Switching control method and device for network construction type power system

Abstract

The invention relates to the technical field of electric power systems, and discloses a switching control method and device for a network construction type electric power system, and the method comprises the steps: obtaining an abnormal signal set, carrying out the broadcasting of adjacent nodes, and obtaining a consensus confirmation result; performing priority regulation and control according to a consensus confirmation result to obtain a global control strategy; carrying out load fluctuation analysis and coordination synchronization according to the global control strategy to obtain a stable voltage distribution scheme; and real-time feedback and iterative optimization are carried out on the scheme, so that the fine adjustment precision is enhanced. According to the method, the control precision and the response speed of the network construction type power system in a dynamic complex scene can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular to a switching control method and device for a network-structured power system. BACKGROUND

[0002] At present, the power system is the core pillar of energy supply in modern society, and its stable operation is crucial for economic development and people's livelihood protection. With the widespread access of renewable energy and the rapid development of distributed power supply, the complexity of the power system has increased significantly. The traditional centralized operation control method has problems of slow response speed and heavy communication pressure when dealing with complex grid states, and gradually cannot meet the dynamic and changing needs.

[0003] In one prior art, in order to realize multi-node cooperation and rapid response, edge intelligent algorithms and consensus algorithms are introduced. Although the edge intelligent algorithm can distribute computing tasks and improve data processing efficiency, its consistency in multi-node coordination still faces challenges. At the same time, when the state judgment of the nodes needs to be unified through the consensus algorithm, the prior art is difficult to ensure that all nodes reach an agreement in a short time in a complex power grid, which may cause deviation or delay in the generation of control instructions, and further exacerbate the instability of the power grid.

[0004] Therefore, in the prior art, there is a problem that it is difficult for the power system to realize fast and accurate switching control among multiple nodes in a dynamic and complex scenario. SUMMARY

[0005] The present application provides a switching control method and device for a network-structured power system to solve the problem that the power system is difficult to realize fast and accurate switching control.

[0006] In a first aspect, to solve the above technical problems, the present application provides a switching control method for a network-structured power system, comprising: Obtaining an abnormal signal set, broadcasting the abnormal signal set to adjacent nodes to obtain a consensus confirmation result; According to the consensus confirmation result, the upper node performs priority control to obtain a global control strategy; If the global control strategy indicates that mode switching is needed, the lower node is notified to receive the global control strategy, then the load fluctuation of the lower node is analyzed to obtain fluctuation characteristic data, and if the fluctuation characteristic data exceeds the preset load fluctuation range threshold, the local resource allocation is adjusted to obtain a refined output instruction set; The refined output instruction set is coordinated and synchronized to obtain a coordinated action instruction set; According to the coordinated action instruction set, the node is distributed with the instruction set and adjusted in real time to obtain a stable voltage distribution scheme; Real-time acquisition of transient fluctuation voltage of each device in the stable voltage distribution scheme, if the transient fluctuation voltage exceeds the preset fluctuation voltage threshold, iterative optimization parameter configuration, obtain enhanced transient fine-tuning precision scheme.

[0007] In a second aspect, the application provides a switching control device for a networked power system, comprising: A consensus confirmation module is configured to obtain an abnormal signal set, broadcast the abnormal signal set to adjacent nodes, and obtain a consensus confirmation result. A strategy generation module is configured to perform priority control based on the consensus confirmation result, and obtain a global control strategy. An instruction refining module is configured to notify a lower-level node to receive the global control strategy if the global control strategy indicates that mode switching is required, then perform load fluctuation analysis on the lower-level node, obtain fluctuation characteristic data, and adjust local resource allocation if the fluctuation characteristic data exceeds a preset load fluctuation range threshold, and obtain a refined output instruction set. A coordination and synchronization module is configured to coordinate and synchronize the refined output instruction set, and obtain a coordinated action instruction set. A voltage distribution module is configured to distribute and real-time adjust the coordinated action instruction set to the nodes based on the coordinated action instruction set, and obtain a stable voltage distribution scheme. An iterative optimization module is configured to real-time acquire the transient fluctuation voltage of each device in the stable voltage distribution scheme, and if the transient fluctuation voltage exceeds the preset fluctuation voltage threshold, iterative optimization parameter configuration is performed to obtain an enhanced transient fine-tuning precision scheme.

[0008] Compared with the prior art, the application has the following beneficial effects: (1) The application broadcasts the data collected by the edge nodes to obtain the consensus confirmation result between nodes, and then transmits the unified fluctuation evaluation to the upper-level node to form a global strategy, avoiding the problem of large communication pressure and slow response caused by processing massive raw data by a single central node, and enabling fast and accurate judgment of the authenticity and impact range of abnormal events, providing high-quality input for the formulation of a global strategy, and significantly improving the response speed and decision accuracy of the system in response to sudden conditions.

[0009] (2) After the lower-level node receives the global strategy, the application first performs local load fluctuation analysis, and when the fluctuation exceeds the safety threshold, the edge intelligent mechanism is activated to perform transient fine-tuning of local resource allocation, thereby obtaining refined output instructions, which can better adapt to the real-time working conditions of each node, effectively avoiding the instability risk that may be caused by the mismatch between the upper-level strategy and the actual local situation due to communication delay, and enhancing the safety of the control instructions and the dynamic adaptability of the system.

[0010] (3) The application can collect the instantaneous fluctuation data of each device to form feedback after the instruction distribution, and when the fluctuation amplitude continuously exceeds the threshold value, the parameter configuration of the consistency algorithm is iteratively optimized through the particle swarm optimization algorithm, the core coordination algorithm is dynamically adjusted according to the control effect, so that the instantaneous fine-tuning precision of subsequent control is continuously enhanced, and the long-term improvement of system performance and the high adaptability to complex power grid environment are realized. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is a switching control method flow diagram of a networked power system provided by the first embodiment of the application; Figure 2 is a switching control device structure diagram of a networked power system provided by the second embodiment of the application. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0013] With reference to Figure 1 The first embodiment of the application provides a switching control method of a networked power system, including the following steps: S11, an abnormal signal set is obtained, and adjacent node broadcasting is performed on the abnormal signal set to obtain a consensus confirmation result; S12, according to the consensus confirmation result, priority control is performed by an upper layer node to obtain a global control strategy; S13, if the global control strategy indicates that mode switching is needed, the lower layer node is notified to receive the global control strategy, then load fluctuation analysis is performed on the lower layer node to obtain fluctuation characteristic data, and if the fluctuation characteristic data exceeds a preset load fluctuation range threshold value, local resource allocation is adjusted to obtain a refined output instruction set; S14, the refined output instruction set is coordinated and synchronized to obtain a coordinated action instruction set; S15, according to the coordinated action instruction set, instruction set distribution and real-time adjustment are performed on the node to obtain a stable voltage distribution scheme; S16, the instantaneous fluctuation voltage of each device in the stable voltage distribution scheme is collected in real time, and if the instantaneous fluctuation voltage exceeds a preset fluctuation voltage threshold value, the parameter configuration is iteratively optimized to obtain an enhanced instantaneous fine-tuning precision scheme.

[0014] In step S11, an abnormal signal set is obtained, the abnormal signal set is broadcasted to neighboring nodes, and a consensus confirmation result is obtained, including: An instantaneous voltage fluctuation data set of the edge node is obtained. The instantaneous voltage fluctuation data set is subjected to noise suppression processing to obtain a first voltage data set. If the voltage value in the first voltage data set exceeds a preset voltage threshold, an abnormal signal mark is performed to obtain a first abnormal signal mark set. The instantaneous voltage fluctuation data set is subjected to fluctuation amplitude analysis to obtain a fluctuation amplitude. If the fluctuation amplitude exceeds a preset fluctuation amplitude threshold, an abnormal signal is identified in combination with the first abnormal signal mark set to obtain an abnormal signal set. According to the abnormal signal set, local state information is generated, the local state information is sent to neighboring nodes, instantaneous fluctuation voltage data fed back by the neighboring nodes is obtained, and the instantaneous fluctuation voltage data is integrated to obtain a consensus confirmation result.

[0015] It should be noted that in a distributed power supply system, the edge node is a front-end unit for data acquisition, responsible for real-time monitoring of the operating state of the power system. By deploying high-precision sensors on the edge node, the voltage instantaneous value in the power grid can be continuously and continuously acquired. For example, in a distributed power supply system of a wind power plant, the sensors integrated on the edge computing node can acquire voltage data at a preset sampling frequency (for example, once per second). Within a certain time period, the collected data form an instantaneous voltage fluctuation data set, for example, containing a series of instantaneous voltage values: {220.5V, 221.2V, 219.8V, 222.0V, 220.8V,...}. This data set is dynamically updated, providing real-time raw data input for subsequent noise suppression and abnormal analysis.

