A network-constructing microgrid control method, device, equipment, medium and product

Abstract

The embodiment of the application provides a network-constructing micro-grid control method, device, equipment, medium and product, relates to the field of power system control, and the method preacquires source-load state data and scene data of the network-constructing micro-grid, uses a preset source-load prediction model to pre-judge source-load changes and fault levels in a future period, and when a trigger condition is met, not only triggers control according to a prediction result, but also calculates virtual inertia compensation amounts required to be provided by each distributed power supply according to the source-load prediction value and the fault level, so that accurate configuration of inertia support capabilities of each power supply is completed before source-load fluctuation or abnormality occurs, the abstract fault level is converted into specific virtual inertia compensation tasks, the system has inertia support capabilities matched with disturbance intensity and dynamic characteristics when an abnormality occurs, the system transient instability problem caused by inertia response lag or uneven distribution in the traditional method is solved, and the control efficiency of the network-constructing micro-grid is improved.

Description

Technical Field

[0001] This application relates to the field of power system control, and in particular to a control method, device, equipment, medium and product for a grid-type microgrid. Background Technology

[0002] Grid-based microgrid control refers to the use of devices such as energy storage converters and photovoltaic inverters as core components. By simulating the inertia and damping characteristics of synchronous generators, it establishes and maintains voltage and frequency stability in the microgrid, thereby achieving dynamic regulation of distributed power sources, energy storage systems, and diversified loads under multiple operating modes, including grid-connected, off-grid, and black-start. Grid-based microgrid control technology undertakes multiple tasks, including ensuring system transient stability, achieving real-time source-load balance, and responding to extreme operating condition disturbances. Its control performance directly affects power supply security, energy utilization efficiency, and user energy experience.

[0003] In existing technologies, the common control method for grid-type microgrids is centralized collaborative control. The central controller collects the operating data of various distributed power sources, energy storage and loads in the microgrid, generates coordinated control commands based on the global operating status, and sends them to each terminal execution unit to achieve power balance and stable operation within the system.

[0004] However, existing technologies suffer from low control efficiency in grid-type microgrids. Current centralized collaborative control methods heavily rely on real-time communication between the central controller and terminal units. Under conditions of rapid fluctuations in source load power and frequent changes in operating scenarios, they struggle to achieve flexible and timely dynamic responses and adaptive adjustments, resulting in low overall control efficiency for grid-type microgrids and difficulty in effectively guaranteeing control accuracy and operational stability. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, medium, and product for controlling a grid-type microgrid, in order to solve the problem of low control efficiency in the prior art for grid-type microgrids.

[0006] In a first aspect, embodiments of this application provide a grid-type microgrid control method, including:

[0007] Acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time;

[0008] The multiple source load status data and the multiple scenario data are input into a preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the output power of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time;

[0009] Triggering conditions are determined based on the predicted source load and the fault level to obtain a judgment result; wherein, the judgment result includes meeting the triggering conditions, which indicates that the preset grid-type microgrid has any one of the preset abnormal conditions within the preset second time period. The triggering condition judgment is based on a preset power threshold optimized according to the load fluctuation frequency and photovoltaic volatility and the predicted source load, and a preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level.

[0010] When the judgment result is that the triggering condition is met, the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control is calculated based on the source load prediction value and the fault level. Based on the virtual inertia compensation amount, a target control strategy is determined from a plurality of preset control strategies and the target control strategy is executed to eliminate the abnormal situation that occurs in the preset grid-type microgrid during the preset second time period.

[0011] In one possible design, the preset source load prediction model includes multiple prediction models. The step of inputting the multiple source load state data and the multiple scenario data into the preset source load prediction model to obtain the source load prediction value and fault level includes:

[0012] The multiple scene data are input into a preset scene twin model to obtain a virtual running trajectory adapted to the multiple scene data; wherein, the virtual running trajectory refers to the multiple scene branches that the current scene evolves into within the preset second time period;

[0013] Calculate the prediction response of each prediction model to each scenario branch, and normalize and weight the prediction response of each prediction model to each scenario branch to obtain the robustness score of each prediction model under the multiple scenario branches.

[0014] Calculate the matching degree between the multiple scene data and the preset scene feature templates corresponding to each prediction model, and perform multi-objective weighted fusion of the robustness score of each prediction model under the multiple scene branches and the matching degree to obtain the comprehensive fit index of each prediction model.

[0015] The multiple source load state data and the multiple scenario data are input into the target prediction model to obtain the source load prediction value and the fault level; wherein, the target prediction model refers to the model corresponding to the maximum value of the comprehensive adaptation index among the multiple prediction models.

[0016] In one possible design, after inputting the multiple source load state data and the multiple scenario data into a preset source load prediction model to obtain the source load prediction value and fault level, the method further includes:

[0017] Acquire multiple real-time data; wherein, the multiple real-time data are used to represent the operating status of the preset grid-type microgrid at the current moment;

[0018] The multiple real-time data, the predicted source load value, and the fault level are input into a preset collaborative decision-making model to obtain the multiple control strategies.

[0019] In one possible design, after executing the target control strategy to eliminate abnormalities occurring in the preset grid-type microgrid during the preset second time period, the method further includes:

[0020] Acquire multiple effect data; wherein, the multiple effect data refers to the actual operating status data of the preset grid-type microgrid after the target control strategy is executed;

[0021] Using the multiple effect data as training samples, the parameters of the preset source load prediction model and the preset collaborative decision-making model are updated.

[0022] In one possible design, the judgment result also includes not meeting the triggering condition. The judgment based on the source load prediction value and the fault level to obtain the judgment result includes:

[0023] If the predicted source load value is greater than the preset power threshold, or the fault level is higher than the preset level threshold, the judgment result is determined as the trigger condition being met.

[0024] If the predicted source load value is less than or equal to the preset power threshold, and the fault level is equal to or lower than the preset level threshold, the judgment result is determined as the triggering condition not being met.

[0025] In one possible design, before determining that the triggering condition is met if the predicted source load value is greater than a preset power threshold or the fault level is higher than a preset level threshold, the method further includes:

[0026] The load fluctuation frequency and photovoltaic volatility of the preset grid-type microgrid are obtained; wherein, the load fluctuation frequency is used to represent the severity of the disturbance to the load, the photovoltaic volatility refers to the change range of photovoltaic power generation within a preset unit time, and the photovoltaic power generation refers to the real-time power generation of the photovoltaic power station in the preset grid-type microgrid;

[0027] The preset initial power threshold is optimized based on the load fluctuation frequency and the photovoltaic volatility to obtain the preset power threshold;

[0028] The historical number of abnormal triggers and the current operating mode of the preset grid-type microgrid are obtained; wherein, the historical number of abnormal triggers refers to the number of times the preset grid-type microgrid has an abnormal situation within the preset first time period, and the current operating mode includes grid-connected mode and off-grid mode;

[0029] The preset initial level threshold is optimized based on the historical number of abnormal triggers and the current operating mode to obtain the preset level threshold.

[0030] In one possible design, the multiple source-load status data includes power supply operation data, load operation data, and system operation data. The acquisition of the multiple source-load status data and multiple scenario data includes:

[0031] The power operation data is collected by a preset distributed power monitoring unit; wherein, the power operation data includes the photovoltaic power generation, energy storage SOC data and charge / discharge power data, the energy storage SOC data is used to represent the remaining power of the energy storage battery in the preset grid-type microgrid, and the charge / discharge power data is used to represent the charging power and discharging power of the energy storage system in the preset grid-type microgrid;

[0032] The load operation data is collected by a preset load monitoring unit; wherein the load operation data is used to represent the operating status of the load in the preset grid-type microgrid;

[0033] The system operation data is collected by a preset system status monitoring unit; wherein, the system operation data includes voltage data and frequency data, the voltage data refers to the voltage amplitude at the grid connection point in the preset grid-type microgrid, and the frequency data refers to the AC frequency at the grid connection point;

[0034] The multiple scene data are obtained through a preset external interface.

[0035] In one possible design, the calculation of the virtual inertia compensation required by each distributed power source participating in the grid-connected microgrid control based on the predicted source load value and the fault level includes:

[0036] Based on the multiple source load state data and the multiple scenario data, a signal model for a multi-virtual synchronous machine parallel system is established; wherein, the multi-virtual synchronous machine parallel system refers to a parallel operation system composed of multiple distributed power sources using virtual synchronous machine control mode in the preset grid-type microgrid;

[0037] The total inertia compensation requirement for the preset grid-type microgrid is determined based on the fault level.

[0038] Based on the signal model and the total inertia compensation requirement, distributed iterative calculations are performed through a preset neighbor communication mechanism to obtain the virtual inertia compensation amount required by each distributed power source.

[0039] In one possible design, determining the total inertia compensation requirement of the preset grid-type microgrid based on the fault level includes:

[0040] The abnormal situation corresponding to the fault level is analyzed to obtain the pre-simulation duration and energy deviation accumulation rate of the abnormal situation corresponding to the fault level; wherein, the pre-simulation duration is used to represent the time span from the occurrence to the peak of the abnormal situation corresponding to the fault level, and the energy deviation accumulation rate refers to the integral value of the power deficit or power surplus of the preset grid-type microgrid in a unit time.

[0041] Based on the rehearsal duration, the energy deviation accumulation rate, and the preset system inertia time constant of the preset grid-type microgrid, calculate the inertia compensation kinetic energy used to offset the impact of the abnormal situation corresponding to the fault level;

[0042] By coupling multiple inertia compensation kinetic energies corresponding to the abnormal conditions of the fault level, the total inertia compensation requirement of the preset grid-type microgrid is obtained.

[0043] In one possible design, based on the signal model and the total inertia compensation requirement, distributed iterative calculations are performed through a preset neighbor communication mechanism to obtain the virtual inertia compensation amount required by each distributed power source, including:

[0044] Based on the signal model, the electrical distance matrix and participation factor vector of each distributed power source under the system oscillation mode are extracted; wherein, the system oscillation mode refers to the inherent motion pattern of relative swaying between the distributed power sources in a multi-virtual synchronous machine parallel system after being disturbed; the electrical distance matrix is ​​used to represent the coupling strength between the distributed power sources in the dynamic response; and the participation factor vector is used to represent the degree of contribution of each distributed power source to the system oscillation mode.

[0045] Based on the total inertia compensation requirement, the electrical distance matrix, and the participation factor vector, a distributed optimization objective function is constructed with the goal of minimizing the non-uniformity of the spatial distribution of system inertia.

[0046] The current iteration value of each distributed power source is initialized to a preset value; where the current iteration value of each distributed power source refers to the current virtual inertia compensation value of each distributed power source.

[0047] The current iteration value is iteratively updated among the distributed power sources through the preset neighbor communication mechanism, and the current iteration value after each iteration is corrected based on the distributed optimization objective function;

[0048] When the difference between the current iteration values ​​of two adjacent iterations is less than the preset convergence threshold, it is determined that the iteration has converged, and the current iteration value of each distributed power source at the time of iteration convergence is determined as the amount of virtual inertia compensation required by each distributed power source.

[0049] Secondly, embodiments of this application provide a grid-type microgrid control device, comprising:

[0050] The first acquisition module is used to acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time;

[0051] The first input module is used to input the multiple source load status data and the multiple scenario data into a preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the output power of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time;

[0052] The judgment module is used to judge the triggering conditions based on the predicted source load value and the fault level, and obtain the judgment result; wherein, the judgment result includes meeting the triggering conditions, which is used to indicate that the preset grid-type microgrid has any one of the preset abnormal conditions within the preset second time period. The triggering condition judgment is based on the preset power threshold optimized according to the load fluctuation frequency and photovoltaic fluctuation rate and the predicted source load value, and on the preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level.

[0053] The determination module is used to calculate the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control based on the source load prediction value and the fault level when the judgment result is that the trigger condition is met, and to determine the target control strategy from a plurality of preset control strategies based on the virtual inertia compensation amount, and to execute the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid during the preset second time period.

[0054] In one possible design, the preset source load prediction model includes multiple prediction models, and the first input module includes:

[0055] The first input unit is used to input the multiple scene data into a preset scene twin model to obtain a virtual running trajectory adapted to the multiple scene data; wherein, the virtual running trajectory refers to multiple scene branches that evolve in the current scene within the preset second time period;

[0056] The first calculation unit is used to calculate the prediction response of each prediction model to each scenario branch, and to normalize and weight the prediction response of each prediction model to each scenario branch to obtain the robustness score of each prediction model under the multiple scenario branches.

[0057] The fusion unit is used to calculate the matching degree between the multiple scene data and the preset scene feature templates corresponding to each prediction model, and to perform multi-objective weighted fusion of the robustness score of each prediction model under the multiple scene branches and the matching degree to obtain the comprehensive fit index of each prediction model.

[0058] The second input unit is used to input the multiple source load state data and the multiple scenario data into the target prediction model to obtain the source load prediction value and the fault level; wherein, the target prediction model refers to the model corresponding to the maximum value of the comprehensive adaptation index among the multiple prediction models.

[0059] In one possible design, the grid-type microgrid control device further includes:

[0060] The second acquisition module is used to acquire multiple real-time data; wherein, the multiple real-time data are used to represent the operating status of the preset grid-type microgrid at the current moment;

[0061] The second input module is used to input the multiple real-time data, the source load prediction value, and the fault level into a preset collaborative decision-making model to obtain the multiple control strategies.

[0062] In one possible design, the grid-type microgrid control device further includes:

[0063] The third acquisition module is used to acquire multiple effect data; wherein, the multiple effect data refers to the actual operating status data of the preset grid-type microgrid after the target control strategy is executed;

[0064] The update module is used to update the parameters of the preset source load prediction model and the preset collaborative decision-making model using the multiple effect data as training samples.

[0065] In one possible design, the judgment result also includes the condition that the triggering condition is not met. The judgment module includes:

[0066] The first determining unit is used to determine the judgment result as the fulfillment of the triggering condition if the predicted source load value is greater than a preset power threshold or the fault level is higher than a preset level threshold.

[0067] The second determining unit is used to determine the judgment result as the failure to meet the triggering condition if the predicted source load value is less than or equal to the preset power threshold and the fault level is equal to or lower than the preset level threshold.

[0068] In one possible design, the determination module further includes:

[0069] The second acquisition unit is used to acquire the load fluctuation frequency and photovoltaic volatility of the preset grid-type microgrid; wherein, the load fluctuation frequency is used to represent the severity of the disturbance to the load, the photovoltaic volatility refers to the change range of photovoltaic power generation within a preset unit time, and the photovoltaic power generation refers to the real-time power generation of the photovoltaic power station in the preset grid-type microgrid;

[0070] The first optimization unit is used to optimize the preset initial power threshold based on the load fluctuation frequency and the photovoltaic fluctuation rate to obtain the preset power threshold;

[0071] The third acquisition unit is used to acquire the historical number of abnormal triggers of the preset grid-type microgrid and the current operating mode of the preset grid-type microgrid; wherein, the historical number of abnormal triggers refers to the number of times the preset grid-type microgrid has an abnormal situation within the preset first time period, and the current operating mode includes grid-connected mode and off-grid mode;

[0072] The second optimization unit is used to optimize the preset initial level threshold based on the number of historical abnormal triggers and the current operating mode to obtain the preset level threshold.

[0073] In one possible design, the plurality of source-load status data includes power supply operation data, load operation data, and system operation data, and the first acquisition module includes:

[0074] The first acquisition unit is used to acquire the power operation data through a preset distributed power monitoring unit; wherein, the power operation data includes the photovoltaic power generation, energy storage SOC data and charge / discharge power data, the energy storage SOC data is used to represent the remaining power of the energy storage battery in the preset grid-type microgrid, and the charge / discharge power data is used to represent the charging power and discharging power of the energy storage system in the preset grid-type microgrid;

[0075] The second acquisition unit is used to acquire the load operation data through the preset load monitoring unit; wherein the load operation data is used to represent the operating status of the load in the preset grid-type microgrid;

[0076] The third acquisition unit is used to acquire the system operation data through the preset system status monitoring unit; wherein, the system operation data includes voltage data and frequency data, the voltage data refers to the voltage amplitude at the grid connection point in the preset grid-type microgrid, and the frequency data refers to the AC frequency at the grid connection point;

[0077] The fourth acquisition unit is used to acquire the multiple scene data through a preset external interface.

[0078] In one possible design, the determining module includes:

[0079] The establishment unit is used to establish a signal model of a multi-virtual synchronous machine parallel system based on the multiple source load state data and the multiple scenario data; wherein, the multi-virtual synchronous machine parallel system refers to a parallel operation system composed of multiple distributed power sources using virtual synchronous machine control mode in the preset grid-type microgrid;

[0080] The third determining unit is used to determine the total inertia compensation requirement of the preset grid-type microgrid based on the fault level.

[0081] The second calculation unit is used to perform distributed iterative calculations based on the signal model and the total inertia compensation requirement, through a preset neighbor communication mechanism, to obtain the virtual inertia compensation amount required by each distributed power source.

[0082] In one possible design, the third determining unit includes:

[0083] The analysis component is used to analyze the abnormal situation corresponding to the fault level to obtain the pre-simulation duration and energy deviation accumulation rate of the abnormal situation corresponding to the fault level; wherein, the pre-simulation duration is used to represent the time span from the occurrence to the peak of the abnormal situation corresponding to the fault level, and the energy deviation accumulation rate refers to the integral value of the power deficit or power surplus of the preset grid-type microgrid in a unit time.

[0084] The calculation component is used to calculate the inertial compensation kinetic energy to offset the impact of the abnormal situation corresponding to the fault level, based on the pre-simulation duration, the energy deviation accumulation rate, and the preset system inertia time constant of the preset grid-type microgrid.

[0085] A coupling component is used to couple multiple inertia compensation kinetic energies corresponding to the abnormal conditions corresponding to the fault level, so as to obtain the total inertia compensation requirement required by the preset grid-type microgrid.

[0086] In one possible design, the second computing unit includes:

[0087] An extraction component is used to extract the electrical distance matrix and participation factor vector of each distributed power source under the system oscillation mode according to the signal model; wherein, the system oscillation mode refers to the inherent motion pattern of relative swaying between distributed power sources in a multi-virtual synchronous machine parallel system after being disturbed; the electrical distance matrix is ​​used to represent the coupling strength between distributed power sources in dynamic response; and the participation factor vector is used to represent the degree of contribution of each distributed power source to the system oscillation mode.

