of load transfer failure is determined to obtain the adjusted dispatch instructions.

[0013] Based on the adjusted scheduling instructions, perform real-time load transfer operations, monitor changes in node voltage distribution, and determine stability improvement paths.

[0014] Secondly, the present invention provides a machine learning-based adaptive control system for power distribution network switchgear, comprising:

[0015] The data processing module is used to acquire real-time data of the power distribution network, analyze the real-time data of the power distribution network, and determine the change information of the dynamic power flow direction.

[0016] The topology construction module is used to construct the current network topology based on the dynamic power flow direction change information, and to perform power flow calculation on the current network topology in combination with preset branch impedance parameters to obtain preliminary results of node voltage and power distribution.

[0017] The alarm generation module is used to calculate the node power balance state based on the preliminary results of the node voltage and power distribution, and obtain the voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, it generates local overvoltage alarm information.

[0018] The scheme formulation module is used to construct a switch state matrix based on the alarm information of the local overvoltage and determine a temporary scheme for equipment load distribution.

[0019] The path optimization module is used to analyze the operation sequence constraints and assess the overload risk based on the temporary scheme of the equipment load allocation, and obtain an optimized combination of load transfer paths.

[0020] The instruction adjustment module is used to simulate power dispatching based on the optimized combination of the load transfer paths, determine the possibility of load transfer failure, and obtain the adjusted dispatching instructions.

[0021] The execution monitoring module is used to perform real-time load transfer operations according to the adjusted scheduling instructions, monitor changes in node voltage distribution, and determine stability improvement paths.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) This invention constructs a network topology reflecting the current state by acquiring real-time data of the distribution network and analyzing the changes in the dynamic power flow direction; and generates preliminary network loss results by combining branch impedance parameters, and then applies power flow analysis technology to calculate the node power balance state to judge the voltage stability index. This approach breaks away from the dependence of existing technologies on fixed rules, can perceive the dynamic changes of the power grid in real time, and thus accurately identify and locate potential local overvoltage risks, providing accurate alarm basis for subsequent adaptive control.

[0024] (2) This invention addresses local overvoltage alarm information by constructing a switch state matrix to determine a temporary load distribution scheme for the equipment, and further utilizes iterative search technology to analyze operational sequence constraints and assess overload risks. This progressive analysis from "temporary scheme" to "optimized combination" achieves coordinated linkage among multiple devices, ensuring that the generated load transfer path not only meets operational constraints but also effectively avoids overload risks, thus solving the problems of operational incoordination and inefficiency caused by a lack of comprehensive consideration in the prior art.

[0025] (3) Before executing load transfer, this invention uses simulated power dispatch to determine the "possibility of load transfer failure" of the optimized path, and makes final adjustments to the dispatch instructions accordingly. This closed-loop control strategy of "simulating first, then adjusting, and then executing" minimizes the risks of actual operation and ensures the safety and feasibility of dispatch instructions. Finally, by executing the adjusted instructions and monitoring node voltage changes in real time, efficient real-time load transfer can be achieved, stability improvement paths can be determined, and the overall stability and resource optimization level of the distribution network under dynamic scenarios can be significantly improved. Attached Figure Description

[0026] Figure 1 This is a schematic flowchart of an adaptive control method for power distribution network switchgear based on machine learning, provided in the first embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of a machine learning-based adaptive control system for a power distribution network switchgear, provided in the second embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Reference Figure 1 The first embodiment of the present invention provides an adaptive control method for distribution network switchgear based on machine learning, comprising the following steps:

[0030] S11, acquire real-time data of the power distribution network, analyze the real-time data of the power distribution network, and determine the change information of the dynamic power flow direction;

[0031] S12, Based on the information on the change in the dynamic power flow direction, construct the current network topology, and perform power flow calculation on the current network topology in combination with the preset branch impedance parameters to obtain preliminary results of node voltage and power distribution;

[0032] S13. Based on the preliminary results of the node voltage and power distribution, calculate the node power balance state and obtain the voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, generate a local overvoltage alarm message.

[0033] S14. Based on the alarm information of the local overvoltage, construct a switch state matrix and determine a temporary scheme for equipment load distribution.

[0034] S15, Based on the temporary scheme for equipment load distribution, analyze the operation sequence constraints, assess the overload risk, and obtain an optimized combination of load transfer paths;

[0035] S16. Based on the optimized combination of the load transfer paths, simulate power dispatch and determine the possibility of load transfer failure to obtain the adjusted dispatch instructions.

[0036] S17. According to the adjusted scheduling instructions, perform real-time load transfer operations, monitor changes in node voltage distribution, and determine a stability improvement path.

[0037] In step S11, real-time data of the power distribution network is acquired and analyzed to determine the changes in the direction of dynamic power flow, including:

[0038] The real-time data of the power distribution network is marked with abnormal data points to obtain the marked abnormal data points;

[0039] The marked abnormal data points are compared over time to obtain continuous change characteristics;

[0040] The continuous change characteristics are matched with preset historical data to obtain a judgment result;

[0041] Based on the judgment results, the abnormal trends in the direction of power flow are analyzed to determine the change information of the dynamic power flow direction.

[0042] It should be noted that the real-time data of the distribution network is obtained through a sensor network deployed on key nodes and lines of the distribution network. The sensor network performs high-frequency sampling of node voltage distribution and line current load, for example, sampling at a frequency of once every 5 seconds, forming a voltage and current fluctuation dataset containing precise timestamps. This dataset is the real-time data of the distribution network.

[0043] It should be noted that marking abnormal data points in the real-time data of the distribution network is a threshold-based screening process. The system compares the real-time collected voltage and current values ​​with a preset threshold range. If the voltage value of a node or the current value of a line exceeds the preset threshold range, the system marks it as an abnormal data point and records its location and time information, thereby obtaining the marked abnormal data points.

[0044] It is worth noting that the preset threshold range is determined based on national power quality standards and power grid safety operation regulations. This aims to ensure that the power grid operates within safe boundaries. For example, for a node voltage rated at 220V, the safety threshold range is typically set at ±5% of its rated value, i.e., 209V to 231V; while for a line rated at 100A, the current load threshold range can be set at ±10% of its rated value, i.e., 90A to 110A.

[0045] It should be noted that those skilled in the art will understand that, in the context of this invention, the terms "load" and "load" are used interchangeably. Both refer to equipment or components connected to an electrical system and consuming electrical energy, or specifically to the amount of power consumed. To avoid ambiguity, it is hereby declared that the two terms have the same technical connotation in this invention.

[0046] For example, in a medium-sized city power distribution network scenario, the system detected that the voltage value of "Node A" remained above 235V from 10:00 to 10:05, exceeding the upper limit threshold of 231V; at the same time, the current of a line connected to Node A reached 120A, exceeding the upper limit threshold of 110A. Therefore, the system marked the data of "Node A" during this time period as the abnormal data point.