[0016] It should be noted that the moving average filtering method is used to suppress noise in the data, and the average value of every 5 consecutive sampling points is taken, so that the processed data (i.e. the first voltage data set) is smoother, and can more truly reflect the fluctuation trend of the voltage. Subsequently, the voltage data points exceeding the voltage threshold range are preliminarily marked as abnormal, thereby quickly screening out potential voltage abnormal events.

[0017] It is worth noting that the voltage threshold range (e.g. ±5% of the rated voltage 220V, i.e. 209V to 231V) is determined based on the power industry's power quality standards. Setting the threshold to ±5% is a generally accepted stricter standard, aiming to effectively identify significant voltage deviation events that can cause damage to electrical equipment or indicate grid instability. For example, taking the arithmetic mean of the voltage values of 5 consecutive sampling points within the window (e.g. 220.5V, 221.2V, 219.8V, 222.0V, 220.8V) gives a smoothed voltage value of 220.86V, which is a data point in the first voltage data set. Then, set the voltage threshold range, for example, ±5% of the rated voltage 220V, i.e. 209V to 231V. If a filtered voltage value is 235V, which is obviously beyond the threshold range, the system will mark it as an abnormal signal and record its timestamp, forming the first abnormal signal marking set.

[0018] It is worth noting that relying solely on the instantaneous voltage value exceeding the threshold to determine the abnormality may have false positives. Therefore, the sliding window algorithm with a window size of 5 is used to calculate the fluctuation amplitude (such as standard deviation) of the voltage data. When the calculated fluctuation amplitude (such as standard deviation) exceeds the fluctuation amplitude threshold, it is combined with the marked abnormal points in the first abnormal signal marking set for double confirmation, so as to more accurately identify the real voltage abnormal signal and form the final abnormal signal set.

[0019] It is worth noting that the fluctuation amplitude threshold is determined by statistical analysis of a large amount of voltage data under normal operating conditions. First, collect voltage data of the equipment under normal operating conditions for a long enough time; then, calculate the standard deviation of all windows using the same size sliding window (size 5) to form a fluctuation amplitude distribution set; finally, set the threshold according to the statistical characteristics of the distribution. For example, set the threshold to the 99.7th percentile of the distribution, which means that any fluctuation exceeding this threshold is an extremely small probability event (probability less than 0.3%) under normal circumstances, and therefore has sufficient reason to be considered as a potential anomaly.

[0020] For example, calculate the standard deviation of the voltage data within the window (e.g. 220.3V, 221.5V, 219.7V, 222.1V, 220.9V) to get the fluctuation amplitude. Assuming the calculated standard deviation is 1.8V. The pre-set fluctuation amplitude threshold is 1.5V, since 1.8V > 1.5V, it indicates that the voltage fluctuation is severe at this time. At the same time, if the data near this time point is also marked in the first abnormal signal marking set (for example, the instantaneous voltage value reaches 235V), the system makes a comprehensive judgment and confirms that it is a real abnormal event, and records its detailed information (such as abnormal timestamp 2025-08-14 12:00:02, fluctuation amplitude 1.8V) to the abnormal signal set.

[0021] It should be noted that when an edge node identifies an abnormal signal, the node will generate local state information containing key information (such as timestamp, voltage and fluctuation amplitude) according to the confirmed abnormal signal set, and broadcast it to other nodes adjacent in geographical position. After receiving the information, the adjacent node analyzes the voltage data collected at the same time to determine whether there is a similar abnormal phenomenon. The specific implementation process is as follows: after receiving the information, the node locates the corresponding time window in the local data according to the timestamp, calculates the fluctuation amplitude of the voltage data in the window, and calculates the relative difference, If the relative difference is less than 20%, it is determined that there is a similar abnormal phenomenon. Through this multi-node cooperative verification method, the reliability of abnormal judgment can be greatly improved, and finally a consensus confirmation result is formed.

[0022] It is worth noting that the adjacent nodes in the geographical position of the node are determined by querying a centrally managed node geographical position database. The database records the positions and states of all edge nodes in the network. The information in the database is mainly recorded when the node is deployed. The record of each node contains four core attributes: node unique identifier, geographical position, communication address (such as IP address, which tells other nodes how to contact it) and running state (such as "online" or "offline", which ensures communication with only normally running nodes). When node A needs to find neighbors, it will initiate a simple query to the database: "Please return the communication addresses of all nodes within 500 meters of me and with the current state of 'online'." After the database performs this spatial query, it returns a list containing the addresses of nodes B and C.

[0023] For example, node A confirms an abnormal signal with a fluctuation amplitude of 1.9V at 2025-08-14 12:00:03. Before broadcasting this information, node A first initiates a query to the node geographic location database: "Please return the communication addresses of all nodes within 500 meters of me and currently in the state of 'online'." After the database query, the communication addresses of adjacent nodes B and C are returned. Node A generates a local state information according to this, for example: "{Time: 2025-08-14 12:00:03, Voltage: 221.8V, Fluctuation Amplitude: 1.9V}", and sends this information to nodes B and C through a low-power wide-area network (such as LoRa). After receiving the information, node B locates the same period time window in its own data according to the timestamp, and calculates that its fluctuation amplitude is 1.85V. Then, it performs similarity judgment and calculates the relative difference: |1.85V - 1.9V| / 1.9V≈2.6%. Since 2.6% is less than 20%, node B judges that there is a similar abnormality and feeds back a confirmation signal to node A. Node C performs the same process and finds that its fluctuation amplitude in the same period is 1.92V, and the calculated relative difference is also much less than 20%, so it also feeds back a confirmation signal to node A. After collecting these feedbacks, node A forms a consensus confirmation result of "3 nodes have detected similar abnormalities, and the fluctuation amplitude set is {A: 1.9V, B: 1.85V, C: 1.92V}", thereby confirming the universality and authenticity of the abnormal event.

[0024] In step S12, according to the consensus confirmation result, the priority is regulated by the upper node to obtain a global control strategy, including: The fluctuation amplitude data of each node in the consensus confirmation result is aggregated to obtain a unified fluctuation amplitude evaluation value; If the unified fluctuation amplitude evaluation value exceeds the preset fluctuation evaluation threshold value, the upper node performs hierarchical control according to the unified fluctuation amplitude evaluation value to obtain a hierarchical control instruction; The running state data of the distributed power supply is obtained according to the hierarchical control instruction, and is classified, and then the priority is allocated according to the classification result to obtain a power supply regulation priority; The power supply regulation priority is broadcast to the distributed power supply node to obtain a global control strategy.

[0025] It should be noted that the arithmetic mean method is used for data aggregation, and the fluctuation amplitude data from different nodes are comprehensively calculated to obtain a macro evaluation of the overall system state. In this way, accidental data differences of individual nodes can be smoothed out, and a unified fluctuation amplitude evaluation value reflecting the overall fluctuation trend of the system can be generated, providing a key basis for subsequent control decisions. For example, assume that nodes A, B, and C in the system detect voltage fluctuation amplitudes of 1.9V, 1.85V, and 1.92V at 2025-08-14 12:01:00. The arithmetic mean of these data is calculated to obtain a unified instantaneous fluctuation evaluation value of about 1.89V. This evaluation value integrates the monitoring results of multiple nodes as a quantitative indicator of the overall voltage stability of the current system.

[0026] It should be noted that the system will set a fluctuation evaluation threshold to determine whether the current system fluctuation is within an acceptable range. The fluctuation evaluation threshold is determined by system simulation and optimization on a historical operation data set. The specific method is as follows: first, a simulation database containing various historical operating conditions (such as normal operation, slight disturbance, critical instability, etc.) is established; then, backtesting simulation is performed on the historical operating conditions under a series of candidate thresholds (such as 1.2V, 1.5V, 1.8V), and the overall benefit of the system after activating the hierarchical control under each threshold is evaluated (for example, a performance indicator that integrates system stability, control cost, and power quality); finally, the candidate value that maximizes the overall benefit indicator of the system (for example, 1.5V) is selected as the final preset threshold. When the calculated unified fluctuation amplitude evaluation value exceeds this threshold, it indicates that the system may be at risk of instability. At this time, the hierarchical control architecture is activated, and the evaluation value is sent to the node of the previous level (such as the regional control center). The upper node generates corresponding hierarchical control instructions according to the over-standard rate of the evaluation value.