[0088] A component is constructed to build a distributed optimization objective function with the goal of minimizing the spatial non-uniformity of the system inertia, based on the total inertia compensation requirement, the electrical distance matrix, and the participation factor vector.

[0089] An initialization component is used to initialize the current iteration value of each distributed power source to a preset value; where the current iteration value of each distributed power source refers to the current virtual inertia compensation value of each distributed power source.

[0090] An update component is used to iteratively update the current iteration value among the distributed power sources through the preset neighbor communication mechanism, and to correct the current iteration value after each iteration based on the distributed optimization objective function;

[0091] The determination component is used to determine that the iteration has converged when the difference between the current iteration values ​​of two adjacent iterations is less than a preset convergence threshold, and to determine the current iteration value of each distributed power source at the time of iteration convergence as the amount of virtual inertia compensation required by each distributed power source.

[0092] Thirdly, embodiments of this application provide an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0093] The memory stores computer-executed instructions;

[0094] When the processor executes the computer execution instructions stored in the memory, it is used to implement the grid-type microgrid control method as described in any of the first aspects.

[0095] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the grid-type microgrid control method as described in any of the first aspects.

[0096] Fifthly, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, is used to implement the grid-type microgrid control method as described in any of the first aspects.

[0097] This application provides a control method, device, equipment, medium, and product for a grid-type microgrid. By pre-acquiring source-load state data and scenario data of the grid-type microgrid, it uses a preset source-load prediction model to predict source-load changes and fault levels in future periods. When triggering conditions are met, it not only triggers control based on the prediction results but also calculates the virtual inertia compensation amount required by each distributed power source based on the predicted source-load value and fault level. It completes the precise configuration of the inertia support capability of each power source before source-load fluctuations or anomalies occur, transforming the traditional control mode that relies on post-event response into an active compensation mode based on inertia demand prediction. By transforming the abstract fault level into a specific virtual inertia compensation task, the system has the inertia support capability that matches the disturbance intensity and dynamic characteristics when anomalies occur. It achieves precise suppression of frequency and voltage fluctuations from the perspective of system transient energy balance, solving the problem of system transient instability caused by inertia response lag or uneven distribution in traditional methods, and improving the control efficiency of the grid-type microgrid. Attached Figure Description

[0098] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0099] Figure 1 This is a schematic diagram illustrating an application scenario of the grid-type microgrid control method provided in the embodiments of this application;

[0100] Figure 2 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 1 ;

[0101] Figure 3 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 2 ;

[0102] Figure 4 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 3 ;

[0103] Figure 5 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 4 ;

[0104] Figure 6 An overall architecture diagram of a grid-type microgrid source-load coordinated control system based on event triggering and edge AI provided in the embodiments of this application;

[0105] Figure 7 A flowchart illustrating the control method of the source-load coordinated control method and system for a grid-type microgrid based on event triggering and edge AI, provided in this application embodiment;

[0106] Figure 8 This is a schematic diagram of the structure of the grid-type microgrid control device provided in the embodiments of this application;

[0107] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application.

[0108] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0109] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0110] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0111] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply difference. It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner. In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more.

[0112] It should be noted that the phrase "at...time" in the embodiments of this application can refer to the instant at which a certain situation occurs, or to a period of time after the occurrence of a certain situation; the embodiments of this application do not specifically limit this. Furthermore, the grid-type microgrid control method, device, equipment, medium, and product provided in the embodiments of this application are merely examples; a grid-type microgrid control method, device, equipment, medium, and product may also include more or fewer elements.

[0113] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0114] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0115] To clearly understand the technical solution of this application, the existing technical solutions will first be described in detail. Grid-based microgrid control refers to the use of equipment such as energy storage converters and photovoltaic inverters as the core, simulating the inertia and damping characteristics of synchronous generators to establish and maintain voltage and frequency stability in the microgrid. This allows for dynamic regulation of distributed power sources, energy storage systems, and diverse loads under multiple operating modes, including grid-connected, off-grid, and black-start. Grid-based microgrid control technology undertakes multiple tasks, including ensuring system transient stability, achieving real-time source-load balance, and responding to extreme operating condition disturbances. Its control performance directly affects power supply security, energy utilization efficiency, and user energy experience.

[0116] In existing technologies, the common control method for grid-type microgrids is centralized collaborative control. A central controller collects operational data from various distributed power sources, energy storage, and loads within the microgrid, generates coordinated control commands based on the overall operating status, and distributes them to each terminal execution unit to achieve power balance and stable operation within the system. However, this centralized collaborative control method heavily relies on real-time communication between the central controller and the terminal units. Under conditions of rapid fluctuations in source and load power and frequent changes in operating scenarios, it struggles to achieve flexible and timely dynamic response and adaptive adjustment. This results in low overall control efficiency for grid-type microgrids, and the control accuracy and operational stability cannot be effectively guaranteed. Therefore, existing technologies suffer from low control efficiency in grid-type microgrids.

[0117] Therefore, addressing the issue of low control efficiency in existing grid-type microgrids, this research found that a proactive triggering control method based on prediction and forecasting can be adopted to solve this problem. This transforms traditional reactive response into proactive prediction and strategy pre-setting, eliminating reliance on real-time communication and improving dynamic response speed and control efficiency: ① By collecting historical operating data and environmental data of the microgrid, a predictive model can be built to predict future source load power, allowing for early perception of future power change trends. This transforms traditional passive response into proactive forecasting, providing sufficient time margin for subsequent control decisions, overcoming control lag caused by rapid source load fluctuations, and improving the system's dynamic response capability. ② Potential abnormal faults during future microgrid operation can be quantitatively evaluated, and the evaluation results can be used as the basis for control triggering. This eliminates the traditional reliance on real-time communication for fault detection and response, enabling automatic activation of corresponding adjustment processes based on different fault levels, giving the system stronger adaptability in the face of changing operating scenarios. ③ Multiple control strategies can be pre-configured for different operating scenarios and abnormal conditions. When a specific operating condition is predicted to occur in the future, the system does not need to wait for real-time instructions from the central controller. Instead, it directly matches and executes the most suitable control scheme from the preset strategy library based on the prediction results. This transforms centralized real-time decision-making into distributed preset execution, reducing dependence on communication links and improving the response speed and execution efficiency of control instructions.

[0118] Specifically, it can collect source and load operation data and scenario data of grid-type microgrids, use predictive models to predict the source and load status and abnormal faults in future periods, set reasonable control trigger conditions based on the prediction results, and actively execute the appropriate control mode from the pre-configured control strategy when the trigger conditions are met. This breaks the high dependence of traditional centralized control on real-time communication, avoids the response lag problem caused by source and load fluctuations and scenario changes, realizes the early control and adaptive adjustment of the system, and improves the control efficiency of grid-type microgrids.

[0119] This application discloses a control method, device, equipment, medium, and product for a grid-type microgrid. By pre-acquiring source-load state data and scenario data of the grid-type microgrid, it uses a preset source-load prediction model to predict source-load changes and fault levels in future periods. When triggering conditions are met, it not only triggers control based on the prediction results but also calculates the virtual inertia compensation amount required by each distributed power source based on the predicted source-load value and fault level. It completes the precise configuration of the inertia support capability of each power source before source-load fluctuations or anomalies occur, transforming the traditional control mode that relies on post-event response into an active compensation mode based on inertia demand prediction. By transforming the abstract fault level into a specific virtual inertia compensation task, the system has the inertia support capability that matches the disturbance intensity and dynamic characteristics when anomalies occur. It achieves precise suppression of frequency and voltage fluctuations from the perspective of system transient energy balance, solving the problem of system transient instability caused by inertia response lag or uneven distribution in traditional methods, and improving the control efficiency of the grid-type microgrid.

[0120] Based on the above-mentioned inventive discovery, the technical solution of this application is proposed.

[0121] The application scenarios of the grid-type microgrid control method provided in the embodiments of the present invention are described below. Figure 1 This is a schematic diagram illustrating an application scenario of the grid-type microgrid control method provided in the embodiments of this application. For example... Figure 1 As shown, the application scenario includes a mobile terminal 101 and a server 102. The mobile terminal 101 collects multiple source-load status data and multiple scenario data, and sends the multiple source-load status data and multiple scenario data to the server 102. The server 102 inputs the multiple source-load status data and multiple scenario data into a preset source-load prediction model to obtain the source-load prediction value and fault level. The server 102 judges the trigger conditions based on the source-load prediction value and fault level, and obtains the judgment result. When the judgment result is that the trigger conditions are met, the server 102 calculates the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control based on the source-load prediction value and fault level, and determines the target control strategy from multiple preset control strategies based on the virtual inertia compensation amount, and executes the target control strategy to eliminate the abnormal situation that occurs in the grid-type microgrid in the second time period.

[0122] The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0123] Figure 2 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 1 .like Figure 2 As shown, in this embodiment, the execution entity of this invention is a server. Therefore, the grid-type microgrid control method provided in this embodiment includes the following steps:

[0124] S201. Acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time.

[0125] Specifically, voltage and current sensors can be deployed at the power output and load access terminals of the grid-type microgrid, respectively. Temperature and humidity sensors, light sensors, and wind speed sensors can be deployed in the area where the microgrid is located. The output voltage and current of the power source, the input voltage and current of the load, and the temperature, humidity, light intensity, and wind speed data of the area can be collected in real time within a preset first time period through each sensor. The collected data are used as source-load status data and scenario data, respectively, and then summarized to obtain multiple source-load status data and multiple scenario data. This step is used to provide comprehensive input data for the subsequent source-load prediction model, ensuring that the data input to the model can accurately reflect the actual operating status and environment of the microgrid in the first time period, and providing data support for obtaining accurate source-load prediction values ​​and fault levels.

[0126] Among them, multiple source-load status data refer to the electrical parameter information generated by the actual operation of power sources and loads within the grid-connected microgrid during the first time period. Specifically, this includes the output voltage, output current, and output power of power sources such as distributed photovoltaics, wind turbines, and energy storage, as well as the access voltage, current consumption, and power consumption of various residential and industrial loads, such as the output voltage of photovoltaic modules, the output power of energy storage batteries, and the power consumption of workshop equipment at a certain moment. Multiple scenario data refer to the parameter information related to the external environment and operating conditions of the grid-connected microgrid, which is independent of the internal electrical operating state of the microgrid. Specifically, this includes the irradiance, ambient temperature, wind speed, humidity, and grid operating period in the area where the microgrid is located, such as the irradiance value, ambient temperature value, and natural wind speed at a certain moment.

[0127] S202. Input multiple source load status data and multiple scenario data into the preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the power output of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time.

[0128] Specifically, the acquired source-load status data and scenario data can be preprocessed first to remove abnormal data and perform standardization transformation. The processed data is then used as the overall input to a pre-set source-load prediction model. This model consists of three components: a data input layer, a feature extraction layer, and a dual-output layer. The data input layer receives the preprocessed source-load status data and scenario data and transmits them to the feature extraction layer. The feature extraction layer performs feature mining and fusion on the power supply operating parameters and load operating parameters in the source-load status data, as well as the environmental parameters in the scenario data, to extract core features that reflect the correlation between source-load operation and the law of environmental impact. These core features are then transmitted to the dual-output layer. The first output branch of the dual-output layer performs power trend analysis on the core features and outputs the source-load prediction values ​​of power output and load power demand in the second time period. The second output branch performs anomaly correlation analysis on the core features and outputs the fault levels of the probability of various abnormal situations occurring in the microgrid in the second time period. This step is used to transform the collected historical operation and environmental data into key information that can guide subsequent control operations, providing accurate source-load prediction values ​​and fault levels, and providing a direct basis for subsequent trigger condition judgment and target control strategy determination.

[0129] Multiple sets of source-load status data and scenario data generated during the historical operation of the grid-connected microgrid can be selected as sample data. These sample data are divided into input samples and control samples. Through repeated training and parameter correction of the input and control samples, a source-load prediction model with stable processing capabilities is formed. The source-load prediction value is a numerical result output by the source-load prediction model based on historical operating data and environmental data, representing the power output and load power demand in a future second time period. Specifically, it includes the predicted output power of distributed generation and the predicted power consumption of the load; for example, if the model outputs a predicted photovoltaic output power of 120kW and a predicted load power demand of 180kW in the next 10 minutes, these values ​​are the corresponding source-load prediction values. The fault level is a result output by the source-load prediction model based on the input data, representing the probability of various abnormal situations occurring in the grid-connected microgrid in a future second time period. It is distinguished by level values ​​or level identifiers, with different levels corresponding to different degrees of abnormality; for example, if the model outputs a fault level of 2, it indicates that the microgrid will experience power imbalance-type abnormalities in the future period, and this level value is the corresponding fault level.

[0130] The pre-set abnormal situations that may occur in the second time period of the grid-connected microgrid mainly refer to various system operation abnormalities caused by source-load power imbalance, grid operating parameters deviating from the rated range, and external power supply link interruption. These cover types such as source-load fluctuation related faults, electrical parameter exceeding the standard faults, and grid connection faults. For example, the source-load fluctuation fault level exceeds the standard, that is, the difference between the power output and the load power demand exceeds the reasonable range, resulting in a power supply and demand imbalance; the voltage deviation exceeds the standard, that is, the voltage amplitude at the microgrid connection point deviates from the rated voltage value and exceeds the allowable deviation range; the main grid loses power, that is, the connection link between the microgrid and the main grid is interrupted, and the main grid cannot supply power to the microgrid. It may also include abnormalities such as the frequency deviation of the AC frequency at the connection point deviating from the rated frequency.

[0131] S203. Determine the triggering conditions based on the source load prediction value and the fault level to obtain the judgment result; wherein, the judgment result includes meeting the triggering conditions. Meeting the triggering conditions is used to indicate that the preset grid-type microgrid has any one of the preset abnormal conditions in a preset second time period. The triggering condition judgment is based on the preset power threshold optimized according to the load fluctuation frequency and photovoltaic fluctuation rate and the source load prediction value, as well as the preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level.

[0132] Specifically, fixed trigger thresholds can be preset, including source-load power difference thresholds and fault level thresholds. First, the difference between the power output of the power source and the power demand of the load in the predicted source-load value is calculated. Then, this difference is compared with the source-load power difference threshold and the fault level is compared with the fault level threshold. If either of them reaches or exceeds the corresponding threshold, it is determined that the trigger condition is met. If neither of them reaches the corresponding threshold, it is determined that the trigger condition is not met. The two judgment results are used together as the final judgment result of this trigger condition judgment. This step is used to determine whether the microgrid will have abnormal situations in the second time period, to divide the scenarios that require the execution of control strategies and those that do not require the execution of control strategies, and to screen out the situations that need to be regulated, so as to provide a clear basis for determining and executing the target control strategy in the future.

[0133] S204. When the judgment result is that the triggering condition is met, calculate the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control based on the source load prediction value and fault level, and determine the target control strategy from multiple preset control strategies based on the virtual inertia compensation amount, and execute the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid within the preset second time period.

[0134] Specifically, a virtual inertia compensation calculation and strategy matching module based on edge computing nodes can be constructed. This module is deployed in the local edge server of the grid-type microgrid. When the trigger condition judgment result based on the source load prediction value and fault level is met, it immediately reads the photovoltaic output prediction value, load power prediction value, and fault level output of the source load prediction model for the next 3-5 minutes. Based on these data, it calculates the total virtual inertia compensation required by the system under the current fault level. Then, combined with the pre-established electrical distance matrix and participation factor vector of each distributed power source, it uses a distributed iterative algorithm to differentiate the total inertia compensation requirement to each distributed power source. The energy storage converter and photovoltaic inverter, each distributed power source matches the corresponding control parameter combination from the preset strategy library according to the allocated virtual inertia compensation amount, and immediately adjusts its own virtual synchronous machine control parameters. This enables the system to complete the precise configuration of inertia support capability before an anomaly occurs. This step is used to convert the abstract fault level into specific, executable virtual inertia compensation instructions through quantitative calculation when an anomaly is predicted to occur. This allows each distributed power source to actively provide inertia support that matches the disturbance intensity when an anomaly occurs, thereby maintaining the frequency and voltage stability of the system during the anomaly process and eliminating transient instability caused by insufficient inertia.

[0135] For example, in a microgrid in an industrial park, a control module deployed on a local edge server receives real-time forecasts of photovoltaic (PV) output, load power, and fault levels for the next three minutes from a source-load prediction model. When, based on these forecasts, it is determined that an abnormal condition of "PV output dropping by 30% and fault level being high" will occur within the next three minutes and the triggering conditions are met, the control module immediately initiates a virtual inertia compensation calculation process. First, based on the fault level and the magnitude of the PV output drop, it calculates the total inertia compensation requirement for the system. The kinetic energy compensation amount is then read from the distributed power sources, including 2 energy storage converters and 3 photovoltaic inverters, which were previously identified offline. The electrical distance matrix and participation factor vector under the system's dominant oscillation mode are then used to process this information through a distributed iterative algorithm. The total inertia compensation requirement is allocated differentially to the five power sources: among them, the No. 1 energy storage converter, which is closest to the oscillation center, is allocated... Distribution of No. 2 energy storage converter The allocation of the No. 3 photovoltaic inverter with the highest participation factor The remaining two photovoltaic inverters are each allocated Each distributed power source matches the corresponding virtual synchronous machine control parameter combination from the preset strategy library according to the allocated virtual inertia compensation amount, and completes the online update of the control parameters within 2 seconds before the actual occurrence of the anomaly. When the actual photovoltaic output drops sharply 3 minutes later, each distributed power source has the inertia support capability matching the allocated amount, and the No. 1 energy storage converter immediately releases its power. The kinetic energy is released by the No. 2 energy storage converter. Photovoltaic inverter No. 3 releases through DC-side capacitor. The other two photovoltaic inverters each released The combined inertia of the five power sources precisely offset the power deficit caused by the sudden drop in photovoltaic power, ensuring that the system frequency remained within the allowable range of 49.8Hz to 50.2Hz throughout the entire anomaly process, without any transient instability.