[0047] It should be noted that the time-series comparison of the marked abnormal data points aims to extract continuous change patterns from discrete anomalies. The system analyzes abnormal data points that appear consecutively within a specific time window (e.g., within 5 minutes) to identify their dynamic change patterns, such as "continuous overvoltage," "periodic overcurrent," or "current direction reversal." These identified patterns with temporal continuity are the continuous change characteristics.

[0048] It should be noted that matching the continuously changing features with the preset historical data is achieved through a pre-trained Support Vector Machine (SVM) classification model. This model undergoes supervised training using the preset historical data; it employs a Gaussian radial basis function kernel with kernel coefficients γ ranging from 0.1 to 1.0, a regularization parameter C optimized between 1.0 and 10.0, and a tolerance parameter ε set to 0.0005 to ensure classification accuracy and generalization ability.

[0049] Specifically, the training dataset uses continuously changing features extracted from historical events (e.g., "persistent overvoltage," "periodic overcurrent," or "current direction reversal") as input features and known diagnostic results of these events (e.g., "equipment failure," "typical load imbalance") as output labels. The SVM model (e.g., using a Gaussian radial basis function (RBF) kernel) learns from this data to construct an optimal classification hyperplane in a high-dimensional feature space to distinguish different fault modes. During actual operation, the system uses the currently acquired continuously changing features as input to the pre-trained SVM classifier. Based on the positional relationship between the feature vector and the classification hyperplane, the model outputs a classification prediction result, which is the judgment result (e.g., "potential distribution network load imbalance exists"). Its confidence level (i.e., distance to the hyperplane) is used to assess the risk level.

[0050] For example, the system inputs the continuously changing features of "node A experiencing continuous overvoltage accompanied by current reversal" (extracted as an input feature vector) into a pre-trained Support Vector Machine (SVM) classifier. The model classifies this feature vector as belonging to the "historical load imbalance precursor" category based on its positional relationship in high-dimensional space with the classification hyperplane corresponding to the "historical load imbalance precursor" category. Therefore, the system's judgment result is "there is a potential distribution network load imbalance hazard," and the model outputs a high confidence score (e.g., 0.92) based on the distance of the vector to the hyperplane to assess the risk level of this hazard.

[0051] It should be noted that analyzing the abnormal trend of the power flow direction based on the judgment result and determining the change information of the dynamic power flow direction is the final output of step S11. This step comprehensively analyzes and confirms the judgment result (such as "potential load imbalance") with the original data trend that led to the judgment (such as "current direction reversal"). The finally determined change information of the dynamic power flow direction is a structured and confirmed abnormal event information, which will serve as the basis for constructing the network topology in the subsequent step S12.

[0052] In step S12, based on the information about the change in the dynamic power flow direction, the current network topology is constructed, and power flow calculations are performed on the current network topology using preset branch impedance parameters to obtain preliminary results of node voltage and power distribution, including:

[0053] Based on the information on the change in the dynamic power flow direction, graph theory analysis is performed to obtain the network topology diagram;

[0054] Based on the network topology diagram, a matching process is performed using the preset branch impedance parameters to obtain the impedance distribution;

[0055] If the line value in the impedance distribution exceeds the preset impedance threshold range, then the load distribution characteristics are obtained;

[0056] Based on the load distribution characteristics and the network topology diagram, an overlay analysis is performed to determine the preliminary results of the node voltage and power distribution.

[0057] It should be noted that graph theory analysis based on the dynamic power flow direction change information is the foundation for constructing the power grid topology model. In this embodiment, the processing treats each node in the distribution network (such as substations and switchgear) as a "vertice" of the graph, and the connecting lines between nodes as "edges". When the dynamic power flow direction change information determined in S11 indicates that a switch state has changed (such as closed or open), the system updates the connection relationship of the "edges" in real time, thereby generating a visualized topology graph reflecting the current actual connection state of the power grid, i.e., the network topology structure graph.

[0058] For example, suppose a distribution network contains 10 key nodes and 15 connecting lines. If the dynamic power flow direction change information indicates that the switch between node B and node C is closed, the system will immediately add an "edge" connecting node B and node C in the topology graph, thereby updating the network topology graph.

[0059] It should be noted that the matching process, performed on the network topology diagram and the preset branch impedance parameters, aims to assign physical parameters to the topology model. The preset branch impedance parameters are stored in a pre-established device database, which records the standard resistance and reactance values ​​of each line. The "matching process" in this step involves traversing each "edge" in the network topology diagram, querying the database, and assigning it the corresponding impedance parameters, thereby obtaining an impedance distribution containing complete electrical parameters.

[0060] It should be noted that if the line values ​​in the impedance distribution exceed the preset impedance threshold range, the acquisition of load distribution characteristics is a conditionally triggered data capture process. The system will check the parameters of each line in the impedance distribution in real time. If an abnormal impedance value (or loss-related value calculated from impedance and current) of a certain line is found, a data processing operation will be triggered to retrieve detailed operating data (such as segmented current, power factor, etc.) of the abnormal line in order to analyze the spatial or temporal distribution pattern of its load. These patterns are the load distribution characteristics.

[0061] It is worth noting that the preset impedance threshold range is determined based on the equipment's factory specifications, line model, and historical operating data statistics. For example, it is determined for a specific line model, such as a 50mm² cross-section. 2 For a 1km cable, historical impedance data under normal operating conditions (e.g., load rate below 70%) was statistically analyzed, and the 95% confidence interval (e.g., 0.5 to 1.5 ohms) was used as the impedance threshold range. Other lines were normalized using per-unit values ​​to ensure general applicability. Any measurement value outside this range was considered a significant anomaly and required further analysis.

[0062] In another implementation, the impedance threshold range is determined based on calculations of the line's thermal stability limit, taking into account the conductor's maximum allowable temperature (e.g., 70°C / 80°C / 90°C depending on the insulation class), ambient temperature (e.g., 40°C), and solar radiation intensity (e.g., 1000W / m). 2 Based on conditions such as these, the steady-state current carrying capacity is calculated using the thermal balance equation provided by IEEE Std 738-2012, and then converted into the impedance safety range by combining the line resistance parameters.

[0063] For example, after matching, the system finds that the impedance of a certain line reaches 2.0 ohms, exceeding the upper limit threshold of 1.5 ohms. The system then triggers a data processing tool to obtain detailed data of the line. The load distribution characteristic obtained after analysis is that "the load is concentrated at the end of the line, resulting in an excessively high overall equivalent impedance of the line."