[0027] It is worth noting that the specific steps for generating hierarchical control instructions by the upper node are as follows: let the fluctuation evaluation threshold be T, and the current calculated evaluation value be V. The system first calculates the over-standard rate Then according to the value of P, different levels of response are triggered. First level instruction: 0% < P≤ 30%. The instruction requires all key nodes (such as distributed power supply, energy storage unit) in the region to report detailed operating state data immediately, including but not limited to real-time power, voltage, current, equipment temperature, etc. The data reporting frequency is increased from the conventional minute level to the second level. Second level instruction: 30% < P≤ 100%. On the basis of executing the first level instruction, additional power adjustment instructions are issued. For example, the nodes in the region are required to temporarily reduce the power by 5% - 10% on the basis of the existing output power. Third level instruction: P > 100%. Emergency control plan is executed immediately. For example, temporarily remove part of the non-critical load in the region, and temporarily disconnect some distributed power supply with severe fluctuations from the grid.

[0028] For example, if the fluctuation evaluation threshold is 1.5V, and the currently calculated unified fluctuation amplitude evaluation value is 1.89V, which is obviously beyond the threshold. After receiving the evaluation value of 1.89V, the upper node (regional control center) calculates the over-standard rate: P = (1.89V - 1.5V) / 1.5V = 0.26 = 26%. Match the response level: 26% is in the interval (0%, 30%], trigger the first level response. The system automatically generates and issues control instructions, the content is: "require to obtain the detailed operating state data of all distributed power supply in the region immediately, and increase the data reporting frequency to every 5 seconds." This prepares for the accurate regulation and control that follows.

[0029] It should be noted that after executing the hierarchical control instructions issued by the upper node, the system obtains the real-time operating state data of each distributed power source, including output power, voltage, and operating time. In order to achieve differentiated and refined regulation and control, the system uses a clustering algorithm (such as the K-means method, with K = 3) to classify these operating state data. First, the operating state data is normalized by maximum and minimum. Then, K power sources are randomly selected from all distributed power sources as the initial three "cluster center points" (Centroids). For each power source that is not selected as a cluster center, calculate its "distance" to the K cluster center points. This distance is usually the Euclidean distance, which takes into account the differences in power, voltage, and operating time. Each power source is assigned to the cluster center point closest to it. In this way, all power sources are divided into K temporary clusters. For each cluster, the center point is recalculated. The new center point is the average of all power source data in the cluster (for example, the average power, average voltage, and average operating time of all power sources in the cluster). This new center point better represents the overall characteristics of the cluster. Iterative updating is performed, and eventually the system obtains K stable clusters, each containing a group of distributed power sources with similar operating states, and a center point describing the common characteristics of the cluster. According to the classification results, assign a priority for regulation and control to each power source or each group of power sources. For example, high-priority cluster: output result is (the center point may represent "high output power, slightly high voltage, moderate operating time"). Medium-priority cluster: output result is (the center point may represent "moderate power, stable voltage, long operating time"). Low-priority cluster: output result is (the center point may represent "low output power, low voltage, or excessively short / long operating time").

[0030] For example, assume that there are three distributed power sources in the region, and their operating state data are: power source 1 (500 kW, 220.5 V), power source 2 (480 kW, 221.2 V), and power source 3 (510 kW, 219.8 V). After the K-means algorithm is run, the three power sources are divided into two groups. Cluster 1: contains power source 1 (500 kW, 220.5 V) and power source 3 (510 kW, 219.8 V). The center point feature of this cluster is “very high output power and large voltage fluctuation”. Cluster 2: contains power source 2 (480 kW, 221.2 V). The feature of this cluster is “high output power, stable and slightly high voltage”. The system analysis considers that cluster 2 (power source 2) has stable and slightly high voltage, so it is appropriate to reduce its power to effectively reduce the overall voltage of the region without causing impact to itself, and thus is assigned a “high priority” regulation. Cluster 1 (power source 1 and power source 3) has higher power, but its voltage has some fluctuation, so direct large-scale adjustment may exacerbate instability. Therefore, they are assigned a “secondary priority” and are used as backup regulation resources or only small-scale adjustment is performed.

[0031] It should be noted that after the regulation priority of each distributed power source is determined, the upper layer scheduling module broadcasts the priority information to the corresponding distributed power source node through the communication network. These information containing priority and specific regulation instructions together constitute the global control strategy. This strategy ensures the coordinated action of all nodes, thereby achieving stable control of the entire power system.

[0032] For example, according to the determined priority, the generated global control strategy can be: the high-priority power source needs to increase the output power by 10%, the secondary-priority power source maintains the current output unchanged, and the low-priority power source needs to reduce the output power by 5%. This series of instructions is issued to each power source node through the wireless communication module, and each node executes the corresponding adjustment after receiving the instructions, thereby collectively completing the balance of system load and ensuring the stable operation of the power grid.

[0033] In step S13, if the global control strategy indicates that mode switching is needed, the lower layer node is notified to receive the global control strategy, and then load fluctuation analysis is performed on the lower layer node to obtain fluctuation feature data. If the fluctuation feature data exceeds the preset load fluctuation range threshold, the local resource allocation is adjusted to obtain a refined output instruction set, including: If the global control strategy indicates mode switching, the lower layer node performs rule analysis on it, and performs integrity verification on the analyzed data to obtain strategy analysis data; Real-time collection of load fluctuation data of the device applying the strategy analysis data is performed, and time series decomposition is performed on the load fluctuation data to obtain fluctuation feature data; If the fluctuation feature data exceeds the preset load fluctuation range threshold, the operation parameters of the distributed power supply are instantaneously fine-tuned to obtain a fine-tuned execution sequence. The fine-tuned execution sequence is subjected to refining processing to obtain a refined output instruction set.

[0034] It should be noted that mode switching refers to the need to adjust the output power of the power supply node, and the lower node first needs to parse the strategy to generate executable local instructions. This process is completed through pre-established parsing rules, for example, "mode = energy saving, power reduction = 10%, time = 2025-08-14 12:30:00". To ensure that the instructions are not tampered with during transmission, especially in a wireless communication environment, security protocols also need to be combined to verify the integrity of the parsed data. SHA-256 algorithm is used to generate a data hash value (the input received is the parsed JSON data, and the output is a 64-bit hexadecimal string), which is compared with the verification value attached to the strategy to effectively ensure the reliability and security of the instructions.

[0035] It should be noted that the strategy is encapsulated into a JSON object, where each "key" represents an instruction field, and each "value" represents a specific parameter. All possible control fields are defined in advance, such as: command_id (instruction ID), target_node (target node), mode (operation mode), power_adjustment_value (power adjustment value), adjustment_type (adjustment type, such as percentage or absolute value), timestamp (effective timestamp). The value of a specific field (such as mode) is standardized, for example, 1 represents "energy saving mode", 2 represents "full power mode", and 3 represents "standby mode". It is determined that all node communication adopts this JSON structure.

[0036] For example, a global control strategy indicates that a node (ID: DN_007) needs to switch to energy saving mode due to load reduction. The JSON text is: {"command_id":"CMD_1693297800","mode":"energy_saving","power_adjustment":{"type":"percentage","value":-10},"target_node":"DN_007","timestamp":"2025-08-14T12:30:00Z"}. The specific data of this policy can be "mode = energy saving, power reduction = 10%, time = 2025-08-14 12:30:00". The lower node extracts these fields by parsing the rule to form policy parsing data. In the verification stage, the node calculates the hash value of the received policy data and compares it with the hash value attached to the policy. If they are consistent, the verification is passed, indicating that the data is complete and accurate, and can proceed to the next step of processing.

[0037] It should be noted that after the policy parsing data is verified, the system needs to monitor the load fluctuation of the device applying the policy in real time. Distributed power nodes are usually equipped with sensors to collect real-time load current, voltage and other parameters to form load fluctuation data. Using time series decomposition method, set the period to 30 minutes, extract the periodic load change characteristics of the time window content data, generate fluctuation characteristic data such as fluctuation amplitude and frequency.

[0038] It is worth noting that the time series decomposition method is a technique that splits time series data into several key components, allowing for a clearer understanding of its internal rules and abnormalities. For a given time window (e.g. set to 30 minutes), the original load data (Yt) is usually decomposed into three main components: trend component (Tt), reflecting the overall change direction of the load in a longer period; periodic component (St), revealing the repetitive, regular fluctuation pattern of the load in a specific period (e.g. set to 30 minutes); residual component (Rt), representing the remaining random, irregular or sudden fluctuations after removing the trend and periodic effects. Through this decomposition, the system can distinguish between "normal, periodic fluctuations" and "abnormal, sudden events".