[0136] Among them, each distributed power source is the core power generation unit that constitutes the grid-type microgrid. They are connected to the common bus of the microgrid in parallel through power lines, and together they form the power supply side system of the microgrid. Physically, they are distributed in different locations within the coverage area of ​​the grid-type microgrid, such as photovoltaic power stations on factory rooftops, energy storage stations in the corners of industrial parks, and wind turbines next to workshops. These power sources communicate with the central control system of the microgrid through local controllers to work together to achieve voltage and frequency support and power balance for the entire microgrid.

[0137] Virtual inertia compensation refers to the virtual kinetic energy support value that each distributed power source is required to provide to simulate the rotor inertia effect of a synchronous generator when the system anticipates an impending anomaly. The virtual inertia compensation required by each distributed power source refers to the specific inertia support task that the system allocates to each power source differently after calculation based on the fault level and power source characteristics. It quantifies the amount of inertia that each power source needs to actively contribute to suppress the impending system oscillation. The compensation amount received by each power source may be different, depending on its participation in the system oscillation mode and its electrical coupling relationship with other power sources.

[0138] The grid-based microgrid control method can be applied to the field of microgrid operation and control technology. It is suitable for independent grid-based microgrid technology scenarios in remote areas. It can solve the technical problems in traditional centralized collaborative control, which rely heavily on real-time communication and suffer from lag in control response, untimely dynamic adjustment, and low control efficiency when power supply and load fluctuate rapidly and the operating environment is complex and changeable. As a result, it is difficult to guarantee the stability and control accuracy of microgrid operation.

[0139] This embodiment provides a control method for a grid-connected microgrid. By pre-acquiring source-load state data and scenario data of the grid-connected microgrid, a preset source-load prediction model is used to predict future source-load changes and fault levels. When triggering conditions are met, control is not only triggered based on the prediction results, but also the virtual inertia compensation required by each distributed power source is calculated based on the predicted source-load values ​​and fault levels. This allows for precise configuration of the inertia support capabilities of each power source before source-load fluctuations or anomalies occur. This transforms the traditional control mode that relies on post-event response into an active compensation mode based on inertia demand prediction. By converting abstract fault levels into specific virtual inertia compensation tasks, the system possesses inertia support capabilities that match the disturbance intensity and dynamic characteristics when anomalies occur. This achieves precise suppression of frequency and voltage fluctuations from the perspective of system transient energy balance, solving the problem of system transient instability caused by inertia response lag or uneven distribution in traditional methods, and improving the control efficiency of the grid-connected microgrid.

[0140] In one possible design, the preset source load prediction model includes multiple prediction models. S202: Input multiple source load state data and multiple scenario data into the preset source load prediction model to obtain the source load prediction value and fault level, including:

[0141] S2021. Input multiple scene data into a preset scene twin model to obtain a virtual running trajectory that is adapted to the multiple scene data; wherein, the virtual running trajectory refers to the multiple scene branches that evolve in the current scene within a preset second time period.

[0142] Specifically, a scene twin model based on a conditional generative adversarial network can be constructed. This model is deployed in the edge computing nodes of a microgrid. Its generator takes multiple scene data collected at present, including light intensity, ambient temperature, wind speed, and humidity, as conditional inputs, and introduces a random noise vector that conforms to a standard normal distribution. By learning the distribution of historical scene data, the generator maps the current scene data and random noise into multiple scene time-series trajectories that may evolve in a future preset second time period. The discriminator then distinguishes between the generated scene trajectory and the historically actual scene trajectory. Through adversarial training between the generator and the discriminator, the generator can eventually output a virtual running trajectory that is adapted to the characteristics of the current scene and conforms to the historical evolution pattern, including multiple scene branches such as light enhancement branch, light reduction branch, intermittent cloud branch, and continuous temperature rise branch. This step is used to expand the static scene data at a single moment into a multi-branch dynamic evolution trajectory for future time periods, so that the subsequent prediction model can be robustly evaluated under multiple possible future scenarios, avoiding model mismatch problems caused by sudden scene changes.

[0143] The pre-defined scenario twin model is a deep generative model built on a conditional generative adversarial network architecture. The model construction process can be as follows: First, collect historical scene data of the microgrid area, including long-term series of parameters such as light intensity, ambient temperature, wind speed, and humidity. Clean and normalize these data. Then, construct an adversarial network consisting of two core components: a generator and a discriminator. The generator takes multiple scene data collected at the current moment as conditional input and introduces a random noise vector that conforms to a standard normal distribution. By learning the distribution pattern of historical scene data, it maps the current scene data and random noise into multiple scene time-series trajectories that will evolve in a pre-defined second time period in the future. The discriminator simultaneously receives the virtual scene trajectory output by the generator and the historically actual scene trajectory, distinguishes between them, and calculates the discriminant loss. Through alternating adversarial training between the generator and the discriminator, the generator continuously optimizes its network parameters until it can output a virtual running trajectory that is consistent with the distribution of historical data and adapted to the characteristics of the current scene. The input of the pre-defined scenario twin model is multiple scene data at the current moment and a random noise vector. The output is multiple virtual running trajectory branches representing the future evolution direction of the scene.

[0144] The virtual running trajectory refers to multiple possible scene evolution paths generated by the scene twin model, starting from the current moment and extending to a preset second time period in the future. Each path is a time-series data sequence composed of scene parameters such as light intensity, ambient temperature, wind speed, and humidity, reflecting the possible changing trends of the scene in the future. For example, under the scene data of light intensity of 800W / m², temperature of 28°C, and wind speed of 3m / s at the current moment, a certain path in the virtual running trajectory may depict the complete process of light intensity first slowly rising to 900W / m² and then falling back within the next 15 minutes.

[0145] Multiple scenario branches refer to specific path categories with different evolution directions in the virtual operation trajectory. Each branch represents a scenario evolution mode with specific physical meaning. The branches are distinguished from each other in terms of change trend, change magnitude and change rate, and together they cover various situations that may occur in the future scenario. For example, in the photovoltaic microgrid scenario, multiple scenario branches may include: a light enhancement branch with continuously increasing light intensity, a light weakening branch with continuously decreasing light intensity, a cloudy intermittent branch with frequent fluctuations in light intensity, a temperature continuously rising branch with synchronous increase in temperature with sunshine, and a wind disturbance branch with sudden increase in wind speed, etc. These branches together constitute a characterization of the future evolution of the current scenario.

[0146] S2022. Calculate the prediction response of each prediction model to each scenario branch, and normalize and weight the prediction response of each prediction model to each scenario branch to obtain the robustness score of each prediction model under multiple scenario branches.

[0147] Specifically, for each scene branch in the generated virtual trajectory, Gaussian white noise matching the characteristics of that scene can be superimposed on the time-series data points of that branch to form a set of perturbed input samples. Then, this set of perturbed samples is input into each prediction model, and the change in the output value of each model before and after the perturbation is calculated. The average of the output change under all perturbed samples is taken as the prediction response of the model for that scene branch. The prediction response of each model under all scene branches is normalized and mapped to the 0-1 interval. Then, the normalized response is weighted and summed according to the occurrence probability weight of each scene branch to obtain the robustness score of each prediction model under multiple scene branches. This step is used to quantify the output stability of each prediction model when facing small perturbations in the input data. By expanding the accuracy index under a single scene to the anti-perturbation ability index under multiple scenes, a quantitative basis reflecting the dynamic stability of the model is provided for subsequent model selection.

[0148] Predictive responsiveness refers to the degree of change in the model's output value when a small perturbation is applied to the input data of the prediction model. Specifically, it is the magnitude of the difference between the output values ​​before and after the perturbation. The larger the difference, the more sensitive the model is to the perturbation and the higher the predictive responsiveness. The smaller the difference, the less sensitive the model is to the perturbation and the lower the predictive responsiveness. For example, for a photovoltaic power output prediction model, if a small perturbation of 1% is added to the input irradiance value, the photovoltaic power output predicted by the model fluctuates drastically from 100kW to 120kW. Then, the predictive responsiveness of the model in this scenario is 20%, indicating that the model is extremely sensitive to small changes in the input data.

[0149] Robustness score is a quantitative indicator obtained by comprehensively processing the prediction response of a prediction model under multiple scenario branches. It is used to measure the model's ability to maintain stable output when facing various possible future scenario disturbances. The higher the score, the more stable the model's output can be under various scenario evolutions. The lower the score, the more likely the model's output is to fluctuate drastically with changes in scenario. For example, a prediction model has a prediction response of 5% under the enhanced illumination branch, 6% under the reduced illumination branch, and 15% under the cloudy intermittent branch. After normalization and weighted summation, the robustness score of this model is 0.85. Another model has prediction responses of 12%, 14%, and 30% under the same scenario branches, respectively, and its robustness score is 0.62, indicating that the former can provide more stable prediction output under variable weather conditions.

[0150] S2023. Calculate the matching degree between multiple scene data and the preset scene feature templates corresponding to each prediction model, and perform multi-objective weighted fusion of the robustness score and matching degree of each prediction model under multiple scene branches to obtain the comprehensive fit index of each prediction model.

[0151] Specifically, the process begins by comparing the collected scene data with the preset scene feature templates corresponding to each prediction model. The Euclidean distance or cosine similarity between the two in the high-dimensional feature space is calculated, and this distance value is converted into a matching score in the range of 0-1. The higher the score, the better the current scene matches the feature template of the model. Then, the robustness scores of each prediction model under various scene branches are read. The matching score and robustness score are used as two independent optimization dimensions. The entropy weight method is used to dynamically calculate the matching weight and robustness weight based on the dispersion of each dimension under the current working condition. Then, the matching score and robustness score are weighted and summed based on these two weights to obtain the comprehensive fit index of each prediction model. This step is used to dynamically balance the current scene fit and the ability to resist disturbances in future scenes, merging the two competing indicators into a unified quantitative index. This ensures that the final selected target prediction model can accurately predict under the current working condition and maintain stable output in various possible future scene evolutions.

[0152] The comprehensive fit index is a quantitative indicator obtained by multi-objective weighted fusion of the matching score of the prediction model in the current scenario and its robustness score in various future scenario branches. It is used to measure the comprehensive performance of a prediction model in two dimensions: the fit of the current scenario and the ability to resist disturbances in the future scenario. The higher the index, the more accurate the prediction is under the current operating conditions and the more stable the output is in possible future scenario evolutions. For example, in a microgrid in a park, there are three candidate prediction models. Model A has a matching score of 0.95 but a robustness score of 0.60, while Model B has a matching score of 0.80 and a robustness score of 0.60. Model C has a matching score of 0.70 and a robustness score of 0.85. The matching weight is determined to be 0.4 and the robustness weight is 0.6 under the current working condition using the entropy weight method. After weighted fusion calculation, the comprehensive fit index of model A is 0.74, the comprehensive fit index of model B is 0.86, and the comprehensive fit index of model C is 0.79. Finally, model B with the highest comprehensive fit index is selected as the target prediction model. Although the prediction accuracy of model B under the current scenario is slightly lower than that of model A, it can maintain more stable output when facing future scenario changes such as cloudy weather and cooling.

[0153] S2024. Input multiple source load state data and multiple scenario data into the target prediction model to obtain the source load prediction value and fault level; whereby the target prediction model refers to the model corresponding to the maximum value of the comprehensive adaptation index among multiple prediction models.

[0154] Specifically, the target prediction model already trained and deployed in the edge computing nodes of the microgrid can be invoked. Multiple source-load status data and multiple scenario data collected in real time are concatenated and normalized according to the format required by the model input layer and then input into the model. The model internally performs forward propagation calculation on the input data through a multi-layer neural network structure with fixed weights. Its first output branch directly outputs the photovoltaic power output value sequence and load power value sequence at each time point in the future preset second time period as the source-load prediction value. The second output branch directly outputs the occurrence probability of each anomaly type in the future preset second time period and takes the anomaly level corresponding to the highest probability as the fault level. This step is used to input the current source-load operation data and environmental data into the optimal prediction model selected after multi-objective trade-offs. Through a single forward calculation, a quantitative prediction result for the future period is quickly obtained, providing input basis for subsequent trigger condition judgment and virtual inertia compensation calculation.

[0155] The technical effect of this scheme in this embodiment is as follows: by introducing a preset scene twin model to generate multiple scene branches of the current scene evolving in the future, and comprehensively calculating the prediction response of each prediction model under these scene branches to calculate its robustness score, and then performing multi-objective weighted fusion with the static matching degree of the scene feature template, the model with the highest comprehensive fit index is finally selected as the target prediction model. This realizes the transformation from static scene matching at a single moment to dynamic robustness evaluation for future evolution, so that the selected model not only has high accuracy in the current scene, but also maintains stable output under various possible scene disturbances in the future. This avoids frequent model switching and control command oscillation caused by small fluctuations or rapid changes in scene data, and solves the problem of discontinuous control and poor system stability caused by ignoring the dynamic response characteristics of the model in traditional model selection methods.

[0156] In one possible design, the judgment result also includes: triggering conditions are not met (S203). The judgment result is obtained by judging the triggering conditions based on the source load prediction value and the fault level, including:

[0157] S2031. If the predicted source load value is greater than the preset power threshold, or the fault level is higher than the preset level threshold, the judgment result will be determined as meeting the triggering condition.

[0158] Specifically, the generated source load prediction value and fault level can be read first, and then the preset power threshold and level threshold can be called respectively. First, the source load prediction value and power threshold are compared, and then the fault level and level threshold are compared. When either the source load prediction value is greater than the power threshold or the fault level is higher than the level threshold, the judgment result of the trigger condition is directly marked as satisfied. This step is used to complete the judgment of the trigger condition based on the prediction result, and provide a clear judgment conclusion for whether to execute the control strategy in the future.

[0159] S2032. If the predicted source load value is less than or equal to the preset power threshold, and the fault level is equal to or lower than the preset level threshold, the judgment result will be determined as not meeting the triggering conditions.

[0160] Specifically, the generated source load prediction value and fault level can be retrieved first, and then the pre-set power threshold and level threshold can be retrieved. The source load prediction value and power threshold are compared numerically, and the fault level is compared level-wise. When both conditions are met, the source load prediction value is less than or equal to the power threshold and the fault level is equal to or lower than the level threshold, the trigger condition is determined to be unmet. This step is used to determine the trigger condition based on the prediction result, providing a clear judgment conclusion for whether to execute the control strategy.

[0161] The technical effect of this solution in this embodiment is as follows: by clarifying the judgment criteria for triggering conditions, and using the comparison between the predicted source load value and the preset power threshold, and the fault level and the preset level threshold as the judgment basis, the precise determination of triggering conditions is achieved. This enables accurate identification of possible abnormal situations in the system in the future, avoids false triggering or missed triggering, provides clear guidance for the timely activation of subsequent control strategies or the stable maintenance of normal operating conditions, ensures the timeliness and accuracy of system response, and solves the problem of ambiguous judgment of triggering conditions and lack of clear standards.

[0162] In one possible design, multiple source-load status data include power supply operation data, load operation data, and system operation data. S201: Acquire multiple source-load status data and multiple scenario data, including:

[0163] S2011. Collect power operation data through a preset distributed power monitoring unit; wherein, the power operation data includes photovoltaic power generation, energy storage SOC data and charge / discharge power data, the energy storage SOC data is used to represent the remaining power of the energy storage battery in the preset grid-type microgrid, and the charge / discharge power data is used to represent the charging power and discharging power of the energy storage system in the preset grid-type microgrid.

[0164] Specifically, a distributed power source monitoring unit can be integrated inside the inverter of a photovoltaic power station. This monitoring unit directly collects the voltage and current signals input to the DC side of the inverter via hard-wiring, calculates the photovoltaic power generation through internal processing, and simultaneously connects to the energy storage battery management system via the communication bus within the photovoltaic power station. It reads the remaining charge percentage output by the battery management system as the energy storage state of charge (SOC) data and collects values ​​from the operating status register of the energy storage converter to obtain the current charging and discharging power data. This step is used to obtain real-time operating data for the two core distributed power sources—photovoltaics and energy storage—in a grid-connected microgrid. Photovoltaic power generation reflects the instantaneous magnitude of photovoltaic output, energy storage SOC data represents the current energy reserve level of the energy storage system, and charging and discharging power data reflects the ongoing charging and discharging behavior of the energy storage system. This data constitutes the necessary basic information on the power source side for subsequent source-load prediction and trigger condition determination.

[0165] S2012. Load operation data is collected through a preset load monitoring unit; wherein, the load operation data is used to represent the operating status of the load in the preset grid-type microgrid.

[0166] Specifically, smart meters can be installed at the incoming line of the distribution box of each critical load branch as load monitoring units. These smart meters collect the three-phase current and voltage waveforms of the branch in real time through current and voltage transformers. The internal metering chip calculates load operation data such as active power, reactive power, and power factor, and uploads the data to a data aggregation terminal via an RS485 bus or wireless communication module. This step is used to obtain the actual power consumption of various loads in the grid-type microgrid. The collected load operation data reflects the power demand and operating characteristics of the load at different times. This data serves as input to the source-load prediction model, providing a basis for predicting the load power demand in future periods.

[0167] S2013. Collect system operation data through the preset system status monitoring unit; wherein, the system operation data includes voltage data and frequency data, the voltage data refers to the voltage amplitude at the grid connection point in the preset grid-type microgrid, and the frequency data refers to the AC frequency at the grid connection point.

[0168] Specifically, a system status monitoring unit can be installed at the grid connection point of a grid-connected microgrid. This unit integrates a voltage transformer and a frequency transmitter. The primary side of the voltage transformer is directly connected in parallel to the three-phase bus at the grid connection point to collect voltage amplitude signals. The frequency transmitter extracts the AC frequency signal from the voltage waveform using phase-locked loop (PLL) technology. The system status monitoring unit converts the collected voltage amplitude and AC frequency signals into digital values ​​and transmits them to the microgrid's local controller via fiber optic or Ethernet interfaces. This step is used to obtain the voltage level and frequency status at the grid connection point of the grid-connected microgrid. The collected voltage data represents the voltage amplitude of the grid connection point bus, and the frequency data represents the current AC frequency of the system. These two data points are core indicators reflecting the overall operating status and power quality of the microgrid, providing a basis for subsequent judgments on whether the system is within the normal operating range and for trigger condition analysis.