[0064] It should be noted that determining the preliminary results of the node voltage and power distribution by overlaying the load distribution characteristics with the network topology diagram is a data visualization and decision support process. In this embodiment, a heatmap is used to overlay the load distribution characteristics (such as high-load areas) onto the network topology diagram (the geographical or logical wiring diagram of the power grid) in a highlighted (e.g., red) form. The result of this overlay analysis visually identifies high-loss, high-risk "key concern areas" in the power grid. This visualized chart represents the preliminary results of the node voltage and power distribution, used for power flow analysis in S13.

[0065] In step S13, based on the preliminary results of the node voltage and power distribution, the node power balance state is calculated to obtain a voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, a local overvoltage alarm message is generated, including:

[0066] Based on the preliminary results of the node voltage and power distribution, the difference between injected and consumed power is calculated to obtain power balance state data;

[0067] Based on the power balance state data, the voltage value of each node is analyzed, and the voltage value is used as the voltage stability index.

[0068] If the voltage stability index exceeds the preset voltage threshold condition, an alarm message for local overvoltage is generated for the abnormal node.

[0069] It should be noted that calculating the difference between injected and consumed power based on the preliminary results of the node voltage and power distribution is a process of initiating power flow simulation calculations. The system obtains the power distribution information of each node based on the preliminary results and pre-established power grid topology data. It then calculates the difference between the injected and consumed power at each node through simulation. The specific calculation formula is as follows:

[0070]

[0071] in For nodes The power difference, For the node Injection power, For the node The power consumed. This calculation result is used to quantify the power balance state of the node, thereby obtaining the power balance state data.

[0072] For example, suppose in a power flow simulation, the injected power of a certain "node A" ( The power consumption is 500MW, while the power consumption is ( If the power difference is 450MW, then the calculated power difference is ( The value of +50MW indicates that the node is in a power surplus state. The injected power of another node, "Node B", is 300MW, and the power consumed is 400MW, with a difference of -100MW, indicating insufficient power.

[0073] It should be noted that analyzing the voltage value of each node based on the power balance state data is a voltage calculation process to assess the impact of power distribution on grid stability. The system calculates the predicted voltage value of each node under the current operating conditions based on the power balance state data (i.e., the power surplus or deficit status of each node), and this voltage value is used as the voltage stability index to quantify the current stability.

[0074] It is worth noting that the preset voltage threshold condition is set based on industry standards for the safe operation of power systems. In distribution networks, to ensure power quality and equipment safety, node voltages must be maintained within a stable range close to their rated values. Therefore, this threshold condition is typically set to ±5% of the rated voltage value, i.e., between 0.95 and 1.05 times the rated value.

[0075] For example, assume the preset voltage threshold condition is 0.95 to 1.05 times the rated value. After calculating the voltage of "Node A" (power surplus), its voltage stability index is found to be 1.08 times the rated value. Since 1.08 times exceeds the upper limit threshold of 1.05 times, the system triggers the next step: for the abnormal node "Node A", it generates the local overvoltage alarm information and clearly marks its location.

[0076] It should be noted that generating the local overvoltage alarm information for abnormal nodes is an automated warning triggering process. When the system determines that the voltage stability index of any node exceeds the preset voltage threshold, the system will immediately generate a structured alarm data. This alarm data typically includes the identifier of the abnormal node, its specific location information, the currently calculated voltage value (e.g., 1.08 times the rated value), and an event timestamp. This alarm information is the local overvoltage alarm information, used to activate the multi-device coordination mechanism in subsequent steps.

[0077] For example, in response to the anomaly of "Node A" mentioned above, the system will generate an alarm message, the content of which may be a piece of structured data, such as:

[0078] {"event_id":"OV-20251102-001","timestamp":"2025-11-02T23:55:00Z","node_id":"Node A","location":"Northern Region - Critical Substation","event_type":"Local Overvoltage","value":"1.08pu","threshold":"1.05pu"}.

[0079] This data is the alarm information for the aforementioned local overvoltage.

[0080] In step S14, based on the alarm information of the local overvoltage, a switch state matrix is ​​constructed to determine a temporary scheme for equipment load distribution, including:

[0081] Based on the alarm information of the local overvoltage and in conjunction with the preset power grid topology data, the voltage distribution in the affected area is determined;

[0082] Based on the voltage distribution, extract equipment operating data and compare and analyze the equipment operating data with preset equipment capacity information to determine abnormal equipment status records;

[0083] Based on the device status anomaly records, construct the switch status matrix;

[0084] Perform matrix calculations on the switch state matrix to determine an adjustable switch combination scheme;

[0085] Based on the adjustable switch combination scheme, the abnormal equipment status records, and the preset power grid topology data, load allocation processing is performed to generate a temporary load allocation scheme for the equipment.

[0086] It should be noted that determining the voltage distribution in the affected area based on the local overvoltage alarm information and in conjunction with preset power grid topology data is a process of location and correlation analysis. The system uses the abnormal node ID (e.g., "Node C") in the alarm information as an index to query the real-time parameters (such as current and power data) of that node and its adjacent branches in the preset power grid topology data. By analyzing the spatial distribution of these data, the voltage distribution in the affected area is obtained.

[0087] It is worth noting that the preset power grid topology data is a digital benchmark for power grid operation, sourced from the power company's asset management database or the configuration file of the dispatch automation system. This data accurately records the electrical parameters of all physical devices in the power grid (such as lines, transformers, and switches) and their unique connections, serving as a well-known foundation for any network analysis and power flow calculation.

[0088] For example, the system receives a local overvoltage alarm from "Node C". By querying the power grid topology data, it is found that "Node C" is located in the southern region, and among the three branches connected to it, the current value of one branch is too high, reaching 120% of the rated value. After comprehensive analysis, the system determines that the voltage distribution is "the overall voltage in the southern region is too high due to local load concentration".

[0089] It should be noted that extracting equipment operating data based on the voltage distribution and comparing it with preset equipment capacity information to identify abnormal equipment status records is a process of diagnosing overloaded equipment. The system retrieves real-time equipment operating data (e.g., current load power, in kW) of key equipment (such as transformers) in the affected area and compares it with preset equipment capacity information (e.g., rated capacity, in kW) to calculate the real-time load rate. If this load rate exceeds a preset load rate threshold, the system determines that the equipment is overloaded and generates the abnormal equipment status record.

[0090] It is worth noting that the preset equipment capacity information comes from the equipment's factory specifications or asset management database, and is an inherent physical attribute of the equipment. The preset load rate threshold (e.g., 70%) is set based on power system operation and maintenance experience. Setting it at a safety margin far below 100% aims to ensure that the equipment has sufficient headroom to cope with sudden fluctuations, which is common knowledge for ensuring the stable operation of the power grid.