[0039] For example, a sensor on a certain distributed power supply node detects that its load current drops from 2000A to 1800A in a short time. The system captures the continuous current data in a 30-minute time window containing this mutation event. The system applies the time series decomposition algorithm to the 30-minute data. After decomposition, three component time series curves are obtained. The system checks the residual component and finds that at the time of the current drop, a significant negative spike of -200A appears. This clearly identifies a burst event. The system analyzes the periodic component and finds that its amplitude is stable at ±50A in a complete cycle (30 minutes). Based on the above analysis, the system converts the original, unstructured time series data into structured feature data for decision-making. For example: {burst event amplitude: -200A, burst event amplitude rate: -10%, regular fluctuation frequency: 2 times / hour, regular fluctuation amplitude: 50A}.

[0040] It should be noted that the system will compare the extracted fluctuation feature data with the preset load fluctuation range threshold (fixed at 15%) to determine whether the current fluctuation is within an acceptable range. If the fluctuation feature data exceeds the threshold, it indicates that the system stability may be threatened, and an activation signal will be generated to trigger the instantaneous fine-tuning of local resources. The local resource allocation algorithm adjusts the operating parameters of the distributed power supply (such as inverter operating frequency) quickly and accurately based on this signal to generate a fine-tuned execution sequence.

[0041] It is worth noting that the essence of the local resource allocation algorithm is a fast "perception-decision-execution" system, and its core is a preset control rule library. When the algorithm is triggered by the activation signal, it will first analyze the fluctuation type (such as "high-frequency oscillation" or "voltage sag") and the extent of the violation (fluctuation amplitude rate-fluctuation amplitude threshold) contained in the signal. Then, it will query the built-in rule library to match the current fluctuation with the preset control strategy. For example, Rule A: if the fluctuation type is "high-frequency oscillation", execute the "active power suppression" strategy; Rule B: if the fluctuation type is "voltage sag", execute the "dynamic reactive power support" strategy. According to the matched strategy, the algorithm calculates a preliminary control target value (for example, new output power) based on the extent of the violation. Before generating the final instruction, the target value must pass safety verification to ensure that it does not exceed the physical limits of the power supply equipment (such as maximum rated power, power ramp rate limit, etc.). Once the verification is passed, the algorithm will package the safe control target into a specific execution sequence and issue it to the local controller (such as the inverter) to complete the instantaneous fine-tuning.

[0042] For example, assume that the system is set to have a safety fluctuation amplitude threshold of no more than 15%, and the current analysis obtains a fluctuation amplitude rate of 20%, which exceeds the threshold. A "trigger fine-tuning" activation signal is generated, which contains information that the fluctuation type is "high-frequency oscillation" and exceeds the threshold by 5%. After receiving this signal, the local resource allocation algorithm of a power supply node with a rated power of 500 kW and a current output of 480 kW immediately matches rule A: "active power flattening". According to the 5% excess, the algorithm calculates that the target output power should be reduced to 420 kW. The target value is verified to be within the safe operating range and ramp rate limit of the device. The algorithm finally generates a specific execution sequence, instructing the inverter to smoothly adjust the output power from 480 kW to 420 kW at a specified rate, thereby quickly stabilizing the system.

[0043] It should be noted that, in order to improve the transmission efficiency and reliability of the instructions, the execution sequence generated by the instantaneous fine-tuning needs to be refined. This process is completed through a distributed control protocol (such as an MQTT communication mechanism), and finally forms a refined output instruction set. The refined instruction set is distributed to each relevant node through an encrypted channel, ensuring that the instructions can be efficiently and accurately executed.

[0044] For example, an original execution sequence may contain multiple steps and intermediate states, and after refinement, these redundant information is merged. The final refined output instruction set may be a simple instruction such as "Node A: power 450 kW, execution time 2025-08-14 12:30:00". This refined instruction is then sent to node A through a secure encrypted channel, guiding it to complete the final resource allocation operation.

[0045] In step S14, the refined output instruction set is coordinated and synchronized to obtain a coordinated action instruction set, including: Obtain instantaneous coordination sequence data from the power supply node executing the refined output instruction set, and perform rule analysis to obtain key coordination feature data; Perform multi-node synchronization operation on the key coordination feature data, and detect node communication delay. If the node communication delay exceeds the preset communication delay range, adjust the synchronization frequency to obtain a sequence consistency state; According to the sequence consistency state, use anomaly detection to obtain an abnormal response frequency. If the abnormal response frequency exceeds the preset abnormal response frequency threshold, a trigger signal is generated; Perform power coordination processing on the trigger signal to obtain a coordinated action instruction set.

[0046] It should be noted that in order to realize the cooperative operation among multiple nodes, it is necessary to first obtain the instantaneous coordination sequence data from each distributed power supply node that has executed the refined output instruction set. These data are real-time operation state information generated by each node, usually including power output, voltage frequency and communication state, etc. After obtaining these raw data, the system will use the pre-established sequence analysis rules to decompose the data to obtain the average power and frequency change rate. For example, the sliding window average rule: the node collects the power output sequence (a total of 60 data points) every second in the past 60 seconds, and sets a 10-second sliding window. From the first point of the sequence, the average value of the 10 data points in the window is calculated. The window slides forward by one time step (1 second), and the calculation is repeated until the end of the sequence, obtaining a smoothed power sequence and the average power in the window. Linear regression slope: take time as X axis and smoothed power as Y axis. Perform one-dimensional linear regression fitting on the smoothed power sequence to obtain the equation w=at+b, where w is power, t is time, a is slope, and b is intercept. Extract the slope a of the equation, which is the power change rate.

[0047] For example, the sensors and edge devices of a certain node can collect its operation state data in real time, such as recorded power output of 510 kW, voltage frequency of 50.2 Hz, and communication delay of 20 ms. Assuming that the power data of the node fluctuates from 510 kW to 490 kW within 10 minutes, the sequence analysis rule will extract the dynamic feature, i.e. the change rate of 2 kW / min, from it. This change rate is a key coordination feature data.

[0048] It should be noted that after obtaining the key coordination feature data of each node, the system will use a consistency algorithm to ensure the coordination consistency among multiple nodes, such as synchronization of power distribution and frequency. This algorithm can use a consensus-based mechanism (such as Raft algorithm) to detect the state of each node in real time through heartbeat signals. In this process, the system will continuously monitor the communication delay between nodes. If the delay of a certain node exceeds the pre-set threshold range, the consistency algorithm will automatically adjust the synchronization frequency, such as reducing the sending frequency of the heartbeat signal, to reduce the network communication pressure, thereby ensuring that the sequence consistency state of the entire system remains stable.

[0049] It is worth mentioning that the main logic of the consistency algorithm can be decomposed into a closed-loop control process based on a "Leader-Follower" model, containing state monitoring and adaptive adjustment. After system startup or election, a leader will be generated in all nodes, and the rest of the nodes will become followers. The leader is the only initiator of synchronization. It sends a heartbeat message to all followers at a preset, fixed heartbeat interval (for example, 100 ms, i.e. 10 Hz). This message not only declares itself "alive", but also carries the data that needs to be synchronized (such as standard frequency reference, power allocation instructions, etc.). The follower passively receives the heartbeat from the leader and immediately replies to the leader with an acknowledgement message (ACK). The leader starts a timer for each follower when sending each heartbeat message. When the ACK from the corresponding follower is received, the timer is stopped. The time difference between sending and acknowledgement, i.e. the communication delay of the follower, is measured. The leader continuously compares the measured communication delay of each node with a preset delay threshold (for example, 120 ms). Once the communication delay of any follower exceeds the threshold, the system will trigger the "frequency adjustment mechanism". After triggering the adjustment mechanism, the leader will reduce the heartbeat frequency of the entire system according to the preset rules. The rules can be very simple, for example: increase the current heartbeat interval by a fixed step (for example, from 100 ms to 125 ms, the frequency is reduced from 10 Hz to 8 Hz). The leader will include the new heartbeat interval value in the next heartbeat message and broadcast it to all followers. The follower will update its own expectations for timeout judgment after receiving it. After the frequency is reduced, the leader continues to monitor the communication delay of all nodes at the new, slower frequency. If the communication delay of all nodes remains below the delay threshold for a sustained period of time (for example, 30 seconds), the leader will consider that the network condition has improved. After meeting the recovery conditions, the leader will gradually and reversely execute the adjustment mechanism, for example, restore the heartbeat interval from 125 ms to 100 ms, so that the system can run as efficiently as possible under the condition allowed by the network.