[0169] S2014. Obtain multiple scene data through preset external interfaces.

[0170] Specifically, an external interface with network communication capabilities can be set up on the local controller of the microgrid. This interface establishes a TCP / IP connection with the meteorological service center's server via Ethernet cable or fiber optic cable. It periodically sends data request commands to the meteorological service center according to an agreed-upon data format, receives forecast data containing meteorological elements such as temperature, light intensity, and wind speed for future periods, and parses and stores this data in the local controller's real-time database. This step is used to acquire scenario data related to the microgrid's operating environment from external systems. The collected meteorological forecast data, such as light intensity, temperature, and wind speed, reflects the external environmental conditions of the microgrid in future periods. This data serves as input to the source-load prediction model, providing environmental background information for the model to predict changes in photovoltaic output and load demand trends.

[0171] The technical effect of this solution in this embodiment is as follows: by using a distributed power supply monitoring unit, a load monitoring unit, and a system status monitoring unit to collect power supply operation data, load operation data, and system operation data respectively, and by obtaining scenario data through an external interface, it is possible to achieve multi-dimensional collection of source and load status and operation scenarios, ensuring the integrity and accuracy of source and load status data and scenario data, providing a reliable data foundation for subsequent prediction, judgment, and strategy execution, improving the overall control accuracy and response reliability of the system, and solving the problems of single data collection dimension and insufficient reliability in traditional data collection.

[0172] In one possible design, S204 calculates the virtual inertia compensation required by each distributed power source participating in the grid-connected microgrid control based on the predicted source load and fault level, including:

[0173] S2041. Based on multiple source-load state data and multiple scenario data, establish a signal model for a multi-virtual synchronous machine parallel system; wherein, the multi-virtual synchronous machine parallel system refers to a parallel operation system composed of multiple distributed power sources using virtual synchronous machine control mode in a pre-defined grid-type microgrid.

[0174] Specifically, a multi-virtual synchronous machine parallel system identification model based on deep neural networks can be constructed. This model consists of four components: an input encoding layer, a time-series feature extraction layer, a dynamic coupling layer, and an output decoding layer. First, the collected source-load state data, including the output voltage, output current, power angle, and frequency of each distributed power source, as well as multiple scene data, including light intensity, ambient temperature, wind speed, and humidity, are concatenated and normalized to generate a multi-dimensional input feature vector. This feature vector is then input to the input encoding layer. The input encoding layer uses a fully connected network to perform dimensionality transformation and feature recombination on the multi-dimensional feature vector, outputting an encoded high-dimensional feature tensor. This high-dimensional feature tensor is input to the time-series feature extraction layer. The time-series feature extraction layer uses a long short-term memory network to perform time-series modeling on the high-dimensional feature tensor, extracting the dynamic characteristics of each power source's output changing over time. The first step involves constructing an end-to-end identification model that simultaneously characterizes the dynamic characteristics of each power source and the coupling relationships between them. This model provides an analytical neural network structure for extracting the electrical distance matrix and participation factor vectors from the model.

[0175] Among them, the multi-virtual synchronous machine parallel system refers to a power generation system consisting of multiple distributed power sources using virtual synchronous machine control mode connected through a common bus and operating together. These power sources are electrically connected in parallel to the microgrid, and in terms of control, they each simulate the operating characteristics of a synchronous generator, jointly undertaking the tasks of voltage establishment and frequency support for the microgrid. For example, an industrial park microgrid contains two energy storage converters and three photovoltaic inverters. All five devices adopt the virtual synchronous machine control mode and are connected in parallel to the park's 380V bus, jointly providing voltage and frequency support for the loads in the park. When one energy storage converter is taken out of operation for maintenance, the remaining four power sources can automatically redistribute power output to maintain stable system operation.

[0176] Virtual synchronous machine control mode is a control strategy for power electronic converters. By introducing an algorithm that simulates the rotor motion equation of a synchronous generator into the inner loop controller, devices such as energy storage converters and photovoltaic inverters exhibit similar inertial response and damping characteristics to traditional synchronous generators in terms of external characteristics. This includes automatic output voltage establishment, frequency droop regulation, and inertial suppression of power fluctuations. For example, when an energy storage converter using virtual synchronous machine control mode detects a drop in microgrid frequency due to a sudden increase in load, its internal control algorithm will automatically increase the output power according to the preset virtual inertia coefficient, simulating the process of a synchronous generator rotor releasing rotational kinetic energy, slowing down the rate of frequency drop, and buying time for power regulation of other power sources.

[0177] S2042. Determine the total inertia compensation requirements of the pre-defined microgrid based on the fault level.

[0178] Specifically, the fault level output by the source-load prediction model can be read first. Based on the preset fault level-anomaly type mapping table, the specific anomaly type corresponding to the fault level can be determined, including photovoltaic power output drop anomaly, load surge anomaly, or frequency limit violation anomaly. Then, the power change data related to the anomaly type can be extracted from the source-load prediction value. The simulation duration from the start time to the peak time of the anomaly and the integral value of power deficit or power surplus per unit time can be calculated as the energy deviation accumulation rate. Then, the system inertia time constant of the microgrid can be read. The simulation duration, energy deviation accumulation rate, and system inertia time constant can be substituted into the kinetic energy calculation formula to obtain the inertia compensation kinetic energy required to offset the impact of the single anomaly. Finally, the inertia compensation kinetic energy of multiple concurrent anomalies corresponding to the fault level can be coupled and solved. The overlapping part between the inertia demand of each anomaly is removed and the complementary part is superimposed to obtain the total inertia compensation demand required by the system to cope with the comprehensive abnormal conditions. This step is used to transform the abstract fault level and source-load prediction value into a quantitative value of total inertia demand with clear physical meaning, providing a total basis for the subsequent allocation of differentiated inertia compensation tasks to each distributed power source.

[0179] The total inertia compensation requirement refers to the total amount of virtual kinetic energy injected into the system to maintain frequency stability when an abnormal operating condition is predicted to occur. This requirement is determined by the severity and duration of the abnormality, as well as the system's own inertia level. For example, in a microgrid in an industrial park, the source-load prediction model outputs an abnormal operating condition in the next 3 minutes where the photovoltaic output will suddenly drop from 200kW to 120kW with a high fault level. Calculations show that the power deficit caused by this abnormality is 80kW, lasting for 3 seconds. Considering the system's inertia time constant of 2 seconds, the final calculated additional virtual kinetic energy required is... This value represents the total inertia compensation required by the system under this abnormal operating condition.

[0180] The fault level-anomaly type mapping table is a pre-built two-dimensional association table stored in the edge computing nodes of a microgrid. This table uses fault level as the index, with each row corresponding to a fault level and each column containing one or more specific anomaly type names and their characteristic parameters corresponding to that fault level. It is used to quickly parse the abstract fault level values ​​output by the source-load prediction model into anomaly categories with clear physical meaning, providing a type basis for subsequent inertia demand calculations. For example, in the mapping table of a certain microgrid, fault level 1 corresponds to a sudden drop in photovoltaic output anomaly, fault level 2 corresponds to a sudden increase in load anomaly, fault level 3 corresponds to a frequency exceeding the limit anomaly, and fault level 4 corresponds to a combined anomaly of sudden drop in photovoltaic output and sudden increase in load. When the source-load prediction model outputs a fault level of 4, the system can look up the table to determine that it needs to handle both the photovoltaic output drop and load increase anomalies simultaneously.

[0181] S2043. Based on the signal model and total inertia compensation requirements, distributed iterative calculations are performed through a preset neighbor communication mechanism to obtain the virtual inertia compensation amount required by each distributed power source.

[0182] Specifically, a distributed cooperative optimization solver can be constructed based on the signal model and the total inertia compensation requirement. This solver is deployed in the local controllers of each distributed power source. First, the electrical distance matrix and participation factor vector of each distributed power source under the system oscillation mode are extracted from the signal model. The electrical distance matrix quantifies the coupling strength between power sources in the dynamic response, and the participation factor vector represents the contribution of each power source to the dominant oscillation mode of the system. Then, based on the total inertia compensation requirement, the electrical distance matrix, and the participation factor vector, a distributed optimization objective function is constructed to minimize the non-uniformity of the spatial distribution of system inertia. This objective function prioritizes the inertia compensation task to power sources with high participation factors and close coupling with other power sources. Each distributed power source sets its own virtual inertia compensation value to zero as the initial value for iteration, and uses preset neighbor... The communication mechanism only exchanges the current virtual inertia compensation calculation value with the directly connected neighbor power sources. Each power source calculates the non-uniformity of the current inertia spatial distribution based on the received neighbor data and the locally stored electrical distance matrix, and performs gradient descent correction on its current iteration value based on the distributed optimization objective function. After each correction, each power source exchanges the updated calculation value with its neighbor again, and repeats the above process. When the difference between the current iteration values ​​of two adjacent iterations is less than the preset convergence threshold, the iteration stops. Each power source determines its current iteration value at the time of convergence as the virtual inertia compensation amount it needs to provide. This step is used to enable each power source to collaboratively converge to a set of differentiated optimal virtual inertia compensation amounts without the need for a central controller through a distributed computing method that only communicates with neighbors, thereby achieving a spatially balanced configuration of inertia support capabilities in the system.

[0183] Among them, the pre-set neighbor communication mechanism refers to a local communication method in which each distributed power source establishes a communication link and exchanges data only with the adjacent power sources directly connected to the power line. Each power source is pre-configured with the communication address and port of its neighboring power sources. During operation, it only receives and sends information from these direct neighbors and does not communicate directly with other non-neighboring power sources, thus forming a decentralized distributed communication network. For example, in a microgrid composed of five distributed power sources, power source A only establishes a communication connection with power sources B and C that are electrically closest to it. Power source A only sends its own virtual inertia compensation calculation value to B and C and only receives values ​​from B and C. Information from power sources D and E needs to be transmitted through multiple hops of B or C to reach A.

[0184] The purpose of using a pre-defined neighbor communication mechanism is to enable local information exchange between distributed power sources in a decentralized architecture. Each power source only needs to establish a communication link with its electrically connected neighbor power source to indirectly obtain information about the entire network through multi-hop transmission. This avoids dependence on a central controller and a global communication network, reduces the complexity and construction cost of the communication system, eliminates the problem of network-wide control failure caused by single point of failure or backbone communication interruption, and reduces transmission delay caused by long-distance communication. It enables each power source to complete local information exchange within milliseconds, providing a communication foundation for the rapid convergence of distributed iterative computing. Even in the event of partial failure of the communication link, the system can still maintain basic information exchange capabilities through the remaining neighbor links, thus improving the communication reliability of the grid-type microgrid control system.

[0185] Distributed iterative computation refers to a computational process in which, without a central controller, each distributed power source repeatedly updates its own state parameters by exchanging local information with its neighbors, gradually approaching the global optimum. In each iteration, each power source calculates its local gradient or deviation based on its current value and the received values ​​from its neighbors, and corrects its own value according to a preset update rule. After multiple iterations, the values ​​of all power sources converge to a stable solution that satisfies the global optimization objective. For example, if five power sources need to coordinate allocation... The total inertia compensation requirement is set to zero for each power source. In the first iteration, power source A calculates the local optimization gradient based on the values ​​of its neighbors B and C, and updates its own value accordingly. In the second iteration, the values ​​are corrected again based on the updated neighbor values. After repeating this process 20 times, the values ​​of all power sources no longer change, eventually converging to... , , , and The stable value.

[0186] The virtual inertia compensation required by each distributed power source refers to the specific inertia support task value allocated to each power source after distributed iterative calculation based on the total inertia compensation requirement, the participation level of each power source in the system oscillation mode, and the electrical coupling relationship. The compensation received by each power source may be different, but their combined value exactly equals the total inertia compensation requirement of the system. For example, the total inertia compensation requirement of the system is... After distributed iterative calculation, the No. 1 energy storage converter was allocated to Energy storage converter No. 2 is allocated to Photovoltaic inverter No. 1 was allocated to Photovoltaic inverter No. 2 was allocated to Photovoltaic inverter No. 3 was allocated to These five values ​​represent the amount of virtual kinetic energy that each power supply needs to provide in the event of an anomaly, and their sum is... This perfectly meets the overall system requirements.

[0187] The technical effect of this scheme in this embodiment is as follows: By establishing a small-signal model of a multi-virtual synchronous machine parallel system, the coupling relationship and oscillation characteristics of each distributed power source in dynamic response are characterized. Based on the fault level, the total inertia compensation requirement required by the system is calculated. Then, through a preset neighbor communication mechanism, distributed iterative calculation is performed to allocate the total inertia requirement to each distributed power source in a differentiated manner. This realizes the transformation from centralized unified allocation to distributed collaborative calculation, so that the virtual inertia compensation amount obtained by each power source matches its actual role and position in the system dynamics. This avoids the local inertia excess or deficiency caused by traditional average allocation or capacity allocation. It solves the problem of local instability and difficulty in effectively suppressing oscillations caused by the mismatch between inertia allocation and system oscillation mode when the grid-type microgrid is dealing with complex disturbances.

[0188] In one possible design, S2042, the total inertia compensation requirements of the pre-defined microgrid configuration are determined based on the fault level, including:

[0189] S20421. Analyze the abnormal situation corresponding to the fault level to obtain the pre-simulation duration and energy deviation accumulation rate of the abnormal situation corresponding to the fault level; wherein, the pre-simulation duration is used to represent the time span from the occurrence to the peak of the abnormal situation corresponding to the fault level, and the energy deviation accumulation rate refers to the integral value of the power deficit or power surplus of the pre-set grid-type microgrid within a unit time.

[0190] Specifically, the fault level output by the source-load prediction model can be read first. Based on the preset fault level-anomaly type lookup table, the specific anomaly type corresponding to the fault level can be determined, including photovoltaic power output drop anomaly, load surge anomaly, or frequency limit violation anomaly. Then, the power or frequency time series data sequence corresponding to the anomaly type can be extracted from the source-load prediction values. First-order difference calculation is performed on the time series data sequence to identify the starting time point when the power or frequency begins to deviate from the normal range and the peak time point when the maximum deviation value is reached. The time difference between the peak time point and the starting time point is determined as the rehearsal duration. At the same time, the instantaneous values ​​of power deficit or power surplus at each time point between the starting time point and the peak time point are calculated, and the cumulative energy deviation is obtained by time integration of these instantaneous values. The cumulative energy deviation is divided by the rehearsal duration to obtain the energy deviation accumulation rate per unit time. This step is used to extract characteristic parameters describing the anomaly development process from the time series data of the source-load prediction values, providing time-scale and energy-scale quantitative inputs for the subsequent calculation of the inertia compensation kinetic energy required to offset the anomaly.

[0191] The rehearsal duration refers to the time span from the initial moment of an anomaly identified in the source-load forecast to the moment the anomaly reaches its peak, indicating the speed of the anomaly's development. The cumulative energy deviation rate is the integral value of the system's power deficit or surplus per unit time, indicating the degree of energy imbalance caused by the anomaly. For example, in the source-load forecast of a microgrid, if the photovoltaic output starts to decrease from the 10th second and reaches its lowest point at the 13th second, the rehearsal duration is 3 seconds. Within these 3 seconds, the cumulative energy deviation is obtained by integrating the difference between the photovoltaic output and the normal value over time. Dividing by 3 seconds yields an energy deviation accumulation rate of 20kW.

[0192] S20422. Based on the rehearsal duration, energy deviation accumulation rate, and preset system inertia time constant of the preset grid-type microgrid, calculate the inertia compensation kinetic energy used to offset the impact of abnormal conditions corresponding to the fault level.

[0193] Specifically, the pre-analysis duration and energy deviation accumulation rate obtained from the anomaly analysis step, as well as the preset microgrid system inertia time constant, can be read first. The pre-analysis duration and the system inertia time constant are multiplied to obtain the time scale factor of the system inertia response. Then, the energy deviation accumulation rate is divided by this time scale factor to obtain the power deviation that can be offset by a unit of inertia. Finally, the total power deviation corresponding to the fault level is divided by the unit inertia offset amount to obtain the virtual kinetic energy value to be injected as the inertia compensation kinetic energy. This step is used to perform physical-level coupling calculations of the anomaly's time characteristics, energy characteristics, and the system's inherent inertia characteristics, transforming the abstract anomaly description into an inertia compensation kinetic energy requirement with clear physical meaning.

[0194] The preset system inertia time constant refers to the length of time during which the kinetic energy stored in all the virtual inertia of the entire microgrid can be continuously output at rated power under rated power. It is used to quantify the sustainability of the system's inertia support capability. Numerically, it is equal to the total system inertia kinetic energy divided by the system's rated power, reflecting the time margin by which the system maintains frequency stability through its own inertia when power deficit occurs. For example, if a microgrid has a rated power of 500kW and a total system inertia kinetic energy of... The system inertia time constant is 2 seconds, meaning that in the event of a 500kW power deficit, the system can maintain its rated power output for 2 seconds without frequency collapse solely due to its own inertia. The inertia compensation kinetic energy to offset the impact of the abnormal situation corresponding to the fault level refers to the total amount of virtual kinetic energy injected into the system to eliminate the power imbalance caused by a specific abnormality. This value is determined by the power deficit caused by the abnormality, the duration of the abnormality, and the system's inertia response characteristics, and is used to quantify the amount of inertia energy that needs to be supplemented. For example, if an abnormality causes a power deficit of 200kW and the rehearsal duration is 3 seconds, the calculated additional virtual kinetic energy required is... ,this This refers to the inertial compensation kinetic energy to offset the impact of the anomaly, meaning that within 3 seconds of the anomaly occurring, each distributed power source needs to work together to provide a total of The energy supply is used to maintain the stability of the system frequency.

[0195] S20423. Couple multiple inertia compensation kinetic energies corresponding to the abnormal conditions corresponding to the fault level to obtain the total inertia compensation requirement required by the preset grid-type microgrid.