[0091] For example, the system extracts the equipment operation data of a transformer near "Node C" as having a current load of 900kW. Its preset equipment capacity information is 1000kW. The system calculates the current load rate to be 90% (900 / 1000). Since 90% exceeds the preset load rate threshold of 70%, the system immediately generates an abnormal equipment status record, including {Equipment Number: T-01, Location: Node C, Status: Overload, Load Rate: 90%}.

[0092] It should be noted that, based on the abnormal device status records, the switch status matrix is ​​constructed. The system locates the abnormal device (such as T-01) based on the abnormal records and identifies all switches associated with that device in the power grid topology. The switch status matrix is ​​a vector or matrix, and its element values ​​(usually 1 or 0) represent the current physical state of these associated switches (1 represents closed, 0 represents open).

[0093] It should be noted that performing matrix calculations on the switch state matrix to determine the adjustable switch combination scheme is a process of finding a feasible solution. The system uses the current switch state matrix as the initial state, and, in conjunction with power grid topology constraints (e.g., prohibiting islanding) and the abnormal equipment status records (i.e., requiring a reduction in the load on T-01), calculates one or more new switch state matrices through search or optimization methods. These new matrix combinations, which can both solve the overload problem and satisfy operational constraints, constitute the adjustable switch combination scheme.

[0094] It is worth noting that the matrix calculation is a constraint optimization-based solution process, the core of which is to solve a network reconfiguration problem. In this embodiment, the calculation is implemented using a heuristic search algorithm, such as a genetic algorithm. The specific implementation of the genetic algorithm involves encoding each potential switch combination scheme (i.e., a new switch state matrix) as a "chromosome." This chromosome is a binary vector with a length of... Equal to the total number of adjustable switches, where the vector of the first rank is... Position Representing the When a switch is closed, for This indicates that it is disconnected. A random number containing... The initial population of chromosomes (e.g.) After each chromosome is generated, a topology constraint check is immediately performed to ensure that the switching state it represents satisfies the "grid topology constraints" (i.e., maintaining network connectivity and not allowing islanding). Invalid chromosomes are discarded and regenerated until the population size reaches a certain threshold. The parameters of the genetic algorithm are determined based on the search space size of the distribution network switch combination. The population size is set to 5-10 times the number of adjustable switches to ensure sufficient exploration of the solution space. The maximum number of iterations is determined through pre-experiments, based on the convergence of the fitness function value to a stable plateau.

[0095] The fitness function of this algorithm (i.e., the optimization objective function) is designed to minimize a comprehensive cost, which primarily penalizes the overload conditions indicated in the "device status anomaly log". For example, the fitness function can be defined as... ,in It is the sum of penalties for the overload of all overloaded equipment (such as T-01). This is the total network loss under this switch combination. and These are preset weighting coefficients (e.g.) (Prioritizing the resolution of overload issues).

[0096] Specifically, the total penalty for the overload The calculation is performed by summing the squares of the excess load rates of all critical equipment (such as transformers and lines) compared to their preset load rate thresholds. The formula is as follows:

[0097]

[0098] in, Traverse all monitored critical device sets , It is equipment The simulated load rate (per unit) under the current switch combination scheme (i.e., the current "chromosome"). This is a preset load rate threshold for the device (e.g., 70% or 0.7 pu). This squared penalty mechanism ensures that severely overloaded devices are given a greater penalty weight, thereby guiding the algorithm to prioritize addressing the most serious security risks.

[0099] Tournament selection is used to select chromosomes with high fitness from the current population as parents. A single-point crossover operation is then applied to the selected parents to generate offspring chromosomes. The offspring chromosomes are then processed with a low mutation probability (e.g., ...). The algorithm performs a bit-flip mutation to introduce new switch combinations. All newly generated offspring chromosomes must again pass the "grid topology constraint" check; offspring that do not meet the constraint are discarded. The algorithm iterates until a preset maximum number of iterations is reached (e.g., ...). The algorithm terminates when the value of the fitness function no longer significantly improves over multiple generations (e.g., 20 generations). Upon termination, the combination of switch states represented by the chromosome with the highest fitness in the population is determined as a "feasible solution" that satisfies all constraints and effectively alleviates overload, and is selected as one of the adjustable switch combination schemes.

[0100] It is worth noting that the above and The weighting coefficients are set based on the well-known technical principle of "safety first, with consideration for losses" in power grid operation. The core objective of this algorithm is to eliminate overloads (safety issues) identified by abnormal equipment status records, and secondarily to optimize network losses (technical performance issues). Therefore, The value was set to be much greater than The value (e.g.) Compare Several orders of magnitude higher, such as The method for determining this setting is to ensure that... The minimum non-zero penalty value is also greater than With the maximum possible network loss The product of these factors ensures mathematically that any evolutionary direction of the algorithm will absolutely prioritize choosing a chromosome with "no overload" rather than a chromosome with "slight overload but extremely low network loss," thus strictly adhering to a safety-first optimization strategy.

[0101] For example, suppose there are 5 key switches in the "Node C" region, and the current state matrix is ​​as follows: The heuristic search algorithm found that by changing the state to... (That is, by disconnecting the second switch and closing the third switch), part of the load on T-01 can be transferred to an adjacent branch. Therefore, It was determined to be the adjustable switch combination scheme.

[0102] It should be noted that the final step in S14 is to perform load allocation processing based on the adjustable switch combination scheme, the equipment status anomaly records, and the preset power grid topology data to generate a temporary load allocation scheme for the equipment. This process simulates the execution of the adjustable switch combination scheme (e.g., ...). The power flow is recalculated based on the power grid topology data and load information contained in the anomaly records. If the calculation results (such as the adjusted voltage and load rate) meet the preset voltage threshold conditions and the preset load rate threshold conditions, the simulation scheme is confirmed as a feasible and safe temporary scheme.

[0103] It is worth noting that the preset voltage threshold condition and the preset load rate threshold condition are the final verification standards to ensure the safety and feasibility of the scheme. The voltage threshold condition is the safe operating range set based on industry standards (e.g., 0.95 to 1.05 times the rated value) mentioned in S13; the load rate threshold condition is the safety margin set based on historical operation and maintenance data (e.g., 70%) mentioned earlier in step S14. Only when all relevant adjusted equipment simultaneously meets these two conditions can the simulation scheme be finally confirmed.

[0104] For example, system simulation Solution. Power flow calculations show that the load on T-01 decreases from 900kW (90%) to 700kW (70%), while the load on the adjacent "Node D" transformer increases from 500kW to 700kW. Since the load rates and voltages of both nodes are within safe thresholds, the system confirms this solution is feasible. This scheduling plan, which includes the specific load transfer path and expected results ("T-01 900kW -> 700kW"), constitutes the temporary solution for load allocation to the aforementioned equipment.