[0050] For example, in a five-node system (one leader and four followers), the steady-state heartbeat interval is 100ms and the latency threshold is 120ms. The leader detects that the ACK from node C arrives after 150ms. Since 150ms > 120ms, this triggers an adjustment mechanism. Based on the rules, the leader adjusts the heartbeat interval from 100ms to 125ms and notifies all nodes of the new interval. The system continues to operate at a frequency of 8Hz (125ms interval). During this period, the leader observes that the communication latency of all nodes, including node C, remains stable at around 90ms for 30 seconds. Determining that the network has recovered, the leader resets the heartbeat interval to 100ms, restoring the system to the optimal synchronization frequency of 10Hz.

[0051] It's important to note that the system uses the anomaly detection model to determine whether an exception response needs to be triggered based on the consistency state. The model generates a trigger signal to initiate the next coordinated action. The number of times an anomaly is triggered is the exception response frequency.

[0052] It is worth noting that the core of the anomaly detection model lies in the use of a deep neural network model to perform high-order temporal feature extraction and pattern recognition on dynamic key coordination feature data from multiple nodes, so as to more accurately judge the real-time stability of the entire system.

[0053] Model Architecture: The deep learning model used in this solution is a long short-term memory network autoencoder. This model is specifically designed to process and learn normal patterns in multivariate time series data. The input layer receives a multi-dimensional, normalized window of time series data. Each dimension corresponds to a key coordination feature (such as the frequency change rate of node A, the power stability of node B, the communication delay between nodes A and B, etc.). The encoder consists of a two-layer LSTM network. It reads the input time series data and gradually compresses it into a fixed-dimensional "state vector." This vector is a condensed representation of the operating state of the entire system within that time window. The decoder, also composed of a two-layer LSTM network, receives the state vector generated by the encoder and attempts to decode it to reconstruct the original input time series.

[0054] Model Training: This model is trained using unsupervised learning. It utilizes only a large amount of coordinated feature data from historically confirmed "healthy" and "stable" system operations. The mean squared error (MSE) is used as the loss function to measure the difference between the original input and the reconstructed output. Iterative training is performed using the Adam optimizer until the model can recover a wide range of normal operating modes with extremely low reconstruction error.

[0055] It is worth mentioning that the preset abnormal response frequency threshold is set based on statistical analysis of false alarm behavior of the anomaly detection model under normal working conditions. The method aims to set an empirical value that can filter out the model's own noise with a high probability. The process is as follows: let the anomaly detection model run on a large amount of historical normal operation data that has been confirmed to have no real faults, and count the distribution of the continuous occurrence times of the isolated abnormal signals (i.e. false alarms) generated by it. For example, if statistics show that 99% of false alarms are in the form of 1 or 2 consecutive occurrences, then setting the threshold to 3 is a reasonable choice. This ensures that only when the persistence of the abnormal signal significantly exceeds the known normal noise level, the system will judge it as a real event that needs attention. For example, the system continuously receives the coordination feature data stream of each node, such as the frequency change rate of a certain node being 0.2 Hz / min at t1, 0.5 Hz / min at t2, and 0.4 Hz / min at t3. The sequence data containing this dramatic change is sent to the trained anomaly detection model. The system therefore determines that an anomaly occurred at t2. If such anomalies identified by high reconstruction error occur continuously for 3 times in a short period of time, exceeding the preset abnormal response frequency threshold (such as 3), the system will generate a final trigger signal indicating that immediate intervention and adjustment of the system are needed.

[0056] It is worth mentioning that once the trigger signal is generated, the distributed power coordination algorithm will intervene to process and generate the final coordination action instruction set. The algorithm will consider the priority of coordination action and the balance rule of resource allocation. For example, the system will prioritize ensuring power supply to critical loads, and then optimize and adjust the output of each power node to achieve load balancing of the entire system and ensure stability.

[0057] It is worth mentioning that the essence of the distributed power coordination algorithm is a constrained optimization problem, and the core logic is to find the optimal resource allocation scheme under the premise of meeting all rigid constraints. The constraints are: the sum of the adjusted output power of all nodes must equal the total demand of the system; the adjusted output power of each node must be within its safe operating range. The objective function is set as: Minimize Var(P_1_new / P_max_1, P_2_new / P_max_2,...), where P_max_1 is the maximum available output power of node 1, P_1_new is the specific target value of node 1, and the remaining symbols can be obtained in the same way. The above objective function and all constraints are given to a standard constrained optimization solver (such as a quadratic programming QP solver) for calculation. The solver will give a solution that meets all constraints and makes the objective function optimal within milliseconds, i.e. a new power allocation scheme {P_1_new, P_2_new,...} for each node.

[0058] For example, in a system with a total power demand of 2000kW, there are three nodes (with maximum available powers of 800kW, 700kW, and 700kW, respectively), currently outputting 700kW, 650kW, and 600kW (totaling 1950kW). A trigger signal is received indicating a total power shortage of 50kW and uneven load. The optimization goal is set as: while making up the power shortage, minimize the variance of the load rate of each node. The constraints are defined as: P_1_new+P_2_new+P_3_new=2000; 0≤P_1_new≤800; 0≤P_2_new≤700; 0≤P_3_new≤700. The solver finds the solution that minimizes Var(P_1 / 800, P_2 / 700, P_3 / 700) under the above constraints. The optimal solution calculated by the solver is: P_1_new=680kW, P_2_new=660kW, P_3_new=660kW. The final coordination action instruction set is transmitted through protocols such as MQTT, which may include a specific instruction such as: "Node 1: power 680kW, execution time 2025-08-1412:45:00".

[0059] In step S15, according to the coordination action instruction set, the node is instructed to distribute the instruction set and adjust in real time to obtain a stable voltage distribution scheme, including: The coordination action instruction set is distributed to each relevant distributed power node, and communication delay data of the power node is collected in real time; if the communication delay data is lower than the preset communication delay range, a mode switching signal is generated; According to the mode switching signal, real-time response mode switching is performed to obtain real-time response configuration data; The real-time response configuration data is subjected to remote area voltage adjustment to generate voltage adjustment instructions; According to the voltage adjustment instructions, instruction distribution is performed to obtain a stable voltage distribution scheme.

[0060] It should be noted that after obtaining the coordinated action instruction set, the first step is to distribute it to each relevant distributed power supply node and synchronously collect the communication delay data of each node. The communication delay is collected through a "request-response" mechanism, in which the control center sends a very small and light data packet, usually referred to as a "heartbeat packet" or "Ping packet", to the target distributed power supply node. At the moment of sending, the control center records the current accurate time stamp T1. The communication module of the target node receives this specific data packet and immediately and automatically replies an acknowledgement data packet (Ping packet) to the control center. The control center receives this acknowledgement packet and records the received time stamp T2. The round-trip time (RTT) is calculated as RTT=T2-T1. If the communication delay of the node is lower than the preset communication delay threshold, it indicates that the network state of the node is good and can support faster response, at which time the system generates a mode switching signal as a trigger condition for entering the next stage.

[0061] It should be noted that the communication delay threshold is determined based on the technical specification of the total response time of the power system fast control task. In the power system, key regulation tasks such as fast frequency response or dynamic voltage support must be completed within a very short time window (e.g. 50-100 milliseconds) from "event occurrence" to "power response" to effectively maintain grid stability. This total response time budget consists of multiple links, including event detection, instruction calculation, communication transmission and physical execution. Therefore, the communication delay threshold is set to a small value (such as 20ms) to ensure that after the communication link consumes time, sufficient time is still reserved for other necessary links.

[0062] For example, in a remote area distributed power supply system containing 4 nodes, the control center periodically sends heartbeat packets to each node and records the response time. Through calculation, the communication delays of each node are obtained as follows: node 1 (15ms), node 2 (18ms), node 3 (22ms) and node 4 (25ms). If the communication delay threshold is 20ms, the system will determine that the communication delays of node 1 and node 2 (15ms and 18ms) are lower than the threshold, and therefore generate mode switching signals for these two nodes, indicating that they can switch to the real-time response mode to undertake more fast response regulation tasks.

[0063] It should be noted that after receiving the mode switching signal, the system will switch the running mode of the corresponding node to the real-time response mode according to the pre-established response mechanism configuration rule. For example, first-level response: for nodes with optimal network state and communication delay less than 20 milliseconds. The system will assign the highest priority (level 1) to such nodes, and set their data sampling frequency to the fastest, once every 200 milliseconds (i.e. 5 Hz). Second-level response: for nodes with good network state and communication delay between 20 milliseconds and 50 milliseconds. These nodes will be assigned a higher priority (level 2), and their data sampling frequency will be set to once every 500 milliseconds (i.e. 2 Hz) to accommodate their slightly higher network delay while ensuring faster response. Third-level response: for nodes with standard network state and communication delay between 50 milliseconds and 100 milliseconds. The system will assign them a medium priority (level 3), and set their data sampling frequency to once every 1000 milliseconds (i.e. 1 Hz) as the basic standard for real-time response mode. This ensures that the system can utilize the most reliable nodes for the fastest and most accurate response. After switching, the system will generate a real-time response configuration data, which defines the running parameters of the nodes in this mode, such as node priority, data sampling frequency, etc., providing a basis for subsequent accurate control.