[0196] Specifically, we can first read the inertia compensation kinetic energy of each of the multiple abnormal situations corresponding to the fault level, along with their corresponding pre-simulation duration and start time. Align the start times of all abnormal situations to the same zero point. Expand the inertia compensation kinetic energy of each abnormality into a time-varying kinetic energy output curve according to its pre-simulation duration. Superimpose the kinetic energy values ​​of the output curves of each abnormality at the same time point to obtain a comprehensive kinetic energy demand curve that varies with time. Extract the maximum kinetic energy value on this curve as the peak inertia demand of the system. Simultaneously, calculate the area of ​​integration of this curve from start to finish as the total inertia energy of the system. The demand is calculated by weighting and combining the peak inertia demand and the total inertia energy demand to obtain the total inertia compensation demand required for the pre-defined grid-type microgrid. This step is used to merge the inertia compensation kinetic energy of multiple possible simultaneous or sequential abnormal situations, integrating the dispersed single abnormal inertia demand into a total inertia compensation demand that can comprehensively reflect the impact of complex abnormal operating conditions. This ensures that the final total demand can not only meet the instantaneous inertia support requirements at the abnormal peak moment, but also cover the total energy consumption during the entire duration of the abnormality, providing an accurate total amount basis for the subsequent allocation of differentiated inertia compensation tasks to various distributed power sources.

[0197] The technical effect of this solution in this embodiment is as follows: by analyzing the abnormal situation corresponding to the fault level, two key characteristic parameters, namely the pre-simulation duration and the energy deviation accumulation rate, are extracted. Combined with the system inertia time constant, the inertia compensation kinetic energy required to offset the abnormality is calculated. Then, the inertia compensation kinetic energy of multiple possible concurrent abnormalities is coupled and solved to finally obtain the total inertia compensation requirement of the system. This makes the total inertia requirement no longer dependent on empirical formulas or simple mappings, but is calculated in a physical sense based on the time scale of the abnormality, the energy accumulation rate, and the inherent inertia characteristics of the system. This avoids the insufficient or excessive waste of inertia compensation caused by the inability to quantify the impact of compound abnormalities in traditional methods. It solves the problem of system transient instability or low resource utilization efficiency caused by inaccurate evaluation of inertia requirement when facing multiple concurrent abnormal operating conditions in grid-type microgrids.

[0198] In one possible design, S2043, based on the signal model and total inertia compensation requirements, performs distributed iterative calculations through a pre-defined neighbor communication mechanism to obtain the virtual inertia compensation amount required by each distributed power source, including:

[0199] S20431. Based on the signal model, extract the electrical distance matrix and participation factor vector of each distributed power source under the system oscillation mode; where the system oscillation mode refers to the inherent motion pattern of relative swaying between distributed power sources in a multi-virtual synchronous machine parallel system after being disturbed; the electrical distance matrix is ​​used to represent the coupling strength between distributed power sources in the dynamic response; and the participation factor vector is used to represent the degree of contribution of each distributed power source to the system oscillation mode.

[0200] Specifically, eigenvalue analysis and mode decomposition can be performed based on the established signal model. First, the state-space expression of the small-signal model is converted into a transfer function matrix. The Laplace transform of this transfer function matrix is ​​then performed to obtain the characteristic polynomial of the system. Solving the characteristic polynomial yields all the eigenvalues ​​of the system. The oscillation modes corresponding to the eigenvalues ​​whose real parts are close to the imaginary axis are the dominant oscillation modes of the system. The left and right eigenvector matrices corresponding to the dominant oscillation modes are extracted. The Hadamard product of the left and right eigenvector matrices is calculated to obtain the participation factor matrix of each state variable with respect to the dominant oscillation mode. From this participation factor matrix, the participation factor values ​​corresponding to the power angle states of each distributed power source are extracted to form the participation factor vector. At the same time, the electrical distance is calculated based on the coupling coefficients between the state variables in the state matrix. The electrical distance matrix is ​​constructed based on the ratio of mutual impedance to self impedance between each power source. This step is used to parse the participation factor vector describing the importance of each power source in the dominant oscillation mode and the electrical distance matrix describing the strength of dynamic coupling between power sources from the small-signal model. This provides the input basis for the subsequent construction of a distributed optimization objective function aimed at minimizing the inhomogeneity of the inertia spatial distribution.

[0201] Among them, the system oscillation mode refers to the inherent motion pattern of relative oscillation among the distributed power sources in a multi-virtual synchronous machine parallel system after being disturbed. It is uniquely determined by the system's eigenvalues ​​and eigenvectors. Each oscillation mode corresponds to a specific oscillation frequency and a set of mode shape coefficients describing the degree of participation of each power source. It reflects the oscillation transfer path of energy between different power sources in the dynamic process of the system. For example, in a microgrid with five distributed power sources, the system identifies a dominant oscillation mode with a frequency of 2Hz. The eigenvector of this mode shows that the power angles of power sources 1 and 3 oscillate in the same direction, while power sources 2, 4 and 5 oscillate in opposite directions. This specific oscillation pattern is the oscillation mode of the system.

[0202] The participation factor vector is a numerical vector describing the contribution of each distributed power source to the dominant oscillation mode of the system. Each element in the vector corresponds to a power source. The larger the element value, the higher the participation of the power source in the oscillation mode, and the more significant its damping or excitation effect on the oscillation mode. For example, in the above-mentioned 2Hz dominant oscillation mode, the calculated participation factor vector is [0.35, 0.12, 0.28, 0.15, 0.10]. This means that the participation factor of power source 1 is 0.35, making it the most important participant in the oscillation mode. The participation factor of power source 3 is 0.28, which is the second largest. The participation factor of power source 5 is 0.10, which is the smallest. This indicates that when suppressing the oscillation mode, prioritizing the adjustment of the control parameters of power sources 1 and 3 can achieve better results.

[0203] The electrical distance matrix is ​​a two-dimensional matrix describing the coupling strength between distributed power sources during dynamic response. Each element in the matrix represents the degree of electrical coupling between two power sources. The smaller the value, the closer the electrical distance and the stronger the dynamic coupling; the larger the value, the farther the electrical distance and the weaker the dynamic coupling. This matrix is ​​calculated from the small-signal model parameters of the system and reflects the electrical path impedance of disturbance propagation from one power source to another. For example, in the electrical distance matrix of five power sources, the element value between power source 1 and power source 2 is 0.15, and the element value between power source 1 and power source 5 is 0.62. This indicates that power source 1 and power source 2 are electrically close and their dynamic responses strongly influence each other, while power source 1 and power source 5 are farther apart and their dynamic responses weakly influence each other. Therefore, when allocating virtual inertia compensation, it is necessary to ensure that the inertia allocation values ​​between power sources with close electrical distances are coordinated and consistent to avoid new oscillations caused by excessive local inertia differences.

[0204] The Hadamard product is an element-wise multiplication operation between two matrices or vectors of the same dimension. The resulting matrix or vector has the same dimensions as the original matrix, and each element is equal to the product of the elements of the two matrices at that position. Unlike matrix multiplication, the Hadamard product does not involve row and column summation; it only performs element-wise point-to-point multiplication. For example, if matrix A is [1, 2; 3, 4] and matrix B is [5, 6; 7, 8], then the Hadamard product of A and B is [1×5, 2×6; 3×7, 4×8] = [5, 12; 21, 32]. In factorization, the Hadamard product of the left eigenvector matrix L and the right eigenvector matrix R is... Each element × That is, the participation factor of the j-th state variable in the i-th oscillation mode. It refers to the element in the i-th row and j-th column of the left eigenvector matrix L. It refers to the element in the i-th row and j-th column of the right eigenvector matrix R.

[0205] S20432. Based on the total inertia compensation requirement, electrical distance matrix, and participation factor vector, a distributed optimization objective function is constructed with the goal of minimizing the non-uniformity of the spatial distribution of system inertia.

[0206] Specifically, we can first read the total inertia compensation requirement as the total constraint of the optimization problem. Then, we read the value of each element in the electrical distance matrix to calculate the dynamic coupling strength coefficient between each distributed power source. We use the reciprocal of the electrical distance between each power source as the weight to construct the first metric of the inertia spatial distribution non-uniformity. This metric measures the degree of difference in inertia allocation values ​​between adjacent power sources. Next, we read the contribution value of each power source to the dominant oscillation mode in the participation factor vector. We use the reciprocal of the participation factor of each power source as the weight to construct the second metric of the inertia spatial distribution non-uniformity. This metric measures the power source inertia allocation value and its contribution to the oscillation mode. The degree of matching importance is determined, and finally, the first and second metric terms are weighted and summed to obtain the complete objective function for the non-uniformity of the spatial distribution of system inertia. The total inertia compensation requirement is equal to the sum of the inertia allocation values ​​of each power source as an equality constraint, and the inertia allocation values ​​of each power source within the range of zero to the upper limit are used as inequality constraints. Together, they constitute the distributed optimization objective function. This step is used to establish a mathematical expression that can quantify the merits of the inertia allocation scheme, so that the subsequent distributed iterative calculations have a clear objective orientation, ensuring that the final allocation result can not only meet the total inertia requirement, but also achieve a spatially balanced configuration of inertia support capacity in the system.

[0207] Among them, the non-uniformity of system inertia spatial distribution refers to the quantitative index of the deviation between the virtual inertia compensation amount actually allocated by each distributed power source and the ideal distribution determined according to the electrical distance and participation factor. This index is measured by calculating the difference in inertia allocation values ​​between adjacent power sources and the matching degree between the inertia allocation values ​​of each power source and the participation factor. The larger the value, the more uneven the distribution of inertia in the system, and the easier it is to cause local oscillations or energy concentration.

[0208] In a distributed computing architecture, the objective function is a mathematical expression constructed by each local node based on local information to guide its own iterative update direction. This function typically consists of a deviation term between the local state variable and the neighbor's state variable, as well as a matching term between the local state variable and the global constraints. Each node achieves consistency of the global optimization objective by minimizing its own objective function.

[0209] S20433. Initialize the current iteration value of each distributed power source to a preset value; wherein, the current iteration value of each distributed power source refers to the current virtual inertia compensation value of each distributed power source.

[0210] Specifically, an initialization module based on local register assignment can be constructed. This module is deployed in the local controller of each distributed power source. When the trigger signal output by the event triggering layer meets the preset conditions, the local controller of each distributed power source immediately reads its internal preset initial value register. The value stored in the register is zero. The controller writes the value into the current iteration value storage unit to complete the initialization of the current iteration value. At the same time, the controller informs the neighboring power sources of the initialization state through a preset neighbor communication mechanism. This step is used to set a unified starting point for distributed iterative computation, so that all power sources start the iterative process from the same zero starting point, ensuring that the subsequent iterative update and convergence process have a clear initial state.

[0211] S20434. The current iteration value is iteratively updated among the distributed power sources through a preset neighbor communication mechanism, and the current iteration value after each iteration is corrected based on the distributed optimization objective function.

[0212] Specifically, at the beginning of each iteration, each power source can send its current iteration value to all directly connected neighbor power sources through a preset neighbor communication mechanism, and simultaneously receive the current iteration values ​​sent by all neighbor power sources. Each power source calculates the spatial non-uniformity gradient component of its inertia allocation with its neighbors based on the received neighbor current iteration values ​​and the locally stored electrical distance matrix, and calculates the matching degree gradient component of its inertia allocation and oscillation mode contribution based on the locally stored participation factor vector. The two gradient components are weighted and summed to obtain the total corrected gradient. The current iteration value is obtained by subtracting the product of the total corrected gradient and the preset step size from its current iteration value. The updated current iteration value is written to the local storage unit as the current iteration value for the next iteration, and simultaneously sent to neighbor power sources again through the neighbor communication mechanism. This step is used so that, without a central controller, each power source can gradually correct its own inertia allocation value by exchanging information with its neighbors, so that the inertia allocation values ​​of all power sources converge to a stable set of values ​​that satisfy the optimization objective and total constraints after multiple iterations.

[0213] S20435. When the difference between the current iteration values ​​of two adjacent iterations is less than the preset convergence threshold, it is determined that the iteration has converged, and the current iteration value of each distributed power source at the time of iteration convergence is determined as the virtual inertia compensation amount required by each distributed power source.

[0214] Specifically, after each iteration update, each power source calculates the absolute value of the difference between the current iteration value and the current iteration value of the previous iteration. This absolute value is compared with a preset convergence threshold stored locally. When the absolute value is less than the preset convergence threshold, the power source sets its convergence flag to true locally and broadcasts its own convergence flag and current iteration value to all neighboring power sources through a preset neighbor communication mechanism. Each power source continuously receives the convergence flags from neighboring power sources. When a power source detects that its own and all neighboring power sources' convergence flags are true, it determines that the local area has reached a convergence state. This power source continues to propagate this local convergence state to its neighbors. After multiple iterations, when all power sources confirm that the convergence flags of all power sources in the entire network are true, each power source locks its current iteration value after the last iteration update and writes it into the final virtual inertia compensation register as the virtual inertia compensation amount required by each distributed power source. This step is used to achieve a unified convergence judgment across the entire network in a distributed architecture, ensuring that the final value obtained by all power sources at the end of the iteration is a set of stable solutions that meet the optimization objectives and are mutually coordinated, providing accurate instruction values ​​for subsequent execution of virtual inertia compensation control.

[0215] The technical effect of this scheme in this embodiment is as follows: By extracting the electrical distance matrix and participation factor vector of each distributed power source in the system oscillation mode from the signal model, the dynamic coupling strength between power sources and their contribution to the dominant oscillation mode are characterized. Based on the total inertia compensation requirement, a distributed optimization objective function is constructed with the goal of minimizing the spatial non-uniformity of system inertia distribution. Then, distributed iterative calculation is performed through a preset neighbor communication mechanism, so that each power source gradually converges to its own differentiated optimal virtual inertia compensation amount by exchanging information only with its neighbors. The scheme realizes the leap from centralized global optimization to distributed collaborative optimization, so that the compensation amount obtained by each power source is accurately matched with its actual status and spatial position in the oscillation mode. This avoids the inertia resource mismatch or difficulty in eliminating local weak points caused by ignoring oscillation characteristics and spatial distribution in traditional methods. It solves the problems of high response delay, strong communication dependence and large spatial non-uniformity caused by reliance on centralized communication and global calculation when the grid-type microgrid allocates differentiated inertia.

[0216] Figure 3 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 2 In this embodiment, in Figure 2 Based on the provided embodiments, the control method for grid-connected microgrids is further explained. The grid-connected microgrid control method includes:

[0217] S301. Acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time.

[0218] S302. Input multiple source load status data and multiple scenario data into the preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the power output of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time.

[0219] S301-S302 are similar to S201-S202, and will not be described again in this embodiment.

[0220] S303. Acquire multiple real-time data; wherein, the multiple real-time data are used to represent the operating status of the preset grid-type microgrid at the current moment.

[0221] Specifically, voltage, current, and power acquisition units can be deployed at distributed power sources, energy storage devices, load nodes, and key locations on lines within a grid-type microgrid. Each acquisition unit directly collects the corresponding electrical operating values ​​at the current moment, and the collected multiple values ​​are aggregated to form multiple real-time data. This step is used to provide the collaborative decision-making model with the real operating values ​​at the current moment, so that the subsequently generated control strategy can be consistent with the current operating state of the microgrid.

[0222] Among them, multiple real-time data refer to electrical parameters that are directly collected at the current moment and reflect the actual operating status of the grid-type microgrid. These include specific operating values ​​such as voltage, current, and power of distributed power sources, energy storage devices, loads, and transmission lines at the current moment. For example, the output voltage and output power of photovoltaic power sources, the charging and discharging current and port voltage of energy storage batteries, the power consumption and access current of industrial loads, and the line voltage and line current of the microgrid backbone lines at the current moment.

[0223] S304. Input multiple real-time data, source load prediction values ​​and fault levels into the preset collaborative decision-making model to obtain multiple control strategies.

[0224] Specifically, multiple real-time data, source load predictions, and fault levels can be input into a pre-defined collaborative decision-making model. The collaborative decision-making model consists of three components: a data input layer, an information integration layer, and a strategy generation layer. The data input layer receives real-time data, source load predictions, and fault levels and transmits these data to the information integration layer. The information integration layer performs state integration processing on the real-time data, trend integration processing on the source load predictions, and level integration processing on the fault levels. Then, it merges the three types of processed information to form comprehensive decision information and transmits it to the strategy generation layer. The strategy generation layer generates and outputs various control strategies adapted to the microgrid's operating state based on the comprehensive decision information. This step is used to provide multiple optional control strategies for the subsequent selection and execution of the target control strategy.

[0225] The collaborative decision-making model is used to comprehensively process real-time data, source load predictions, and fault levels, and output multiple control strategies. This model consists of three parts: a data input layer, an information integration layer, and a strategy generation layer. It can receive various types of operational information and generate corresponding control strategies. It can select multiple sets of real-time data, source load predictions, fault levels, and corresponding control strategies from historical operation as samples. By repeatedly learning from these samples and correcting the parameters, a collaborative decision-making model with information processing and strategy output capabilities can be formed. For example, a directly usable collaborative decision-making model can be formed by using real-time electrical data, predicted power, fault level information, and actual control strategy combinations from multiple sets of different operating states of a microgrid in the past.

[0226] S305. Based on the source load prediction value and the fault level, the triggering condition is judged to obtain the judgment result; wherein, the judgment result includes meeting the triggering condition, which is used to indicate that the preset grid-type microgrid has any one of the preset abnormal conditions in a preset second time period. The triggering condition judgment is based on the preset power threshold optimized according to the load fluctuation frequency and photovoltaic fluctuation rate and the source load prediction value, as well as the preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level.

[0227] S306. When the judgment result is that the triggering condition is met, calculate the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control based on the source load prediction value and fault level, and determine the target control strategy from multiple preset control strategies based on the virtual inertia compensation amount, and execute the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid within the preset second time period.

[0228] S305-S306 are similar to S203-S204, and will not be described again in this embodiment.

[0229] The technical effect of this solution in this embodiment is that by introducing real-time operating data at the current moment after obtaining the source load prediction value and fault level, and combining it with a preset collaborative decision-making model to dynamically generate multiple control strategies, the predicted results can be organically combined with the real-time operating status, so that the generation of control strategies fits the actual operating conditions of the system, improves the pertinence and reliability of the control strategies, ensures the accuracy of subsequent control execution, and solves the problem of insufficient matching between control strategies and real-time operating conditions.