[0105] In step S15, based on the temporary load distribution scheme for the equipment, the operation sequence constraints are analyzed, and the overload risk is assessed to obtain an optimized combination of load transfer paths, including:

[0106] A temporary solution for the device load distribution is analyzed in conjunction with preset operation sequence constraints to obtain a combination of operation sequences;

[0107] Based on the combination of the aforementioned operation sequence, an overload risk assessment is performed, and a risk assessment report is obtained.

[0108] Based on the risk assessment report, path planning is performed to determine the optimal combination of load transfer paths.

[0109] It should be noted that the analysis of the temporary load allocation scheme for the aforementioned equipment, combined with preset operation sequence constraints, is a process of generating an executable sequence through iterative search techniques. The temporary scheme defines the objective (e.g., reducing the load on node A by 200kW), while the preset operation sequence constraints define the rules. This analysis process, through an iterative search method, plans a specific and safe sequence of switching operations to achieve the objective under the constraints of these rules; this sequence is the operation sequence combination.

[0110] It is worth noting that the preset operation sequence restrictions are set based on equipment safety specifications and power grid operation experience. For example, the determination method is based on statistical analysis of historical operating data. If the data shows that when the load rate of a certain transformer (e.g., T-02) exceeds 90%, the response delay of its associated switch increases by 30%, then the system will generate a restriction: "When the load rate of T-02 is higher than 90%, the fast switching operation associated with it is prohibited." These restrictions ensure that the operation sequence does not exceed the safe operating range of the equipment.

[0111] For example, suppose the temporary solution is "to transfer the load of node A to node B". The preset operation sequence constraints include "the high-load branch switch (Switch-A1) of node A must be disconnected before the access switch (Switch-B1) of node B can be closed". The system analyzes and generates the operation sequence combination as a list of instructions with a clear timing, [1. Disconnect Switch-A1, 2. Close Switch-B1].

[0112] It should be noted that the overload risk assessment based on the aforementioned sequence of operations is a process of verifying the safety of each step in the sequence. The system acquires each step in the sequence and uses a risk calculation method to simulate the power flow state of the grid after executing that step. Through this simulation, the system obtains relevant data on the overload risk level associated with that step (e.g., predicted line current values), determines whether there is a potential overload risk, and finally summarizes the assessment results of all steps to obtain the risk assessment report.

[0113] For example, the system evaluates the operation sequence combination generated in the previous step: [1. Disconnect Switch-A1], risk calculation shows that after performing this step, the load on all lines is below 100%. [2. Close Switch-B1], risk calculation shows that after performing this step, the current value of "Line L5" which bears the load will reach 130% of its rated value. The system determines that there is a high overload risk in step 2, therefore the generated risk assessment report is: {"Status":"Unsafe","Step":2,"Risk":"High","Detail":"Line L5 is expected to be overloaded to 130%"}.

[0114] It should be noted that the path planning process, based on the risk assessment report, to determine the optimal combination of load transfer paths is a risk feedback-based scheme correction process. If the risk assessment report displays "Unsafe," the system will activate the path planning method. This method is based on the original temporary scheme's objective (e.g., "reducing the load at node A"), but incorporates the risks identified in the risk assessment report (e.g., "overload on line L5") as new hard constraints. The system uses the path planning method to re-search the power grid topology, seeking alternative paths that achieve the original objective while circumventing the new constraints. These alternative paths constitute the optimal combination of load transfer paths.

[0115] For example, the risk assessment report indicates that "Line L5" (the path from node A to node B) is high-risk. The path planning method is activated, and it re-searches with the goal of "reducing the load on node A". Analysis reveals that although the path "A to B" (Line L5) is unavailable, there are two alternative paths: "transferring part of the load from node A to node C" (Line L6) and "transferring to node D" (Line L7). The method assesses that neither of these new paths poses an overload risk. Therefore, the system determines the optimal combination of the load transfer paths as {Path1:[A->C, via L6],Path2:[A->D, via L7]}.

[0116] In step S16, based on the optimized combination of load transfer paths, power dispatch is simulated, and the probability of load transfer failure is determined to obtain adjusted dispatch instructions, including:

[0117] Based on the optimized combination of the load transfer paths, power dispatch simulation is performed to obtain the load distribution status;

[0118] Based on the aforementioned load distribution status, a load transfer failure risk assessment is performed, and a risk level report is obtained.

[0119] Based on the optimized combination of the risk level report and the load transfer path, an instruction matching degree analysis is performed to determine the adjusted scheduling instruction.

[0120] It should be noted that the power dispatch simulation based on the optimized combination of load transfer paths is a "pre-play" or "deduction" process conducted in a digital twin environment. The system takes the optimized combination of load transfer paths determined in S15 as input and performs dynamic power flow calculations within a preset network security boundary to predict the expected power distribution and voltage level of each key node in the network after the scheme is implemented. These predicted steady-state results are the load distribution state.

[0121] It is worth noting that the preset network security boundary is set based on power system operation guidelines. Its determination method comprehensively considers the rated capacity of equipment (such as transformers and lines), the thermal stability limits provided by the manufacturer, and the maximum and minimum voltage deviations (e.g., ±5%) specified in industry safety regulations. These parameters together constitute a multi-dimensional set of constraints, ensuring that any simulated dispatching scheme remains within the physically and procedurally permissible safety limits.

[0122] For example, suppose the optimized combination of load transfer paths output by S105 is "transferring part of the load of node D to nodes E and F". Power dispatch simulation calculations show that the load distribution state after the implementation of this scheme is as follows: the load of node D drops to 82%, the load of node E rises to 78%, and the load of node F rises to 85%.

[0123] It should be noted that the load transfer failure risk assessment for the aforementioned load distribution status is a process of quantifying dynamic execution risk using a pre-trained Random Forest Regression model. This assessment considers not only the static risks in S15 but also focuses on analyzing dynamic factors not taken into account in S15. The model is trained in a supervised manner using log data containing historical load transfer operations. The training dataset uses the aforementioned dynamic factors from historical operations as input features, which specifically include equipment response time (e.g., switchgear closing delay), branch reliability (e.g., line aging or number of faults extracted from historical fault files), current load rate, and related environmental factors (e.g., temperature and humidity). The output labels of the dataset are the actual results of these historical operations (e.g., whether the transfer was successful, 1 for failure, 0 for success).

[0124] It is worth noting that the random forest regression model sets the number of base learners to 200 to 300, the maximum tree depth to 15 to 20, the minimum number of samples for node splitting to 5, the feature selection strategy to adopt the square root rule, and the hyperparameters to be determined through Bayesian optimization. In cross-regional testing, the prediction error is reduced by 25%, and the single inference time is less than 10 milliseconds, ensuring the real-time performance and reliability of scheduling instructions.