[0064] For example, in a system containing 4 nodes, node 1 and node 2 have received the mode switching signal, and their current communication delays are 15ms and 18ms respectively. For node 1: its delay of 15ms matches the "first-level response" level in the rule table. For node 2: its delay of 18ms also matches the "first-level response" level. The system generates a configuration for node 1: {Node_ID: 1, mode: real-time, priority: 1, sampling frequency: 200ms}. The system generates a configuration for node 2: {Node_ID: 2, mode: real-time, priority: 1, sampling frequency: 200ms}. These two configuration data are issued to the corresponding nodes. Node 1 and 2 immediately switch to the highest level of real-time response mode, start reporting data at a frequency of every 200ms, and are given the highest priority in the control algorithm, ensuring that the system can utilize the best nodes in the network state to respond the fastest and most accurately.

[0065] It should be noted that after obtaining the new real-time response configuration data, the system will process the voltage state in remote areas to generate specific voltage adjustment instructions. This process can use a support vector machine (SVM) machine learning algorithm. Support vector machines can determine whether the current voltage is within the normal range and identify nonlinear voltage trends by classifying the collected voltage data. When an anomaly is detected, the algorithm can quickly generate accurate voltage adjustment instructions to correct the deviation.

[0066] It is worth mentioning that the core of the support vector machine algorithm is to use a trained classification model to find an optimal decision boundary (hyperplane) in a multi-dimensional feature space to accurately distinguish different voltage states (such as stable, low, high, oscillation, etc.), and generate control instructions accordingly.

[0067] Model structure: The model used in this solution is a support vector classifier. The input of the model is a feature vector extracted from the original voltage time series data. For example, not only the current voltage value, but also "average voltage in the past 10 seconds", "voltage change rate (slope)", "voltage fluctuation standard deviation" and other dimensions. The model uses a radial basis function (RBF) kernel, which is good at handling nonlinear problems and can map low-dimensional input features to high-dimensional space, so as to find an optimal linear boundary (hyperplane) in this high-dimensional space to effectively classify various complex voltage states. The direct output is a class label representing the current voltage state.

[0068] Model training: The training of this model is supervised learning, and the goal is to find a decision boundary that can distinguish different class samples with the maximum "margin". The training data comes from the historical operation data of the power grid, which contains a large number of voltage sequences labeled by automated rules. Each training sample (i.e. a feature vector of a voltage sequence) is assigned a clear class label, for example: label 0: stable normal; label 1: voltage low (trend downward); label 2: voltage high (trend upward); label 3: voltage oscillation. The algorithm is trained on a large amount of labeled data, and the goal is to find a hyperplane that can best separate these class data points in feature space. The trained model has the ability to classify new, unseen voltage data.

[0069] It is worth mentioning that the system defines a fixed mapping rule in advance, which maps each class label directly to one or a set of standardized instructions. The rule is as follows: when the SVM model outputs label 0, it means that the current voltage state is stable and normal, so the system will execute the "no operation" instruction to maintain the existing running state. When the output label is 1 (indicating voltage low) or label 2 (indicating voltage high), the system maps the same instruction, which is to execute SET_VOLTAGE (220.0). The purpose of this instruction is to correct the detected voltage deviation by actively adjusting the voltage to the standard 220.0V. When the output label is 3, it means that the system detects a voltage oscillation state. At this time, the rule will trigger the ACTIVATE_DAMPING_MODE () instruction to start a special damping control mode to suppress voltage fluctuations and restore system stability.

[0070] For example, suppose the voltage data sequence for a node is 220V, 219V, and 218V. The system extracts the feature vector: {Current voltage: 218V, Voltage rate of change: -1V}. This feature vector is fed into the trained SVM model. Because this feature (particularly the negative rate of change) is similar to a large number of "low voltage" samples in the training data, the model outputs the prediction: Label 1. Upon receiving Label 1, the system queries the instruction mapping rules and generates a clear voltage adjustment instruction: "Adjust the output voltage of this node to 220V to restore stability."

[0071] It's important to note that after generating voltage adjustment commands, the system uses node distribution rules to package these commands, along with other coordinated action instructions, into a final stable voltage distribution plan and distribute it to each distributed power generation node. These node distribution rules take into account factors such as node location and load capacity to ensure balanced task distribution and optimal resource utilization. The command set is transmitted via a reliable communication protocol, ensuring accurate reception and execution by each node, ultimately achieving stable voltage control across the entire remote region.

[0072] It's worth noting that the core logic of the node distribution rule is a quantifiable task allocation mechanism based on weighted scoring. It converts abstract "factors" into specific weights, thereby determining the extent of each node's responsibility for collaborative tasks. The algorithm first quantifies a macro-control objective (e.g., "restoring the regional voltage to 220V") into a "total control task magnitude," denoted as T_total. The system then calculates a weight factor, W_i, for each participating node i, which determines the proportion of the task it should undertake. Multiple performance metrics are obtained for each node, such as: load capacity (C_i): the node's maximum available power; geographic location (D_i): the node's distance from critical load centers; health status (H_i): the node's health score or availability (between 0 and 1); and communication quality (L_i): the node's communication latency. All of these performance metrics are then normalized to their maximum and minimum values. A pre-defined weighting formula is used to calculate a comprehensive score, Score_i, for each node. A simplified example formula is as follows: Score_i = (w1*C_i) + (w2 / D_i) + (w3*H_i) + (w4 / L_i). For example, if load capacity is prioritized, w1 will have the highest value. Each node's score is then normalized to its maximum and minimum values ​​to obtain the final weight factor W_i, ensuring that the sum of all node weights is 1. The calculated task load for node i is T_i = T_total*W_i.

[0073] It is worth mentioning that the weight coefficients (w1, w2, w3, w4) in the weighting formula are determined by simulation optimization on the historical control scene database. Specifically, the system will preset one or more quantitative control performance indicators, such as "total voltage recovery time", "network loss" or "device call cost". Then, in a search space containing multiple candidate weight combinations (i.e. "grid search"), the system will simulate the task allocation and control process for each scene in the historical database with each weight combination, and calculate its final performance indicator score. Finally, the weight combination that optimizes the performance indicator (e.g. shortest recovery time) on the historical average performance is selected as the fixed preset weight. This data-driven approach ensures that the task allocation rule is statistically optimal, maximizing the overall operational efficiency of the system.

[0074] For example, in a system that needs to adjust the voltage, the total control task is T_total. There are 4 nodes in the system. According to the node allocation rule, node 1 and node 2 have a weight factor of W_1=0.3, W_2=0.3 because they are close in distance and have strong load capacity. Node 3 and node 4 have a weight factor of W_3=0.2, W_4=0.2 because they are far apart and have slightly weaker capacity. Finally, the instruction set containing these specific information, such as "node 1: voltage 220V, execution time...", is packaged and sent to each node through the CoAP protocol.

[0075] In step S16, the instantaneous fluctuation voltage of each device in the stable voltage distribution scheme is collected in real time, and if the instantaneous fluctuation voltage exceeds the preset fluctuation voltage threshold, the parameter configuration is iteratively optimized to obtain an enhanced instantaneous fine-tuning precision scheme, including: The instantaneous fluctuation voltage of each device in the stable voltage distribution scheme is collected in real time, and if the instantaneous fluctuation voltage exceeds the preset fluctuation voltage threshold, an abnormality is marked to obtain a voltage abnormality signal set; The voltage abnormality signal set is analyzed in a time window to obtain abnormal duration state data; According to the abnormal duration state data, the parameter adjustment is performed to obtain optimized parameter configuration data; According to the optimized parameter configuration data, real-time adjustment instructions are distributed to the distributed power supply nodes to obtain an enhanced instantaneous fine-tuning precision.

[0076] It is worth mentioning that after the initial stable voltage distribution scheme is implemented, the system will enter a continuous optimization feedback loop phase to ensure long-term stability of the voltage. The key to this phase is to collect the instantaneous fluctuation voltage data of each device in the scheme in real time. Once the fluctuation voltage amplitude is detected to exceed the fluctuation voltage threshold, it will be marked as abnormal and a voltage abnormality signal will be generated to form a voltage abnormality signal set.