[0230] Figure 4 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 3 In this embodiment, in Figure 2 Based on the provided embodiments, the control method for grid-connected microgrids is further explained. The grid-connected microgrid control method includes:

[0231] S401. Acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time.

[0232] S402. Input multiple source load status data and multiple scenario data into the preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the power output of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time.

[0233] S403. Determine the triggering conditions based on the source load prediction value and the fault level to obtain the judgment result; wherein, the judgment result includes meeting the triggering conditions. Meeting the triggering conditions is used to indicate that the preset grid-type microgrid has any one of the preset abnormal conditions in a preset second time period. The triggering condition judgment is based on the preset power threshold optimized according to the load fluctuation frequency and photovoltaic fluctuation rate and the source load prediction value, as well as the preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level.

[0234] S404. When the judgment result is that the triggering condition is met, calculate the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control based on the source load prediction value and fault level, and determine the target control strategy from multiple preset control strategies based on the virtual inertia compensation amount, and execute the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid within the preset second time period.

[0235] S401-S404 are similar to S201-S204, and will not be described again in this embodiment.

[0236] S405. Obtain multiple effect data; where multiple effect data refers to the actual operating status data of the preset grid-type microgrid after the target control strategy is executed.

[0237] Specifically, after the target control strategy is executed, the actual operating values ​​such as voltage, current, and power corresponding to the strategy execution can be collected by the data acquisition units on the distributed power sources, energy storage, loads, and lines in the microgrid. These values ​​are then centrally summarized and formatted to form multiple effect data. This step is used to provide real operating samples required for training the source-load prediction model and the collaborative decision-making model, and to provide a basis for updating the model parameters.

[0238] Among them, many effect data are electrical parameters that are directly collected after the target control strategy is implemented, reflecting the actual operating effect of the grid-type microgrid. These include actual operating values ​​such as voltage, current, power, and operating status of distributed power sources, energy storage devices, loads, and main lines after the strategy is implemented. They can truly reflect the stable state of the microgrid after the strategy is implemented. After the strategy is implemented, relevant actual operating values ​​can be collected and summarized by various acquisition units in the microgrid. For example, data such as the actual output power of photovoltaic power sources, the actual charging and discharging power of energy storage batteries, the actual power consumption of loads, and the stable voltage and current of lines after the strategy is implemented.

[0239] S406. Using multiple effect data as training samples, update the parameters of the preset source load prediction model and the preset collaborative decision-making model.

[0240] Specifically, multiple performance data can be matched with the previously output source load prediction values ​​and fault levels of the source load prediction model, as well as the multiple control strategies previously output by the collaborative decision-making model. Using the performance data as a reference standard, the parameters related to feature extraction and data fusion in the source load prediction model are adjusted sequentially, and the parameters related to information integration and strategy generation in the collaborative decision-making model are adjusted in turn. After the correction is completed, the operating parameters of the model are redefined and saved. This process updates the parameters of the source load prediction model and the collaborative decision-making model. This step allows the model to correct its own parameters based on the actual operating data, providing a parameter basis that is more in line with the actual operation of the microgrid for subsequent data processing and output results.

[0241] The technical effect of this solution in this embodiment is as follows: After eliminating system anomalies by executing the target control strategy, the source load prediction model and the collaborative decision-making model are updated online by collecting actual operating effect data as training samples. This enables continuous iterative optimization of the model, continuously improving the accuracy of source load prediction and the reliability of control strategy generation. It allows the system to maintain good dynamic response and adaptive adjustment capabilities during long-term operation and complex scenario changes, thereby improving control efficiency and operational stability. This solves the problem that fixed model parameters are difficult to adapt to long-term operating condition changes.

[0242] Figure 5 A flowchart illustrating the grid-type microgrid control method provided in this application embodiment. Figure 4 In this embodiment, in Figure 2 Based on the provided embodiments, the control method for grid-connected microgrids is further explained. The grid-connected microgrid control method includes:

[0243] S501. Obtain the load fluctuation frequency and photovoltaic volatility of the preset grid-type microgrid; wherein, the load fluctuation frequency is used to indicate the severity of the disturbance to the load, the photovoltaic volatility refers to the change range of photovoltaic power generation within a preset unit time, and the photovoltaic power generation refers to the real-time power generation of the photovoltaic power station in the preset grid-type microgrid.

[0244] Specifically, voltage and current signals at key nodes in a grid-type microgrid can be acquired in real time. Time-frequency domain characteristics of load power are extracted using signal processing methods such as Fourier transform or wavelet transform, and the load fluctuation frequency, characterizing the severity of load disturbance, is calculated. Simultaneously, DC-side voltage, current, and AC-side output power data of the photovoltaic power station are continuously acquired. The rate of change of photovoltaic power generation per unit time is calculated using a differential algorithm and used as the photovoltaic volatility. This step dynamically corrects the preset initial power threshold, allowing the power threshold used as the trigger condition to adaptively adjust according to the actual fluctuation characteristics of the load and the real-time changes in photovoltaic output. This avoids overly frequent control actions or slow responses due to a fixed threshold when load fluctuations are severe or photovoltaic output changes frequently, improving the accuracy of trigger condition judgment and the rationality of control strategy execution.

[0245] Load fluctuation frequency refers to the number of times the load power fluctuates around its average value per unit time, used to quantify the severity of disturbances to the load. In actual power systems, the load is not constant but fluctuates due to random changes in user electricity consumption, the start-up and shutdown of large equipment, and external grid disturbances. By collecting real-time load power data sequences and extracting their fluctuation components using signal processing methods, the load fluctuation frequency can be calculated. The higher the fluctuation frequency, the more drastic the load changes in a short period of time, and the greater the impact on the power balance and frequency stability of the microgrid.

[0246] Photovoltaic volatility refers to the variation in the power output of a photovoltaic (PV) power plant within a preset unit of time. It is usually expressed as an absolute value or percentage and is used to quantify the instability of PV output. Since PV power generation depends on solar irradiance, factors such as cloud cover, shadow movement, and sudden weather changes can cause rapid changes in PV output within a short period. By continuously collecting real-time power output data from the PV power plant and calculating the power difference between adjacent time points or within a unit of time, PV volatility can be obtained. A higher volatility indicates more drastic changes in PV output within a unit of time, placing higher demands on the power regulation capabilities of the microgrid.

[0247] S502. Optimize the preset initial power threshold based on the load fluctuation frequency and photovoltaic fluctuation rate to obtain the preset power threshold.

[0248] Specifically, a dynamic correction function can be constructed based on load fluctuation frequency and photovoltaic (PV) volatility. These two factors are used as input variables, and a dynamic adjustment coefficient is calculated using pre-set weighting factors. This dynamic adjustment coefficient is then multiplied by a preset initial power threshold to obtain a power threshold suitable for the current operating state. For example, when the load fluctuation frequency or PV volatility increases, the dynamic adjustment coefficient increases accordingly, ensuring the optimized power threshold is higher than the initial threshold. This avoids triggering unnecessary control actions due to an excessively low threshold when system disturbances are significant. This step adjusts the power threshold in the triggering conditions in real time according to the current level of system disturbance, allowing the threshold to adaptively change with the actual fluctuations in load and PV. This ensures a more reasonable starting point for control actions during periods of severe load fluctuations, enabling timely responses when truly needed while effectively suppressing frequent adjustments caused by small fluctuations, thus improving the accuracy and effectiveness of the control strategy.

[0249] S503. Obtain the historical number of abnormal triggers and the current operating mode of the preset grid-type microgrid; wherein, the historical number of abnormal triggers refers to the number of times abnormal situations occur in the preset grid-type microgrid within the preset first time period, and the current operating mode includes grid-connected mode and off-grid mode.

[0250] Specifically, the historical operation logs stored in the local controller of the grid-connected microgrid can be read. The timestamps and exception types of each triggered exception handling procedure within a preset first time period are extracted from the log records. The number of historical exception triggers is obtained by counting these records. Simultaneously, the open / close status signal of the circuit breaker at the grid connection point and the operating mode flag of the microgrid main controller are read. If the circuit breaker at the grid connection point is closed and the mode flag is "grid connected," the current operating mode is determined to be grid-connected; if the circuit breaker is open and the mode flag is "off-grid," the current operating mode is determined to be off-grid. This step is used to obtain key basic data for optimizing the level threshold. The number of historical exception triggers reflects the frequency of disturbances the system has experienced over a period of time, and the current operating mode represents the current grid-connected or off-grid status of the system. These two parameters together serve as the basis for subsequent adjustments to the level threshold, enabling the optimization of the level threshold to combine historical operating experience with the current operating scenario, thereby accurately setting the judgment criteria for triggering exception handling.

[0251] The historical anomaly trigger count refers to the total number of times that a grid-type microgrid triggers corresponding processing procedures due to anomalies such as power imbalance, frequency exceeding limits, and voltage exceeding limits within a preset first time period. This data can be obtained by reading the operation log stored in the local controller, which records the timestamp and anomaly type of each anomaly event. By counting these records, the historical anomaly trigger count can be obtained, which directly reflects the actual frequency of disturbances and anomalies experienced by the system over a past period.

[0252] The current operating mode refers to the grid-connected or off-grid status of the microgrid at the current moment. This is determined by reading the open / closed status signal of the circuit breaker at the grid connection point and the operating mode flag of the microgrid's main controller. If the circuit breaker at the grid connection point is closed and the flag displays "grid-connected," the current operating mode is determined to be grid-connected, and the microgrid operates connected to the main grid. If the circuit breaker at the grid connection point is open and the flag displays "off-grid," the current operating mode is determined to be off-grid, and the microgrid operates independently. These two modes represent distinctly different operating scenarios for the microgrid, with different requirements for control strategies and different criteria for anomaly detection.

[0253] S504. Optimize the preset initial level threshold based on the number of historical abnormal triggers and the current operating mode to obtain the preset level threshold.

[0254] Specifically, a two-dimensional lookup table can be pre-constructed. The rows of this table correspond to different historical anomaly trigger frequency ranges, and the columns correspond to the two current operating modes: grid-connected and off-grid. Each cell in the table stores a preset correction coefficient. After obtaining the historical anomaly trigger frequency and the current operating mode, the range to which the historical anomaly trigger frequency belongs is first determined. Then, combined with the current operating mode, the corresponding cell in the two-dimensional lookup table is located, and the correction coefficient in that cell is read. This correction coefficient is then multiplied by a preset initial level threshold to calculate the optimized level threshold. This step is used to specifically adjust the level threshold in the triggering conditions based on the historical frequency of anomalies in the system over a period of time and the current grid-connected or off-grid status. This makes the level threshold value more closely match the current actual operating background of the system, avoiding excessively high or low sensitivity to anomalies due to a fixed threshold after frequent historical anomalies or operating mode switching. This improves the accuracy of trigger condition judgment and the rationality of the timing of control strategy activation.

[0255] S505. If the predicted source load value is greater than the preset power threshold, or the fault level is higher than the preset level threshold, the judgment result will be determined as meeting the triggering condition.

[0256] S506. If the predicted source load value is less than or equal to the preset power threshold, and the fault level is equal to or lower than the preset level threshold, the judgment result will be determined as not meeting the triggering conditions.

[0257] S505-S506 are similar to S2031-S2032, and will not be described again in this embodiment.

[0258] The technical effect of this solution in this embodiment is as follows: by combining the load fluctuation frequency and photovoltaic volatility of the grid-type microgrid to optimize the power threshold, and optimizing the level threshold based on the number of historical abnormal triggers and the current operating mode, the judgment criteria for triggering conditions can adaptively match the real-time operating conditions and historical operating patterns of the system, thereby improving the accuracy of triggering condition judgment. This effectively avoids the problems of false triggering and missed triggering caused by fixed thresholds, provides a realistic judgment basis for the accurate activation of subsequent control strategies, ensures the timeliness and rationality of the system's dynamic response, and solves the problem that fixed thresholds are difficult to adapt to changes in operating conditions, leading to inaccurate trigger judgment.

[0259] With the advancement of new power system construction, grid-connected microgrids, possessing autonomous voltage / frequency support capabilities, have become a core carrier for distributed energy consumption and grid resilience enhancement. Their core requirement is to achieve precise coordinated control between distributed power sources (photovoltaics, energy storage, etc.) and loads, ensuring friendly interaction when connected to the grid and autonomous, stable operation when disconnected from the grid. Especially in the scenario of a "black start" due to a main grid fault, rapid source-load balance and system recovery are necessary.

[0260] While major companies and institutions have already deployed related technologies, significant shortcomings remain: Existing AC / DC hybrid microgrid series-parallel interconnection structures and multi-entity collaborative control architectures, while solving the challenge of large-scale networking, rely on traditional centralized control modes and backbone communication links, making them prone to control failures when edge nodes malfunction. Furthermore, they lack dynamic optimization of control strategies based on load characteristics. Existing string-type grid-based photovoltaic-storage solutions focus on improving the performance of core grid components, but their collaborative control relies on passive adjustments based on preset thresholds, lacking the ability to predict and proactively respond to sudden changes in source and load, resulting in insufficient dynamic adaptability. Existing technologies can achieve millisecond-level "black start" for grid-based energy storage, but after startup, source-load collaboration relies on fixed power allocation logic, unable to dynamically adjust based on load type (important / normal) and power output fluctuations, limiting energy utilization efficiency. Existing grid-based solutions primarily focus on optimizing the grid characteristics of single devices (inverters / energy storage converters), failing to form a collaborative control system covering all elements of "source-load-storage," and employing periodic data transmission modes, leading to high communication redundancy and control latency.

[0261] The core pain points of existing technologies can be summarized as follows: First, the control mode is rigid, with centralized control reliability relying on communication and distributed control lacking global coordination; second, the adjustment response is lagging, as control based on periodic sampling or fixed threshold triggering cannot predict sudden changes in source and load (such as sudden photovoltaic drops or load start-ups and shutdowns); third, artificial intelligence (AI) empowerment is insufficient, failing to combine edge computing to achieve real-time derivation and self-learning of AI models, resulting in poor adaptability of control strategies; and fourth, scenario adaptation is limited, lacking differentiated collaborative logic design for multiple scenarios such as grid-connected / off-grid switching and black start. Therefore, developing a grid-based microgrid source-load collaborative control technology that combines high reliability, fast response, and strong adaptability has become the key to industry breakthroughs.

[0262] Current source-load coordinated control technologies for grid-connected microgrids are mainly divided into three categories:

[0263] 1. Centralized Coordinated Control Scheme: A central controller collects source and load data from the entire microgrid, generates unified control commands based on a preset algorithm, and distributes them to each distributed power source and load controller to achieve global power balance. This scheme relies on real-time data transmission through a backbone communication network and is suitable for large-scale microgrid network deployments.

[0264] 2. Equipment-level grid construction + distributed regulation scheme: Inverters / energy storage converters have independent grid construction capabilities (virtual synchronous machine technology). They achieve autonomous power regulation based on algorithms such as droop control by collecting voltage and frequency data locally. They only receive unified dispatch instructions from the grid side when connected to the grid, and lack active coordination between source and load.

[0265] 3. Fixed threshold triggering control scheme: When a preset event such as main grid power failure occurs, the grid-type energy storage is triggered to start automatically, and the power is allocated according to a fixed ratio to drive the power supply and load to recover. After starting, it maintains a fixed power output mode without dynamic optimization mechanism.

[0266] The shortcomings of existing technology are:

[0267] 1. The contradiction between reliability and real-time performance: Centralized solutions rely on backbone communication, and link interruption leads to control failure; distributed solutions lack global coordination and are prone to source-load imbalance. Furthermore, both use periodic data transmission, resulting in large communication redundancy and control delays generally exceeding 100ms, making them unable to cope with instantaneous disturbances such as sudden drops in photovoltaic power or load fluctuations.

[0268] 2. Lack of predictive and proactive adjustment capabilities: Existing solutions are all "passive response after an event occurs", requiring the detection of voltage / frequency deviations before adjustment is initiated. The "black start" solution is only triggered after the main grid loses power, making it impossible to predict source load fluctuations in advance, which can easily lead to transient instability of the system.

[0269] 3. Limited AI application scenarios: Without deploying AI models in conjunction with edge computing, offline data analysis is only performed in the cloud, which makes it impossible to achieve real-time optimization and self-learning of control strategies and makes it difficult to adapt to the dynamic needs of different load types and different operating scenarios.

[0270] 4. Poor adaptability to multiple scenarios: The control logic for scenarios such as grid-connected / off-grid switching, black start, and normal operation is independent of each other, requiring manual parameter switching. Power surges are prone to occur during the switching process, affecting system stability.

[0271] This application also provides a method and system for source-load coordinated control of a grid-connected microgrid based on event triggering and edge AI, the specific objectives of which include:

[0272] 1. Improve control reliability and real-time performance: By adopting the "edge AI node + event triggering" architecture, the reliance on backbone communication is reduced, communication redundancy is reduced, and the control latency is shortened to less than 20ms, so as to cope with instantaneous source load disturbances.

[0273] 2. Achieve early prediction of power generation and load fluctuations: Based on AI prediction models deployed at the edge, predict the trend of photovoltaic output and load changes 3-5 minutes in advance, transforming passive response into active adjustment and avoiding system transient instability.

[0274] 3. Enhanced multi-scenario adaptive capabilities: Through AI models, control patterns in different scenarios (grid-connected / off-grid / black start) are learned and collaborative strategies are dynamically optimized, enabling smooth switching without manual intervention.

[0275] 4. Improve energy utilization efficiency: accurately match source and load output, prioritize power supply to important loads, optimize energy storage charging and discharging rhythm, and maximize the absorption rate of distributed energy while ensuring stability.