[0125] In actual operation, the system takes the dynamic factors (current equipment response time, branch reliability, etc.) corresponding to the current "optimized combination of load transfer paths" to be evaluated as input to the pre-trained random forest regressor. The model averages the prediction results of multiple decision trees within it and finally outputs a continuous prediction value (between 0 and 1). This value is the potential failure probability (e.g., "25%) due to delay or component failure when executing the path. This probability is compared with a preset risk threshold (e.g., "10%)" to finally generate the risk level report.

[0126] It is worth noting that the preset risk threshold (e.g., 10%) is set based on operational risk preferences. Through statistical analysis of historical power grid operation data, operations are categorized by "criticality" (e.g., the threshold for hospital power supply operations would be extremely low, while the threshold for routine load balancing could be slightly higher). This 10% threshold is an empirical value that strikes a balance between safety redundancy and scheduling efficiency for routine load transfer scenarios.

[0127] For example, when evaluating the path "Node D -> Node E", the system first obtains the dynamic factors corresponding to the path as input features. For instance, it extracts "branch reliability" ("severe aging" is quantified as a feature value of 0.2) from historical archives and combines it with the current "device response time" (e.g., 0.9 seconds) and "current load rate" (e.g., 70%). The system then inputs a vector composed of these features into a pre-trained random forest regression model. The model makes a prediction and ultimately outputs a "potential failure probability" of 25% for the path. Since 25% is higher than the preset risk threshold of 10%, the risk level report generated by the system is {"Status":"HighRisk","Path":"D->E","Probability":"25%","Reason":"Model prediction (based on low branch reliability)"}.

[0128] It should be noted that the final decision-making step in S16 is to perform instruction matching analysis based on the optimized combination of the risk level report and the load transfer path to determine the adjusted scheduling instruction. The system analyzes the risk level report. If the risk is "LowRisk," the optimized combination of the load transfer path is directly confirmed as the adjusted scheduling instruction. If the risk is "HighRisk," the system activates the path planning method, removes high-risk paths (such as "D->E") from the original optimized combination, and selects the path with the lowest risk and best matching the original scheduling objective from the remaining suboptimal paths (such as "D->F"), which is then used as the adjusted scheduling instruction.

[0129] For example, the system performs an instruction matching analysis based on the "HighRisk" risk level report. The analysis determines that the high-risk "D->E" path should be eliminated. The system then analyzes the alternative path "D->F" in the optimized combination and finds that its failure probability is only 5% (below the 10% threshold). Although the overhead is slightly higher, the instruction matching degree is high and it is safe. Therefore, the system finally determines the adjusted scheduling instruction as "Execute the load transfer operation of 'Node D->Node F'".

[0130] In step S17, according to the adjusted scheduling instructions, a real-time load transfer operation is performed, and changes in node voltage distribution are monitored to determine a stability improvement path, including:

[0131] According to the adjusted scheduling instructions, the real-time load transfer operation is executed, and the voltage fluctuation range of the nodes is monitored in real time. If the fluctuation range exceeds the preset voltage fluctuation threshold, the distribution location of the abnormal nodes is determined.

[0132] Based on the distribution location of the abnormal nodes, path planning is performed to obtain a preliminary solution for the stability improvement path;

[0133] Based on the preliminary scheme of the stability improvement path, instruction execution efficiency analysis is performed to obtain the adaptation results of the dynamic adjustment mechanism;

[0134] Based on the adaptation results of the dynamic adjustment mechanism, the final path is determined to obtain the stability improvement path.

[0135] It should be noted that the execution of the real-time load transfer operation based on the adjusted scheduling instructions, and the real-time monitoring of node voltage fluctuations, constitute a closed-loop execution and feedback process. The system first sends and executes the adjusted scheduling instructions (e.g., "execute load transfer from node D to node F") to the underlying switchgear. Simultaneously, the system uses a high-frequency sensor network to monitor "node D" and "node F" involved in the operation in real time to track their node voltage fluctuations.

[0136] It is worth noting that the preset voltage fluctuation threshold is determined based on transient stability analysis. By statistically analyzing massive amounts of historical normal load transfer operation data, the instantaneous oscillation amplitude and decay time of the voltage during operation (e.g., within 500 milliseconds after closing the circuit breaker) are analyzed. This threshold (e.g., instantaneous fluctuation not exceeding ±3%) is set as the upper limit of the 99% confidence interval to distinguish between normal operational transients and abnormal fluctuations caused by slow equipment response or oscillation instability.

[0137] In another embodiment, the voltage fluctuation threshold is determined based on the voltage drop tolerance in transient stability analysis. The power distribution system should be able to withstand the instantaneous voltage drop caused by the closing operation, with the amplitude not exceeding 10% to 15% of the rated voltage and the duration not exceeding 0.5 to 1.0 seconds. The ±3% threshold used in this embodiment provides an additional safety margin to ensure that the protection device is not falsely triggered.

[0138] For example, when the system performs the transfer operation from "Node D to Node F", it monitors that the voltage of "Node D" drops sharply by -7% at the moment of closing, exceeding the preset voltage fluctuation threshold of ±3%. The system immediately triggers an alarm and identifies "Node D" as the location of the abnormal node.

[0139] It should be noted that the path planning process, which involves determining the distribution location of the abnormal nodes to arrive at a preliminary stability improvement path, is a real-time fault avoidance and replanning process. After determining the distribution location of the abnormal node (e.g., "Node D"), the system immediately analyzes the cause of the node's abnormal response (e.g., by comparing its actual response time with a "preset response time standard"). If the system finds that the node's response is slow (e.g., below the standard), it immediately lowers the node's priority in the path planning and replans an alternative path; this new path is the preliminary stability improvement path.

[0140] It is worth noting that the preset response time standard (e.g., 1 second) is set based on the device's factory specifications and network communication protocols (such as IEC 61850). It defines the maximum acceptable delay that the device must complete a physical action after receiving an instruction.

[0141] For example, system analysis revealed that the response time of "Node D" was 3.2 seconds, far slower than the preset response time standard of 1 second. The system determined that "Node D" was responding abnormally, immediately terminated the "D->F" path, and reduced its priority in path planning. The system replanned and found that transferring the load from "Node D" to "Node G" was the second-best option. Therefore, the initial solution for the stability improvement path obtained by the system is to "terminate D->F and use the D->G path instead."

[0142] It should be noted that the instruction execution efficiency analysis based on the preliminary scheme of the stability improvement path is a comprehensive performance indicator (KPI) evaluation process performed on the new scheme before execution. The system obtains the preliminary scheme (such as the "D->G path") and quickly simulates a set of key performance indicators for executing the scheme, mainly including time delay (the expected total time from instruction issuance to completion of the switching action). Resource loss (expected network loss of the path after the transfer is executed) ), and voltage stability (the expected instantaneous voltage drop at "node G"). The system uses a preset objective function to perform weighted calculations on these indicators to obtain a comprehensive performance score, which is the adaptation result of the dynamic adjustment mechanism.