[0077] It is worth noting that the fluctuation voltage threshold is determined by statistical analysis of system data under a large number of "ideal stable" operating conditions. The method is as follows: first, collect the instantaneous fluctuation voltage data of the system under ideal conditions such as no disturbance and stable load for a long time, form a benchmark database representing "healthy state"; then, analyze the statistical distribution characteristics of the data in the database, and select a higher quantile (such as 99% or 99.5% quantile) as the fluctuation voltage threshold. Setting the threshold value to 0.7V means that under ideal conditions, only less than 1% of the voltage fluctuations will exceed this amplitude. Therefore, any fluctuation beyond this threshold is a good reason to be considered as an abnormal signal worthy of attention, which needs to trigger further analysis or adjustment.

[0078] For example, in a remote area distributed power supply system containing 5 nodes, the fluctuation amplitude threshold is 0.7V. The system collects the instantaneous voltage fluctuations of nodes 1 to 5 as 0.5V, 0.8V, 1.2V, 0.3V and 0.9V respectively. Since the fluctuation amplitudes of nodes 2, 3 and 5 (0.8V, 1.2V and 0.9V) all exceed the threshold of 0.7V, the system will generate voltage abnormal signals for these three nodes, indicating that there may be potential voltage instability at these locations.

[0079] It should be noted that, in order to avoid overreaction to transient, non-continuous interference, the system does not immediately process individual voltage anomaly signals. Instead, a feedback loop mechanism is adopted, which transmits the voltage anomaly signal set to a signal processing algorithm that determines whether the anomaly is continuously triggered through time window analysis. The specific steps are as follows: First, a clear judgment rule needs to be set for the algorithm: time window (T_w): define a sliding time window size for analysis, for example, 5 seconds. Trigger threshold (N_t): define the number of anomaly events that need to be reached within the time window, for example, 3 times. The system creates an independent event buffer Buffer_i in memory for each monitored node i. This buffer is usually a queue (Queue) that stores the timestamps of recent abnormal events. The algorithm continuously monitors individual voltage anomaly signals, and when it receives an anomaly signal from node i, the algorithm immediately adds the event timestamp to the corresponding buffer Buffer_i. After adding a new event, the algorithm immediately checks all timestamps in Buffer_i, removing all "expired" timestamps older than (current time - time window T_w) from the buffer. This ensures that the buffer only contains data within the latest time window. After cleaning up expired events, the algorithm calculates the number of remaining events Count in Buffer_i. Compare the count value Count with the preset trigger threshold N_t, if Count≥N_t (for example, the number of events in the buffer is ≥3), the condition of continuous triggering is met. Once the condition is met, the algorithm determines that the anomaly of the node is in the "continuous triggering state", and generates the corresponding abnormal continuous state data. At the same time, in order to avoid repeated triggering within the same window, the buffer Buffer_i of the node is usually emptied at this time, and the next round of counting begins.

[0080] For example, assume the rules are: time window T_w=5 seconds, trigger threshold N_t=3 times. T=1.0s: node 3 has a first anomaly, the algorithm stores the timestamp 1.0s in Buffer_3. At this time, there is 1 event in the buffer. 1<3, not triggered. T=3.2s: node 3 has a second anomaly, 3.2s is stored. There are [1.0s, 3.2s] in the buffer, a total of 2 events. 2<3, not triggered. T=4.5s: node 3 has a third anomaly, 4.5s is stored. The algorithm cleans up expired events (no events are older than 4.5s-5s), and there are [1.0s, 3.2s, 4.5s] in the buffer, a total of 3 events. The algorithm immediately determines that the anomaly of node 3 is in the "continuous triggering state", generates the corresponding data, and clears Buffer_3, waiting for new anomaly events.

[0081] It is worth noting that when the system confirms the existence of an abnormal persistent state, it will start the iterative optimization process of parameters. This process can use the particle swarm optimization algorithm to iteratively adjust the key parameters in the consistency algorithm, such as coordination weight and convergence speed. The particle swarm optimization algorithm can efficiently search for the optimal parameter combination by simulating group intelligence, thereby generating an optimized parameter configuration data. This optimization ensures that the coordination between nodes is more efficient and the response to voltage fluctuations is more rapid.

[0082] It is worth noting that the particle swarm optimization algorithm needs a clear evaluation standard to judge the pros and cons of any set of parameters. This standard is the fitness function. The fitness function is a mathematical formula that receives a set of candidate parameters (i.e. the position of a "particle") as input and outputs a numerical value ("fitness value") to quantify the performance of this set of parameters in solving the current specific abnormal problem. The goal of the PSO algorithm is to find the parameter combination that can make the fitness function value optimal (maximize or minimize). In the voltage control scenario, a good parameter combination should achieve a "fast, accurate, and stable" adjustment effect. Therefore, the fitness function is usually designed as a minimization target, and the smaller the value, the better the parameter effect. A typical fitness function may consist of the following weighted parts: . (Bias error): After regulation, the root mean square error or integral error between the node voltage and the nominal value (such as 220V). This item is used to evaluate the "accuracy" of the regulation. (Rise time): The time taken from the start of regulation to the voltage entering and remaining within the allowed error band (e.g. 220V ± 1%). This item is used to evaluate the "speed" of the regulation. (Overshoot): In the regulation process, the maximum amplitude of the voltage exceeding its final stable value. This item is used to evaluate the "stability" of the regulation to avoid new oscillations. , , is a weight coefficient, which is set by system designers according to control priorities (e.g., whether to value speed more or stability more). The algorithm randomly generates a swarm of “particles”, each of which represents a set of candidate parameters (e.g., {coordination weight: 0.7, convergence speed: 0.08}). For each particle, the system applies the parameter set it represents to attempt to correct the detected persistent anomaly (in a fast simulation environment or a controlled real environment). Then, the fitness value of each particle is calculated according to the fitness function formula. The particles update their “flight” direction and speed according to their own historical optimal position and the historical optimal position of the entire swarm, to find the space where there may be better solutions. After multiple iterations (e.g., 10 times), the entire swarm gradually converges in the area with the lowest fitness value. The algorithm finally outputs the parameter combination represented by the global optimal particle.

[0083] For example, for the persistent anomaly detected on node 3, the particle swarm optimization algorithm is started. After 10 iterations of calculation, the algorithm may decide to adjust the coordination weight of the area consistency algorithm from 0.6 to 0.8, and shorten the convergence speed from 0.1 seconds to 0.05 seconds. This optimized parameter configuration data containing the new weight and convergence speed will make the next collaborative adjustment more rapid and powerful.

[0084] It should be noted that after the optimized parameter configuration data is generated, the system will distribute real-time adjustment instructions to each distributed power supply node according to the new configuration through the node distribution rule.

[0085] For example, according to the optimized parameter configuration data, the node distribution rule may allocate 70% of the adjustment tasks to nodes 1 and 2, which are close to the load center and have strong performance, and instruct them to fine-tune the output voltage to 220V; while the remaining nodes 3, 4 and 5 share the remaining 30% of the tasks and maintain their original output. The specific adjustment instructions will be transmitted through protocols such as MQTT, for example, “node 1: voltage 220V, execution time 2025-08-1414:00:00”, so as to achieve accurate and efficient voltage fine-tuning.

[0086] In summary, the present application constructs a full-closed-loop control process from “perception-consensus” to “hierarchical decision-making” to “collaborative execution and feedback optimization”, deeply integrates the hierarchical control architecture with distributed nodes, and innovatively introduces the instruction refinement based on local load analysis and the parameter iterative optimization mechanism based on real-time feedback, solving the technical problems of slow response of centralized control, difficulty of multi-node collaboration and control strategy unable to adapt to local real-time working conditions in the prior art, and significantly improving the control accuracy, response speed and system stability of the network-type power system in dynamic complex scenarios.

[0087] Reference Figure 2The second embodiment of the present application provides a switching control device of a networked power system, comprising: A consensus confirmation module is configured to acquire an abnormal signal set, broadcast the abnormal signal set to adjacent nodes, and obtain a consensus confirmation result. A strategy generation module is configured to perform priority regulation by an upper layer node according to the consensus confirmation result, and obtain a global control strategy. An instruction refining module is configured to, if the global control strategy indicates that mode switching is required, notify a lower layer node to receive the global control strategy, then perform load fluctuation analysis on the lower layer node, obtain fluctuation characteristic data, and if the fluctuation characteristic data exceeds a preset load fluctuation range threshold, adjust local resource allocation to obtain a refined output instruction set. A coordination and synchronization module is configured to coordinate and synchronize the refined output instruction set to obtain a coordinated action instruction set. A voltage distribution module is configured to distribute and real-time adjust an instruction set to nodes according to the coordinated action instruction set to obtain a stable voltage distribution scheme. An iterative optimization module is configured to collect instantaneous fluctuation voltages of each device in the stable voltage distribution scheme in real time, and if the instantaneous fluctuation voltage exceeds a preset fluctuation voltage threshold, iteratively optimize parameter configuration to obtain an enhanced instantaneous fine-tuning precision scheme.