[0276] The overall technical architecture of a grid-based microgrid source-load coordinated control method and system based on event triggering and edge AI includes a perception layer, an edge AI control layer, an event triggering layer, an execution layer, and a collaborative linkage layer. These layers work together to achieve end-to-end source-load coordinated management of the entire process, from prediction to triggering to control to linkage. The core logic is as follows: the perception layer collects source-load and system operation data in real time; the edge AI control layer deploys predictive and decision-making models to predict source-load changes and generate basic control strategies; the event triggering layer dynamically monitors triggering conditions and initiates precise control as needed; the execution layer executes control commands to achieve source-load coordination; and the collaborative linkage layer ensures smooth switching across multiple scenarios and fault self-healing.

[0277] Figure 6 This is an overall architecture diagram of a grid-based microgrid source-load coordinated control system based on event triggering and edge AI, provided in an embodiment of this application. Figure 6In this structure, 1 is the perception layer, 11 is the distributed power monitoring unit, 12 is the load monitoring unit, and 13 is the system status monitoring unit; 2 is the edge AI control layer, 21 is the source-load prediction model, 22 is the collaborative decision-making model, and 23 is the model self-learning unit; 3 is the event triggering layer, 31 is the trigger condition configuration unit, 32 is the event recognition unit, and 33 is the control mode switching unit; 4 is the execution layer, 41 is the energy storage PCS control unit, 42 is the photovoltaic inverter control unit, and 43 is the flexible load control unit; 5 is the collaborative linkage layer, 51 is the main grid communication interface, 52 is the black start coordination unit, and 53 is the fault self-healing unit; 6 is the grid-type microgrid body, and 7 is the main grid.

[0278] Figure 6 In the process, the perception layer 1 is connected to the main body 6 of the grid-type microgrid to collect all elements of operation data. The edge AI control layer 2 is deployed at the local edge node and communicates bidirectionally with the perception layer 1. The event triggering layer 3 is connected in series between the edge AI control layer 2 and the execution layer 4 to dynamically trigger control actions. The execution layer 4 is connected to the energy storage, photovoltaic, load and other equipment in the microgrid. The collaborative linkage layer 5 is connected to the edge AI control layer 2 and the main grid 7 respectively to realize multi-scenario collaboration and fault response.

[0279] The perception layer 1 includes a distributed power monitoring unit 11, a load monitoring unit 12, and a system status monitoring unit 13. It adopts a "high-precision acquisition + edge preprocessing" mode: the distributed power monitoring unit 11 collects data such as photovoltaic output, energy storage SOC, and charging and discharging power in real time, with a sampling frequency of 1kHz; the load monitoring unit 12 collects power data according to the "important / ordinary" load classification, and at the same time monitors the adjustable capacity of flexible loads (such as air conditioners and charging piles); the system status monitoring unit 13 collects data such as microgrid voltage, frequency, and grid connection switch status. After outlier removal and standardization preprocessing, the data is transmitted to the edge AI control layer 2 to avoid invalid data occupying communication resources.

[0280] Edge AI control layer 2 is the core of the event-triggered and edge AI-based microgrid source-load collaborative control method and system. It is deployed on local edge computing nodes (without relying on cloud computing power) and includes a source-load prediction model 21, a collaborative decision-making model 22, and a model self-learning unit 23 to achieve a closed loop of "prediction-decision-optimization".

[0281] The source-load prediction model 21 adopts a Transformer+Light GBM hybrid model architecture. Input data includes historical source-load data, real-time operational data, and scenario characteristic data (weather, load type, production shifts, etc.). Outputs are the photovoltaic power output prediction (error ≤ ±2%), load power prediction (error ≤ ±3%), and load fluctuation fault level (high / medium / low) for the next 3-5 minutes. Compared to traditional Long Short-Term Memory (LSTM) network models, prediction accuracy is improved by more than 15%, and the derivation latency is ≤ 5ms, making it suitable for edge node computing power requirements.

[0282] The collaborative decision-making model 22 takes "system stability + maximum absorption + minimum cost" as multiple objectives, inputs predicted data and real-time system status data, and dynamically generates control strategies: In grid-connected mode, it optimizes the charging and discharging power of energy storage and the operating status of flexible loads to ensure that the power fluctuation at the grid connection point is ≤±5%; In off-grid mode, it allocates the output of each distributed power source, prioritizes the power supply of important loads, and maintains the voltage deviation ≤±1% and the frequency deviation ≤±0.2Hz; In the black start scenario, it generates instructions such as the energy storage start-up power and the order of load phased commissioning.

[0283] The model self-learning unit 23 is based on reinforcement learning algorithm to continuously learn control effect data (such as stabilization time and absorption rate) under different scenarios (grid-connected / off-grid / black start), dynamically optimize the parameters of the prediction model and decision model, and adapt to new source load characteristics (such as new photovoltaic units or changes in load type) without manual intervention.

[0284] Event triggering layer 3 breaks away from the traditional periodic control model, achieving precise regulation through a dual logic of "predictive triggering + abnormal triggering":

[0285] Triggering condition configuration unit 31: Preset multiple types of triggering conditions, including: ① Prediction conditions (the load fluctuation fault level output by the source load prediction model is "high", and the photovoltaic output sudden change is ≥20%); ② Abnormal conditions (voltage deviation > ±1%, frequency deviation > ±0.2Hz, and main grid power failure); ③ Scenario switching conditions (grid-connected / off-grid switching command, and black start command).

[0286] Event recognition unit 32: compares the data from the perception layer with the predicted data output by the edge AI control layer in real time, accurately identifies the triggering event, and generates triggering signals (such as "photovoltaic sudden drop prediction triggering" and "main grid power failure anomaly triggering"), with a triggering delay of ≤10ms.

[0287] Control mode switching unit 33: Upon receiving a trigger signal, it quickly invokes the corresponding scene control strategy generated by the edge AI control layer and sends it to the execution layer. Simultaneously, it shields unnecessary data transmission, reducing communication redundancy. Compared to periodic control, communication volume is reduced by more than 60%.

[0288] The execution layer 4 includes an energy storage inverter control unit 41, a photovoltaic inverter control unit 42, and a flexible load control unit 43, all of which adopt a grid-type control architecture (with virtual synchronous machine functionality): After receiving control commands from the event triggering layer, the energy storage inverter control unit 41 quickly adjusts the charging and discharging power and virtual inertia; the photovoltaic inverter control unit 42 optimizes the maximum power point tracking (MPPT) strategy to smooth out power output fluctuations; and the flexible load control unit 43 adjusts the operating status of adjustable loads according to commands (such as reducing air conditioning power and suspending unnecessary charging pile charging), with an execution delay of ≤5ms.

[0289] The collaborative linkage layer 5 ensures stable operation in multiple scenarios: it interacts with the main grid 7 through the main grid communication interface 51, receives grid connection and dispatch instructions, and uploads the microgrid operation status at the same time; the black start coordination unit 52 cooperates with the execution layer to complete the black start process, coordinating in the order of "energy storage start-up → driving photovoltaic → batch load input" to achieve millisecond-level system autonomous start-up; the fault self-healing unit 53 quickly redistributes source load output when equipment fails to ensure continuous system operation.

[0290] Figure 7 The flowchart of the control method for the source-load coordinated control method and system of the grid-type microgrid based on event triggering and edge AI provided in the embodiments of this application is as follows: Figure 7 As shown, the control method steps of the source-load coordinated control method and system for grid-type microgrids based on event triggering and edge AI are as follows:

[0291] S1: Data Acquisition and Preprocessing: The perception layer collects distributed power sources, loads, and system status data according to categories. After outlier removal and standardization, the data is transmitted to the edge AI control layer and event triggering layer.

[0292] S2: Source Load Prediction and Strategy Generation: The source load prediction model of the edge AI control layer outputs source load prediction data and fluctuation fault level for the next 3-5 minutes. The collaborative decision-making model combines real-time data to generate basic control strategies for multiple scenarios, and the model self-learning unit continuously optimizes model parameters.

[0293] S3: Event Recognition and Triggering: The event triggering layer compares data in real time, identifies predicted, abnormal, or scene switching triggering events, generates triggering signals, and calls the corresponding control strategies.

[0294] S4: Source-load coordinated execution: The execution layer adjusts the operating status of energy storage, photovoltaic and flexible loads according to the control strategy to achieve precise source-load matching.

[0295] S5: Collaborative Linkage and Closed-Loop Optimization: The collaborative linkage layer ensures multi-scenario switching and fault self-healing, while feeding back execution effect data to the edge AI control layer to complete model self-learning and strategy optimization.

[0296] For example, taking an industrial microgrid (including a 200kW photovoltaic system, an energy storage system (200kWh / 100kW), four DC charging piles (60kW / unit), and two heat pumps (20kW / unit), all with power regulation and status feedback functions) as an example, the application process of the technical solution is explained in detail:

[0297] 1. System Setup: The perception layer deploys voltage / current sensors, SOC monitors, and other equipment, with a data acquisition frequency of 1kHz; the edge AI control layer is deployed on a local edge server, loading source-load prediction models and collaborative decision-making models; the event triggering layer presets trigger conditions (PV sudden change ≥20%, voltage deviation >±1%, main grid power failure); the execution layer connects to grid-type energy storage converters, PV inverters, and flexible load controllers; and the collaborative linkage layer connects to the main grid communication system.

[0298] 2. Normal operating scenario:

[0299] (1) Data acquisition: The sensing layer collects real-time data on photovoltaic output of 180kW, energy storage SOC of 70%, important load power of 90kW, flexible load power of 60kW, system voltage of 380V±0.5%, and frequency of 50Hz±0.1Hz.

[0300] (2) AI prediction: The source-load prediction model, combined with meteorological data, predicts that the photovoltaic output will drop sharply to 120kW in the next 3 minutes (fluctuation ≥33%, fault level "high"); the collaborative decision-making model generates the strategy: energy storage discharge 30kW, flexible load reduced by 20kW.

[0301] (3) Event triggering: The event recognition unit recognizes "photovoltaic sudden drop prediction triggering" and triggers the control mode switch.

[0302] (4) Execution and optimization: The execution layer adjusts according to the strategy, the energy storage discharge is 30kW, and the power of the flexible load is reduced to 40kW; after the photovoltaic output is reduced to 120kW, the system power is balanced, and the voltage and frequency do not fluctuate significantly; the model self-learning unit records the control effect and optimizes the prediction parameters.

[0303] 3. Black boot scenario:

[0304] (1) Event triggering: When the main network loses power, the event triggering layer recognizes "abnormal triggering of main network power loss" and calls the black start control strategy.

[0305] (2) Coordinated execution: The energy storage converter starts automatically at 50kW power according to the instruction to establish a stable voltage / frequency; it drives the photovoltaic inverter to start and gradually increases the photovoltaic output to 80kW; according to the principle of "putting important loads first and putting them into operation in batches", 60kW of important loads are put into operation first, and then 30kW of important loads are put into operation.

[0306] (3) Stable operation: After black start is completed (time < 500ms), the system operates stably off-grid with a voltage of 380V ± 0.8% and a frequency of 50Hz ± 0.15Hz, which is 50% faster than the existing black start technology (stable time > 1s).

[0307] 4. Comparative test: Compared with the existing grid-type photovoltaic-storage scheme, the control delay of this application is reduced from 120ms to 18ms, the voltage fluctuation amplitude under photovoltaic sudden drop scenario is reduced from ±2.3% to ±0.7%, the communication volume is reduced by 65%, and it can still be stably controlled when the backbone communication is interrupted, and the reliability is significantly improved.

[0308] A source-load coordinated control method and system for grid-connected microgrids based on event triggering and edge AI achieves multi-dimensional breakthroughs compared to existing technologies through the core design of "edge AI prediction + event triggering control," with the specific effects as follows:

[0309] 1. Improved reliability and real-time performance: Control latency ≤20ms, more than 80% improvement over existing solutions (≥100ms), can cope with instantaneous source load disturbances; edge deployment + event triggering reduces dependence on backbone communication, and can still control independently when communication is interrupted, improving reliability by more than 90%.

[0310] 2. Achieve early prediction of source load fluctuations: The AI ​​prediction model predicts source load changes 3-5 minutes in advance, transforming passive response into active adjustment. In scenarios such as sudden drop in photovoltaic power and sudden load changes, the system fluctuation amplitude is reduced by more than 70%, avoiding transient instability.

[0311] 3. Enhanced multi-scenario adaptive capability: Through model self-learning and event-triggered mode switching, it can adapt to grid-connected / off-grid / black start scenarios without manual intervention. The switching process is free of power impact, and the black start time is less than 500ms, which is better than the industry average.

[0312] 4. Optimize energy utilization efficiency: Accurately match source and load output, and increase the distributed energy consumption rate by 5%-8%; prioritize power supply to important loads, and improve the flexible load regulation accuracy to ±1kW, reducing energy waste.

[0313] 5. Easy integration and industrialization: The core modules are compatible with existing network-type devices without the need for major hardware modifications; combined with edge computing and AI algorithm accumulation, industrialization can be achieved quickly.

[0314] The key points of a source-load coordinated control method and system for a grid-connected microgrid based on event triggering and edge AI are as follows:

[0315] 1. Edge AI dual-model architecture: The Transformer + Light GBM hybrid prediction model achieves high-precision source load prediction, and the collaborative decision-making model achieves multi-objective optimization to adapt to the computing power requirements of edge nodes.

[0316] 2. Dual event triggering logic: It integrates "predictive + abnormal + scene switching" triggering conditions to break the periodic control mode and reduce communication redundancy and control delay.

[0317] 3. Multi-scenario self-adaptive collaboration: Through model self-learning and linkage control, it can achieve smooth switching between scenarios such as grid connection / off-grid / black start without manual intervention.

[0318] 4. Networked execution layer collaboration: Execution layer devices have network capabilities, fast response speed, and ensure accurate implementation of control commands.

[0319] The protection point of the control method of a grid-connected microgrid source-load coordinated control method and system based on event triggering and edge AI is:

[0320] 1. A grid-based microgrid source-load coordinated control system based on event triggering and edge AI, comprising a sensing layer, an edge AI control layer, an event triggering layer, an execution layer, and a coordinated linkage layer; the sensing layer is used to collect source-load and system status data; the edge AI control layer is deployed at local edge nodes and includes a source-load prediction model, a collaborative decision-making model, and a model self-learning unit, used to predict source-load changes and generate multi-scenario control strategies; the event triggering layer is used to identify triggering events and call corresponding control strategies; the execution layer is used to execute control commands; and the coordinated linkage layer is used for multi-scenario switching and fault self-healing.

[0321] In the grid-type microgrid source-load collaborative control system based on event triggering and edge AI, the source-load prediction model adopts a Transformer+Light GBM hybrid architecture. The input data includes historical source-load data, real-time operation data and scene feature data. The output is the source-load prediction value and fluctuation fault level for the next 3-5 minutes. The prediction error is ≤±3% and the derivation delay is ≤5ms.

[0322] In the grid-type microgrid source-load coordinated control system based on event triggering and edge AI, the event triggering layer includes a trigger condition configuration unit, an event identification unit, and a control mode switching unit; the trigger conditions include prediction type (source-load fluctuation fault level "high"), abnormal type (voltage / frequency deviation exceeds the standard) and scene switching type (grid-connected / off-grid / black start) conditions, with a trigger delay ≤10ms.

[0323] In the grid-based microgrid source-load coordinated control system based on event triggering and edge AI, the execution layer adopts a grid-based control architecture, including an energy storage converter control unit, a photovoltaic inverter control unit, and a flexible load control unit. The execution delay is ≤5ms and it has a virtual synchronous machine function.

[0324] In the grid-based microgrid source-load coordinated control system based on event triggering and edge AI, the coordinated linkage layer includes the main grid communication interface, the black start coordination unit, and the fault self-healing unit; the black start coordination unit can realize the coordinated control of energy storage self-start → photovoltaic start → load batch connection, with a black start time of <500ms.

[0325] The core of the grid-based microgrid source-load coordinated control system based on event triggering and edge AI is "edge AI prediction + event triggering control". Without deviating from the core concept, the following alternative solutions can be adopted:

[0326] 1. AI model replacement: The source load prediction model can adopt a GRU+XG Boost hybrid architecture to reduce the computing power requirements of edge nodes; the collaborative decision-making model can combine reinforcement learning and model predictive control (MPC) to further improve the accuracy of multi-objective optimization.

[0327] 2. Replacement of triggering logic: Fuzzy control algorithms can be introduced to optimize trigger condition thresholds and avoid false triggering in extreme scenarios; a "manual emergency trigger" mode can be added to adapt to special operation and maintenance needs.

[0328] 3. Execution Layer Alternatives: Execution layer devices can employ wide-bandgap semiconductor devices (such as SiC) to further reduce execution latency to <3ms; flexible load control can introduce wireless communication modules to adapt to distributed load scenarios. It should be noted that all the above alternative solutions must meet the core logic of "edge AI + event triggering," and their technical effects are all within the scope of protection of this application.

[0329] Figure 8 This is a schematic diagram of the structure of the grid-type microgrid control device provided in an embodiment of this application. Figure 8 As shown, the grid-type microgrid control device includes:

[0330] The first acquisition module 801 is used to acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the deadline of the preset first time period is no later than the current time.

[0331] The first input module 802 is used to input multiple source load status data and multiple scenario data into a preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the power output of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time.

[0332] The judgment module 803 is used to judge the triggering conditions based on the source load prediction value and the fault level, and obtain the judgment result. The judgment result includes meeting the triggering conditions. Meeting the triggering conditions means that the preset grid-type microgrid has any one of the preset abnormal conditions in a preset second time period. The triggering condition judgment is based on the preset power threshold optimized by the load fluctuation frequency and photovoltaic volatility and the source load prediction value, as well as the preset level threshold optimized by the historical number of abnormal triggers and the current operating mode and the fault level.

[0333] The determination module 804 is used to calculate the virtual inertia compensation amount required by each distributed power source participating in the control of the grid-type microgrid based on the source load prediction value and the fault level when the judgment result is that the triggering condition is met. Based on the virtual inertia compensation amount, the module determines the target control strategy from multiple preset control strategies and executes the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid within the preset second time period.

[0334] In one possible design, the preset source load prediction model includes multiple prediction models, and the first input module 802 includes:

[0335] The first input unit is used to input multiple scene data into a preset scene twin model to obtain a virtual running trajectory that is adapted to the multiple scene data; wherein, the virtual running trajectory refers to the multiple scene branches that evolve in the current scene within a preset second time period.