[0143] It is worth noting that the preset objective function is determined based on the technical principle of "safety first, with efficiency as a secondary consideration" in power grid operation. First, an objective "benchmark value" and "penalty weight" are set for each performance indicator. For example, the benchmark value for time delay (e.g., ...) (seconds) is set based on communication protocols and device response specifications; voltage stability reference values ​​(such as...) Based on transient stability and safety procedures; the baseline value of resource loss (such as...) This is based on historical average losses. The objective function (e.g.) The weights in ) (e.g., corresponding to voltage stability) ) is set to be much higher than other weights ( and This ensures that any solution that might threaten voltage stability receives an extremely high cost score and is therefore rejected in the evaluation. The weighting coefficients are set based on a safety-first principle; for example, they can be set to... : : The ratio is 100:1:1 to ensure that the voltage stability index has a decisive influence in the evaluation; all physical quantities in the objective function are calculated using relative deviations from the benchmark value to ensure uniformity of dimensions and comparability of numerical values. A scheme's "fitting result" is judged as "Optimal" if and only if its overall performance score is below a preset threshold representing "acceptable technical performance".

[0144] For example, the system simulates a preliminary scheme for the "D->G path". The comprehensive performance indicator (KPI) evaluation shows the expected time delay of "node G". The expected resource consumption is 0.8 seconds (lower than the 1.5-second baseline). An increase of 5kW (within economic limits) is expected, with anticipated voltage fluctuations. The result is -2.5% (less than the 3% safety threshold). The overall performance score calculated by the objective function is very low, but still within an acceptable range. Therefore, the system obtains the following adaptation result for the dynamic adjustment mechanism: {"Path":"D->G","Result":"Optimal","Estimated_Cost_Score":"25.4"}.

[0145] It should be noted that determining the final path based on the adaptation results of the dynamic adjustment mechanism, thus obtaining the stability improvement path, is the final decision step in S17. Based on the adaptation results (e.g., confirming the "D->G" path as "Optimal"), the system solidifies it as the final execution scheme. The system will replace the instructions in S16 with this scheme and continuously monitor the execution effect of the new path. If the voltage is stable, the execution path ("D->G path") verified as safe, stable, and efficient is ultimately determined as the stability improvement path.

[0146] It should be noted that the method described in this invention is a continuously operating closed-loop adaptive control system. After the "stability improvement path" is determined and executed in step S17, the actual execution effect of this path, including but not limited to the final node voltage distribution, actual equipment response time, and load transfer results, will be recorded, quantified, and fed back to a dynamically updated historical database. This database containing the latest execution results will, on the one hand, serve as updated "preset historical data" for iterative training of the machine learning classification model (such as SVM) in step S11 to improve the accuracy of subsequent fault mode matching; on the other hand, it will also serve as new training samples to optimize the parameters of the machine learning regression model (such as random forest) in step S16, making its prediction of "load transfer failure probability" more accurate. After completing the feedback update of this scheduling, the system will automatically return to step S11, reacquire the latest real-time data of the distribution network, and start a new round of adaptive control loop based on the learned and optimized model and historical data. This closed-loop mechanism of continuous learning and iteration from S17 to S11 ensures that this method can continuously adapt to changes in the power grid state and dynamically optimize its control strategy.

[0147] In summary, this invention constructs a full-chain adaptive control process, from "real-time data perception and dynamic topology construction" to "network loss calculation and voltage stability judgment," then to "switch state matrix construction and load transfer path optimization," and finally to "simulated scheduling adjustment and closed-loop execution." This deeply integrates dynamic network state analysis with multi-device coordinated control and innovatively introduces an optimization mechanism that assesses and adjusts the operation sequence, overload risk, and failure probability before execution. This solves the technical problems of delayed response, difficulty in coordinating multiple device actions, and stability risks in load transfer when facing dynamic power flow in existing technologies. It realizes adaptive control of distribution network switchgear and significantly improves the stability and reliability of power grid operation.

[0148] Reference Figure 2 The second embodiment of the present invention provides an adaptive control system for power distribution network switchgear based on machine learning, comprising:

[0149] The data processing module is used to acquire real-time data of the power distribution network, analyze the real-time data of the power distribution network, and determine the change information of the dynamic power flow direction.

[0150] The topology construction module is used to construct the current network topology based on the dynamic power flow direction change information, and to perform power flow calculation on the current network topology in combination with preset branch impedance parameters to obtain preliminary results of node voltage and power distribution.

[0151] The alarm generation module is used to calculate the node power balance state based on the preliminary results of the node voltage and power distribution, and obtain the voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, it generates local overvoltage alarm information.

[0152] The scheme formulation module is used to construct a switch state matrix based on the alarm information of the local overvoltage and determine a temporary scheme for equipment load distribution.

[0153] The path optimization module is used to analyze the operation sequence constraints and assess the overload risk based on the temporary scheme of the equipment load allocation, and obtain an optimized combination of load transfer paths.

[0154] The instruction adjustment module is used to simulate power dispatching based on the optimized combination of the load transfer paths, determine the possibility of load transfer failure, and obtain the adjusted dispatching instructions.

[0155] The execution monitoring module is used to perform real-time load transfer operations according to the adjusted scheduling instructions, monitor changes in node voltage distribution, and determine stability improvement paths.

[0156] It should be noted that the machine learning-based adaptive control system for distribution network switchgear provided in this embodiment of the invention is used to execute all the process steps of the machine learning-based adaptive control method for distribution network switchgear in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0157] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a machine learning-based adaptive control program for distribution network switchgear. When the processor executes the computer program, it implements the steps described in the various machine learning-based adaptive control method embodiments for distribution network switchgear, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the data processing module.

[0158] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0159] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0160] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0161] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0162] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0163] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0164] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A machine learning based adaptive control method for power distribution network switchgear, characterized in that, include: Acquire real-time data of the power distribution network and analyze the real-time data of the power distribution network to determine the change information of the dynamic power flow direction; Based on the information on the change in the dynamic power flow direction, the current network topology is constructed, and power flow calculation is performed on the current network topology in combination with preset branch impedance parameters to obtain preliminary results of node voltage and power distribution. Based on the preliminary results of the node voltage and power distribution, the node power balance state is calculated to obtain the voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, a local overvoltage alarm message is generated. Based on the alarm information of the local overvoltage, a switch state matrix is ​​constructed to determine a temporary scheme for equipment load distribution; Based on the temporary load distribution scheme of the equipment, the operation sequence constraints are analyzed and the overload risk is assessed to obtain an optimal combination of load transfer paths; Based on the optimized combination of the load transfer paths, power dispatch is simulated, and the possibility of load transfer failure is determined to obtain the adjusted dispatch instructions. According to the adjusted scheduling instructions, perform real-time load transfer operations and monitor changes in node voltage distribution to determine stability improvement paths; The step of performing real-time load transfer operations according to the adjusted scheduling instructions, monitoring changes in node voltage distribution, and determining a stability improvement path includes: performing the real-time load transfer operations according to the adjusted scheduling instructions, and monitoring the node voltage fluctuation range in real time; if the fluctuation range exceeds a preset voltage fluctuation threshold, determining the distribution location of abnormal nodes; performing path planning based on the distribution location of abnormal nodes to obtain a preliminary stability improvement path scheme; performing instruction execution efficiency analysis based on the preliminary stability improvement path scheme to obtain the adaptation result of the dynamic adjustment mechanism; and performing final path determination based on the adaptation result of the dynamic adjustment mechanism to obtain the stability improvement path.