[0088] It should be noted that the switching control device of the networked power system provided by the embodiment of the present application is used to execute all process steps of the switching control method of the networked power system of the above-mentioned embodiment, and the working principles and beneficial effects of the two are one-to-one corresponding, and thus will not be repeated.

[0089] The embodiment of the present application further provides an electronic device. The electronic device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a switching control program of a networked power system. The processor implements the steps in each of the switching control method embodiments of the networked power system described above when executing the computer program, such as the step S11 shown in the figure. Figure 1 Alternatively, the processor implements the functions of each module / unit in each of the above-mentioned device embodiments when executing the computer program, such as the consensus confirmation module.

[0090] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0091] The electronic device can be a desktop computer, a notebook computer, a palm computer, a smart tablet, and the like. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device, and do not constitute a limitation on the electronic device, and can include more or less components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, and the like.

[0092] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is the control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.

[0093] The memory can be used to store the computer programs and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function, and the like), and the like; the data storage area can store data created according to the use of the electronic device (such as audio data, a phone book, and the like), and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0094] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0095] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0096] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only for the specific embodiments of the present application and do not limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A switching control method of a meshed power system, characterized by, The method comprises the following steps: acquiring an abnormal signal set, performing adjacent node broadcasting on the abnormal signal set, and obtaining a consensus confirmation result; performing priority regulation on the consensus confirmation result by an upper layer node to obtain a global control strategy; if the global control strategy indicates that mode switching is required, informing a lower layer node to receive the global control strategy, then performing load fluctuation analysis on the lower layer node to obtain fluctuation characteristic data, if the fluctuation characteristic data exceeds a preset load fluctuation range threshold, adjusting local resource allocation to obtain a refined output instruction set; performing coordination and synchronization on the refined output instruction set to obtain a coordinated action instruction set; performing instruction set distribution and real-time adjustment on the nodes according to the coordinated action instruction set to obtain a stable voltage distribution scheme; real-time collection of instantaneous fluctuation voltages of each device in the stable voltage distribution scheme, if the instantaneous fluctuation voltage exceeds a preset fluctuation voltage threshold, iterative optimization of parameter configuration to obtain an enhanced instantaneous fine-tuning precision scheme.

2. The switching control method of a network configuration type power system according to claim 1, characterized by, The method comprises the following steps: acquiring an abnormal signal set, performing adjacent node broadcasting on the abnormal signal set, and obtaining a consensus confirmation result; acquiring an instantaneous voltage fluctuation data set of an edge node; performing noise suppression processing on the instantaneous voltage fluctuation data set to obtain a first voltage data set; if the voltage value in the first voltage data set exceeds a preset voltage threshold, performing abnormal signal marking to obtain a first abnormal signal marking set; performing fluctuation amplitude analysis on the instantaneous voltage fluctuation data set to obtain a fluctuation amplitude value; if the fluctuation amplitude value exceeds a preset fluctuation amplitude threshold, combining the first abnormal signal marking set to perform abnormal signal identification to obtain an abnormal signal set; 3. The switching control method of a network configuration type power system according to claim 1, characterized by, generating local state information according to the abnormal signal set, sending the local state information to adjacent nodes, acquiring instantaneous fluctuation voltage data fed back by the adjacent nodes, and integrating the instantaneous fluctuation voltage data to obtain a consensus confirmation result. The method comprises the following steps: performing data aggregation on the fluctuation amplitude data of each node in the consensus confirmation result to obtain a unified fluctuation amplitude evaluation value; if the unified fluctuation amplitude evaluation value exceeds a preset fluctuation evaluation threshold, performing hierarchical control according to the unified fluctuation amplitude evaluation value by an upper layer node to obtain a hierarchical control instruction; acquiring operation state data of a distributed power supply according to the hierarchical control instruction, classifying the operation state data, then performing priority allocation according to the classification result to obtain a power supply regulation priority; 4. The network configuration type power system switching control method according to claim 1, characterized by, broadcasting the power supply regulation priority to a distributed power supply node to obtain a global control strategy. The method comprises the following steps: If the global control strategy indicates that mode switching is required, the lower layer node performs rule analysis and integrity verification on the parsed data to obtain policy analysis data; Real-time collection of load fluctuation data of devices applying the policy analysis data, time series decomposition of the load fluctuation data, and obtaining fluctuation characteristic data; If the fluctuation characteristic data exceeds the preset load fluctuation range threshold, the operating parameters of the distributed power supply are instantaneously fine-tuned to obtain a fine-tuned execution sequence; The fine-tuned execution sequence is refined to obtain a refined output instruction set.

5. The network configuration type power system switching control method according to claim 1, characterized by, The refined output instruction set is coordinated and synchronized to obtain a coordinated action instruction set, including: Obtaining instantaneous coordination sequence data from the power supply node executing the refined output instruction set, performing rule analysis on the data, and obtaining key coordination characteristic data; Performing multi-node synchronization operation on the key coordination characteristic data and detecting node communication delay, if the node communication delay exceeds the preset communication delay range, performing synchronization frequency adjustment to obtain sequence consistency state; According to the sequence consistency state, abnormal detection is performed to obtain an abnormal response frequency, if the abnormal response frequency exceeds the preset abnormal response frequency threshold, a trigger signal is generated; The trigger signal is processed by power coordination to obtain a coordinated action instruction set.

6. The network configuration type power system switching control method according to claim 1, characterized by, According to the coordinated action instruction set, the node is distributed with an instruction set and real-time adjusted to obtain a stable voltage distribution scheme, including: Distribute the coordinated action instruction set to each related distributed power supply node, and real-time collect the communication delay data of the power supply node; if the communication delay data is lower than the preset communication delay range, a mode switching signal is generated; According to the mode switching signal, real-time response mode switching is performed to obtain real-time response configuration data; The real-time response configuration data is processed by remote area voltage adjustment to generate voltage adjustment instructions; According to the voltage adjustment instructions, the instructions are distributed to obtain a stable voltage distribution scheme.

7. The network configuration type power system switching control method according to claim 1, characterized by, Real-time collection of instantaneous fluctuation voltage of each device in the stable voltage distribution scheme, if the instantaneous fluctuation voltage exceeds the preset fluctuation voltage threshold, iterative optimization parameter configuration is performed to obtain an enhanced instantaneous fine-tuning precision scheme, including: Real-time collection of instantaneous fluctuation voltage of each device in the stable voltage distribution scheme, if the instantaneous fluctuation voltage exceeds the preset fluctuation voltage threshold, abnormal marking is performed to obtain a voltage abnormal signal set; Time window analysis is performed on the voltage abnormal signal set to obtain abnormal duration state data; According to the abnormal duration state data, parameter adjustment is performed to obtain optimized parameter configuration data; According to the optimized parameter configuration data, real-time adjustment instructions are distributed to the distributed power supply node to obtain enhanced instantaneous fine-tuning precision.

8. A switching control device for a grid-type power system, characterized in that: Including: A consensus confirmation module for obtaining an abnormal signal set, performing adjacent node broadcasting on the abnormal signal set, and obtaining a consensus confirmation result; A strategy generation module for obtaining a global control strategy from an upper layer node by priority control according to the consensus confirmation result; The instruction refining module is configured to, if the global control strategy indicates that mode switching is required, notify a lower node to receive the global control strategy, then perform load fluctuation analysis on the lower node to obtain fluctuation characteristic data, and if the fluctuation characteristic data exceeds a preset load fluctuation range threshold, adjust local resource allocation to obtain a refined output instruction set; The coordination and synchronization module is configured to coordinate and synchronize the refined output instruction set to obtain a coordinated action instruction set; The voltage distribution module is configured to distribute the coordinated action instruction set to the nodes and adjust the nodes in real time according to the coordinated action instruction set to obtain a stable voltage distribution scheme; The iterative optimization module is configured to collect instantaneous fluctuation voltages of each device in the stable voltage distribution scheme in real time, and if the instantaneous fluctuation voltages exceed a preset fluctuation voltage threshold, iteratively optimize parameter configuration to obtain an enhanced instantaneous fine-tuning precision scheme.

Images