2. The machine learning based power distribution network switchgear adaptive control method of claim 1, wherein, The acquisition and analysis of real-time data from the power distribution network to determine changes in the direction of dynamic power flow includes: The real-time data of the power distribution network is marked with abnormal data points to obtain the marked abnormal data points; The marked abnormal data points are compared over time to obtain continuous change characteristics; The continuous change characteristics are matched with preset historical data to obtain a judgment result; Based on the judgment results, the abnormal trends in the direction of power flow are analyzed to determine the change information of the dynamic power flow direction.

3. The adaptive control method for distribution network switchgear based on machine learning according to claim 1, characterized in that, The process of constructing the current network topology based on the dynamic power flow direction change information, and performing power flow calculations on the current network topology in conjunction with preset branch impedance parameters to obtain preliminary results of node voltage and power distribution includes: Based on the information on the change in the dynamic power flow direction, graph theory analysis is performed to obtain the network topology diagram; Based on the network topology diagram, a matching process is performed using the preset branch impedance parameters to obtain the impedance distribution; If the line value in the impedance distribution exceeds the preset impedance threshold range, then the load distribution characteristics are obtained; Based on the load distribution characteristics and the network topology diagram, an overlay analysis is performed to determine the preliminary results of the node voltage and power distribution.

4. The adaptive control method for distribution network switchgear based on machine learning according to claim 1, characterized in that, Based on the preliminary results of the node voltage and power distribution, the node power balance state is calculated to obtain a voltage stability index. If the voltage stability index does not meet a preset voltage threshold condition, a local overvoltage alarm message is generated, including: Based on the preliminary results of the node voltage and power distribution, the difference between injected and consumed power is calculated to obtain power balance state data; Based on the power balance state data, the voltage value of each node is analyzed, and the voltage value is used as the voltage stability index. If the voltage stability index exceeds the preset voltage threshold condition, an alarm message for local overvoltage is generated for the abnormal node.

5. The adaptive control method for distribution network switchgear based on machine learning according to claim 1, characterized in that, The step of constructing a switch state matrix based on the local overvoltage alarm information and determining a temporary load distribution scheme for the equipment includes: Based on the alarm information of the local overvoltage and in conjunction with the preset power grid topology data, the voltage distribution in the affected area is determined; Based on the voltage distribution, extract equipment operating data and compare and analyze the equipment operating data with preset equipment capacity information to determine abnormal equipment status records; Based on the device status anomaly records, construct the switch status matrix; Perform matrix calculations on the switch state matrix to determine an adjustable switch combination scheme; Based on the adjustable switch combination scheme, the abnormal equipment status records, and the preset power grid topology data, load allocation processing is performed to generate a temporary load allocation scheme for the equipment.

6. The adaptive control method for distribution network switchgear based on machine learning according to claim 1, characterized in that, The step of analyzing operation sequence constraints and assessing overload risk based on the temporary load allocation scheme of the equipment, and obtaining an optimized combination of load transfer paths, includes: A temporary solution for the device load distribution is analyzed in conjunction with preset operation sequence constraints to obtain a combination of operation sequences; Based on the combination of the aforementioned operation sequence, an overload risk assessment is performed, and a risk assessment report is obtained. Based on the risk assessment report, path planning is performed to determine the optimal combination of load transfer paths.

7. The adaptive control method for distribution network switchgear based on machine learning according to claim 1, characterized in that, The step of simulating power dispatch based on the optimized combination of load transfer paths, determining the probability of load transfer failure, and obtaining adjusted dispatch instructions includes: Based on the optimized combination of the load transfer paths, power dispatch simulation is performed to obtain the load distribution status; Based on the aforementioned load distribution status, a load transfer failure risk assessment is performed, and a risk level report is obtained; Based on the optimized combination of the risk level report and the load transfer path, an instruction matching degree analysis is performed to determine the adjusted scheduling instruction.

8. An adaptive control system for power distribution network switchgear based on machine learning, characterized in that, include: The data processing module is used to acquire real-time data of the power distribution network, analyze the real-time data of the power distribution network, and determine the change information of the dynamic power flow direction. The topology construction module is used to construct the current network topology based on the dynamic power flow direction change information, and to perform power flow calculation on the current network topology in combination with preset branch impedance parameters to obtain preliminary results of node voltage and power distribution. The alarm generation module is used to calculate the node power balance state based on the preliminary results of the node voltage and power distribution, and obtain the voltage stability index. If the voltage stability index does not meet the preset voltage threshold condition, it generates local overvoltage alarm information. The scheme formulation module is used to construct a switch state matrix based on the alarm information of the local overvoltage and determine a temporary scheme for equipment load distribution. The path optimization module is used to analyze the operation sequence constraints and assess the overload risk based on the temporary scheme of the equipment load allocation, and obtain an optimized combination of load transfer paths. The instruction adjustment module is used to simulate power dispatching based on the optimized combination of the load transfer paths, determine the possibility of load transfer failure, and obtain the adjusted dispatching instructions. The execution monitoring module is used to perform real-time load transfer operations according to the adjusted scheduling instructions, monitor changes in node voltage distribution, and determine stability improvement paths. The step of performing real-time load transfer operations according to the adjusted scheduling instructions, monitoring changes in node voltage distribution, and determining a stability improvement path includes: performing the real-time load transfer operations according to the adjusted scheduling instructions, and monitoring the node voltage fluctuation range in real time; if the fluctuation range exceeds a preset voltage fluctuation threshold, determining the distribution location of abnormal nodes; performing path planning based on the distribution location of abnormal nodes to obtain a preliminary stability improvement path scheme; performing instruction execution efficiency analysis based on the preliminary stability improvement path scheme to obtain the adaptation result of the dynamic adjustment mechanism; and performing final path determination based on the adaptation result of the dynamic adjustment mechanism to obtain the stability improvement path.

Images