[0336] The first calculation unit is used to calculate the prediction response of each prediction model to each scenario branch, and to normalize and weight the prediction response of each prediction model to each scenario branch to obtain the robustness score of each prediction model under multiple scenario branches.

[0337] The fusion unit is used to calculate the matching degree between multiple scene data and the preset scene feature templates corresponding to each prediction model, and to perform multi-objective weighted fusion of the robustness score and matching degree of each prediction model under multiple scene branches to obtain the comprehensive fit index of each prediction model.

[0338] The second input unit is used to input multiple source load state data and multiple scenario data into the target prediction model to obtain the source load prediction value and fault level; wherein, the target prediction model refers to the model corresponding to the maximum value of the comprehensive adaptation index among multiple prediction models.

[0339] In one possible design, the grid-type microgrid control device also includes:

[0340] The second acquisition module is used to acquire multiple real-time data; wherein, the multiple real-time data are used to represent the operating status of the preset grid-type microgrid at the current moment.

[0341] The second input module is used to input multiple real-time data, source load prediction values, and fault levels into a preset collaborative decision-making model to obtain multiple control strategies.

[0342] In one possible design, the grid-type microgrid control device also includes:

[0343] The third acquisition module is used to acquire multiple effect data; among them, multiple effect data refers to the actual operating status data of the preset grid-type microgrid after the target control strategy is executed.

[0344] The update module is used to update the parameters of the preset source load prediction model and the preset collaborative decision-making model using multiple effect data as training samples.

[0345] In one possible design, the judgment result also includes the determination that the trigger condition is not met. The judgment module 803 includes:

[0346] The first determining unit is used to determine the triggering condition by judging the result if the predicted source load value is greater than the preset power threshold or the fault level is higher than the preset level threshold.

[0347] The second determining unit is used to determine the triggering condition as not being met if the predicted source load value is less than or equal to a preset power threshold and the fault level is equal to or lower than a preset level threshold.

[0348] In one possible design, the decision module 803 also includes:

[0349] The second acquisition unit is used to acquire the load fluctuation frequency and photovoltaic volatility of a preset grid-type microgrid; wherein, the load fluctuation frequency is used to represent the severity of the disturbance to the load, the photovoltaic volatility refers to the change range of photovoltaic power generation within a preset unit time, and the photovoltaic power generation refers to the real-time power generation of the photovoltaic power station in the preset grid-type microgrid.

[0350] The first optimization unit is used to optimize the preset initial power threshold based on the load fluctuation frequency and photovoltaic volatility to obtain the preset power threshold.

[0351] The third acquisition unit is used to acquire the historical number of abnormal triggers and the current operating mode of the preset grid-type microgrid; wherein, the historical number of abnormal triggers refers to the number of times abnormal situations occur in the preset grid-type microgrid within a preset first time period, and the current operating mode includes grid-connected mode and off-grid mode.

[0352] The second optimization unit is used to optimize the preset initial level threshold based on the number of historical anomaly triggers and the current operating mode to obtain the preset level threshold.

[0353] In one possible design, multiple source-load status data include power supply operation data, load operation data, and system operation data. The first acquisition module 801 includes:

[0354] The first acquisition unit is used to acquire power operation data through a preset distributed power monitoring unit. The power operation data includes photovoltaic power generation, energy storage SOC data, and charge / discharge power data. The energy storage SOC data is used to represent the remaining power of the energy storage battery in the preset grid-type microgrid, and the charge / discharge power data is used to represent the charging power and discharging power of the energy storage system in the preset grid-type microgrid.

[0355] The second acquisition unit is used to acquire load operation data through a preset load monitoring unit; wherein, the load operation data is used to represent the operating status of the load in the preset grid-type microgrid.

[0356] The third acquisition unit is used to acquire system operation data through the preset system status monitoring unit. The system operation data includes voltage data and frequency data. The voltage data refers to the voltage amplitude at the grid connection point in the preset grid-type microgrid, and the frequency data refers to the AC frequency at the grid connection point.

[0357] The fourth acquisition unit is used to acquire multiple scene data through a preset external interface.

[0358] In one possible design, module 804 is defined as including:

[0359] The establishment unit is used to establish a signal model of a multi-virtual synchronous machine parallel system based on multiple source load state data and multiple scenario data; wherein, the multi-virtual synchronous machine parallel system refers to a parallel operation system composed of multiple distributed power sources using virtual synchronous machine control mode in a pre-defined grid-type microgrid.

[0360] The third determining unit is used to determine the total inertia compensation requirement of the preset grid-type microgrid based on the fault level.

[0361] The second computing unit is used to perform distributed iterative calculations based on the signal model and total inertia compensation requirements, through a preset neighbor communication mechanism, to obtain the virtual inertia compensation amount required by each distributed power source.

[0362] In one possible design, the third determining unit includes:

[0363] The analysis component is used to analyze the abnormal situation corresponding to the fault level, and obtain the pre-simulation duration and energy deviation accumulation rate of the abnormal situation corresponding to the fault level. The pre-simulation duration is used to represent the time span from the occurrence to the peak of the abnormal situation corresponding to the fault level, and the energy deviation accumulation rate refers to the integral value of the power deficit or power surplus of the pre-set grid-type microgrid within a unit time.

[0364] The calculation component is used to calculate the inertial compensation kinetic energy to offset the impact of abnormal conditions corresponding to the fault level, based on the pre-simulation duration, energy deviation accumulation rate, and preset system inertia time constant of the preset grid-type microgrid.

[0365] The coupling component is used to couple multiple inertia compensation kinetic energies corresponding to the abnormal conditions of the fault level to obtain the total inertia compensation requirement of the preset grid-type microgrid.

[0366] In one possible design, the second computing unit includes:

[0367] An extraction component is used to extract the electrical distance matrix and participation factor vector of each distributed power source under the system oscillation mode based on the signal model. The system oscillation mode refers to the inherent motion pattern of relative oscillation among the distributed power sources in a multi-virtual synchronous machine parallel system after being disturbed. The electrical distance matrix is ​​used to represent the coupling strength between the distributed power sources in the dynamic response, and the participation factor vector is used to represent the degree of contribution of each distributed power source to the system oscillation mode.

[0368] Components are constructed to build a distributed optimization objective function based on the total inertia compensation requirement, the electrical distance matrix, and the participation factor vector, with the goal of minimizing the non-uniformity of the spatial distribution of system inertia.

[0369] An initialization component is used to initialize the current iteration value of each distributed power source to a preset value; where the current iteration value of each distributed power source refers to the current virtual inertia compensation value of each distributed power source.

[0370] The update component is used to iteratively update the current iteration value among distributed power sources through a preset neighbor communication mechanism, and to correct the current iteration value after each iteration based on the distributed optimization objective function.

[0371] The determination component is used to determine that the iteration has converged when the difference between the current iteration values ​​of two adjacent iterations is less than a preset convergence threshold, and to determine the current iteration value of each distributed power source at the time of iteration convergence as the amount of virtual inertia compensation required by each distributed power source.

[0372] The grid-type microgrid control device provided in this embodiment can execute... Figures 2 to 5The technical solution of the embodiment of the grid-type microgrid control method shown is implemented in accordance with the principle and technical effect of the method. Figures 2 to 5 The embodiments of the grid-type microgrid control method shown are similar and will not be described in detail here.

[0373] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application. Figure 9 As shown, the electronic device 90 includes at least one processor 901 and a memory 902. The electronic device 90 also includes a communication component 903. The processor 901, memory 902, and communication component 903 are connected via a bus 904.

[0374] In the specific implementation process, at least one processor 901 executes computer execution instructions stored in memory 902, so that at least one processor 901 is used to implement a grid-type microgrid control method of the above embodiment.

[0375] The specific implementation process of processor 901 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0376] In the above embodiments, it should be understood that the processor 901 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0377] The memory 902 may include high-speed RAM memory, and may also include non-volatile memory (NVM), such as at least one disk storage.

[0378] Bus 904 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 904 can be divided into address bus, data bus, control bus, etc. For ease of illustration, the bus 904 in the accompanying drawings of this application is not limited to only one bus or one type of bus.

[0379] The above description of the functions implemented by electronic devices and main control devices has introduced the solutions provided by the embodiments of the present invention. It is understood that, in order to implement the above functions, the electronic device or main control device includes hardware structures and / or software modules corresponding to the execution of each function. By combining the units and algorithm steps of the various examples described in the embodiments of the present invention, the embodiments of the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of the embodiments of the present invention.

[0380] This application also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, these instructions are used to implement a grid-based microgrid control method as described in the above embodiments. In the specific implementation of the aforementioned grid-based microgrid control method, each module can be implemented as a processor.

[0381] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0382] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the readable storage medium can exist as discrete components in an electronic device or a host device.

[0383] This application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement a grid-type microgrid control method according to the above embodiments.

[0384] The computer program is stored in a readable storage medium, and at least one processor can read the computer program from the readable storage medium and execute the computer program to perform the scheme provided in any of the above embodiments.

[0385] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0386] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A control method for a grid-type microgrid, characterized in that, include: Acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time; The multiple source load status data and the multiple scenario data are input into a preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the output power of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time; Triggering conditions are determined based on the predicted source load and the fault level to obtain a judgment result; wherein, the judgment result includes meeting the triggering conditions, which indicates that the preset grid-type microgrid has any one of the preset abnormal conditions within the preset second time period. The triggering condition judgment is based on a preset power threshold optimized according to the load fluctuation frequency and photovoltaic volatility and the predicted source load, and a preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level. When the judgment result is that the triggering condition is met, the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control is calculated based on the source load prediction value and the fault level. Based on the virtual inertia compensation amount, a target control strategy is determined from a plurality of preset control strategies and the target control strategy is executed to eliminate the abnormal situation that occurs in the preset grid-type microgrid during the preset second time period.

2. The microgrid control method according to claim 1, characterized in that, The preset source load prediction model includes multiple prediction models. The step of inputting the multiple source load state data and the multiple scenario data into the preset source load prediction model to obtain the source load prediction value and fault level includes: The multiple scene data are input into a preset scene twin model to obtain a virtual running trajectory adapted to the multiple scene data; wherein, the virtual running trajectory refers to the multiple scene branches that the current scene evolves into within the preset second time period; Calculate the prediction response of each prediction model to each scenario branch, and normalize and weight the prediction response of each prediction model to each scenario branch to obtain the robustness score of each prediction model under the multiple scenario branches. Calculate the matching degree between the multiple scene data and the preset scene feature templates corresponding to each prediction model, and perform multi-objective weighted fusion of the robustness score of each prediction model under the multiple scene branches and the matching degree to obtain the comprehensive fit index of each prediction model. The multiple source load state data and the multiple scenario data are input into the target prediction model to obtain the source load prediction value and the fault level; wherein, the target prediction model refers to the model corresponding to the maximum value of the comprehensive adaptation index among the multiple prediction models.

3. The microgrid control method according to claim 1, characterized in that, After inputting the multiple source load state data and the multiple scenario data into a preset source load prediction model to obtain the source load prediction value and fault level, the method further includes: Acquire multiple real-time data; wherein, the multiple real-time data are used to represent the operating status of the preset grid-type microgrid at the current moment; The multiple real-time data, the predicted source load value, and the fault level are input into a preset collaborative decision-making model to obtain the multiple control strategies.

4. The method according to claim 3, characterized in that, After executing the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid during the preset second time period, the method further includes: Acquire multiple effect data; wherein, the multiple effect data refers to the actual operating status data of the preset grid-type microgrid after the target control strategy is executed; Using the multiple effect data as training samples, the parameters of the preset source load prediction model and the preset collaborative decision-making model are updated.

5. The method according to claim 1, characterized in that, The judgment result also includes not meeting the triggering condition. The judgment based on the source load prediction value and the fault level to obtain the judgment result includes: If the predicted source load value is greater than the preset power threshold, or the fault level is higher than the preset level threshold, the judgment result is determined as the trigger condition being met. If the predicted source load value is less than or equal to the preset power threshold, and the fault level is equal to or lower than the preset level threshold, the judgment result is determined as the triggering condition not being met.

6. The method according to claim 5, characterized in that, Before determining that the triggering condition is met if the predicted source load value is greater than a preset power threshold or the fault level is higher than a preset level threshold, the method further includes: The load fluctuation frequency and photovoltaic volatility of the preset grid-type microgrid are obtained; wherein, the load fluctuation frequency is used to represent the severity of the disturbance to the load, the photovoltaic volatility refers to the change range of photovoltaic power generation within a preset unit time, and the photovoltaic power generation refers to the real-time power generation of the photovoltaic power station in the preset grid-type microgrid; The preset initial power threshold is optimized based on the load fluctuation frequency and the photovoltaic volatility to obtain the preset power threshold; The historical number of abnormal triggers and the current operating mode of the preset grid-type microgrid are obtained; wherein, the historical number of abnormal triggers refers to the number of times the preset grid-type microgrid has an abnormal situation within the preset first time period, and the current operating mode includes grid-connected mode and off-grid mode; The preset initial level threshold is optimized based on the historical number of abnormal triggers and the current operating mode to obtain the preset level threshold.

7. The method according to claim 6, characterized in that, The multiple source-load status data include power supply operation data, load operation data, and system operation data. Acquiring the multiple source-load status data and multiple scenario data includes: The power operation data is collected by a preset distributed power monitoring unit; wherein, the power operation data includes the photovoltaic power generation, energy storage SOC data and charge / discharge power data, the energy storage SOC data is used to represent the remaining power of the energy storage battery in the preset grid-type microgrid, and the charge / discharge power data is used to represent the charging power and discharging power of the energy storage system in the preset grid-type microgrid; The load operation data is collected by a preset load monitoring unit; wherein the load operation data is used to represent the operating status of the load in the preset grid-type microgrid; The system operation data is collected by a preset system status monitoring unit; wherein, the system operation data includes voltage data and frequency data, the voltage data refers to the voltage amplitude at the grid connection point in the preset grid-type microgrid, and the frequency data refers to the AC frequency at the grid connection point; The multiple scene data are obtained through a preset external interface.

8. The method according to claim 1, characterized in that, The calculation of the virtual inertia compensation required by each distributed power source participating in the grid-type microgrid control based on the predicted source load value and the fault level includes: Based on the multiple source load state data and the multiple scenario data, a signal model for a multi-virtual synchronous machine parallel system is established; wherein, the multi-virtual synchronous machine parallel system refers to a parallel operation system composed of multiple distributed power sources using virtual synchronous machine control mode in the preset grid-type microgrid; The total inertia compensation requirement for the preset grid-type microgrid is determined based on the fault level. Based on the signal model and the total inertia compensation requirement, distributed iterative calculations are performed through a preset neighbor communication mechanism to obtain the virtual inertia compensation amount required by each distributed power source.

9. The method according to claim 8, characterized in that, Determining the total inertia compensation requirement for the preset grid-type microgrid based on the fault level includes: The abnormal situation corresponding to the fault level is analyzed to obtain the pre-simulation duration and energy deviation accumulation rate of the abnormal situation corresponding to the fault level; wherein, the pre-simulation duration is used to represent the time span from the occurrence to the peak of the abnormal situation corresponding to the fault level, and the energy deviation accumulation rate refers to the integral value of the power deficit or power surplus of the preset grid-type microgrid in a unit time. Based on the rehearsal duration, the energy deviation accumulation rate, and the preset system inertia time constant of the preset grid-type microgrid, calculate the inertia compensation kinetic energy used to offset the impact of the abnormal situation corresponding to the fault level; By coupling multiple inertia compensation kinetic energies corresponding to the abnormal conditions of the fault level, the total inertia compensation requirement of the preset grid-type microgrid is obtained.

10. The method according to claim 8, characterized in that, Based on the signal model and the total inertia compensation requirement, a distributed iterative calculation is performed through a preset neighbor communication mechanism to obtain the virtual inertia compensation amount required by each distributed power source, including: Based on the signal model, the electrical distance matrix and participation factor vector of each distributed power source under the system oscillation mode are extracted; wherein, the system oscillation mode refers to the inherent motion pattern of relative swaying between the distributed power sources in a multi-virtual synchronous machine parallel system after being disturbed; the electrical distance matrix is ​​used to represent the coupling strength between the distributed power sources in the dynamic response; and the participation factor vector is used to represent the degree of contribution of each distributed power source to the system oscillation mode. Based on the total inertia compensation requirement, the electrical distance matrix, and the participation factor vector, a distributed optimization objective function is constructed with the goal of minimizing the non-uniformity of the spatial distribution of system inertia. The current iteration value of each distributed power source is initialized to a preset value; where the current iteration value of each distributed power source refers to the current virtual inertia compensation value of each distributed power source. The current iteration value is iteratively updated among the distributed power sources through the preset neighbor communication mechanism, and the current iteration value after each iteration is corrected based on the distributed optimization objective function; When the difference between the current iteration values ​​of two adjacent iterations is less than the preset convergence threshold, it is determined that the iteration has converged, and the current iteration value of each distributed power source at the time of iteration convergence is determined as the amount of virtual inertia compensation required by each distributed power source.

11. A grid-type microgrid control device, characterized in that, include: The first acquisition module is used to acquire multiple source-load status data and multiple scenario data; wherein, the multiple source-load status data are used to represent the operating status of the power source and load in the preset grid-type microgrid during a preset first time period, and the end time of the preset first time period is no later than the current time; The first input module is used to input the multiple source load status data and the multiple scenario data into a preset source load prediction model to obtain source load prediction values ​​and fault levels; wherein, the source load prediction values ​​are used to represent the output power of the power supply and the power demand of the load within a preset second time period, and the start time of the preset second time period is later than the current time; The judgment module is used to judge the triggering conditions based on the predicted source load value and the fault level, and obtain the judgment result; wherein, the judgment result includes meeting the triggering conditions, which is used to indicate that the preset grid-type microgrid has any one of the preset abnormal conditions within the preset second time period. The triggering condition judgment is based on the preset power threshold optimized according to the load fluctuation frequency and photovoltaic fluctuation rate and the predicted source load value, and on the preset level threshold optimized according to the historical number of abnormal triggers and the current operating mode and the fault level. The determination module is used to calculate the virtual inertia compensation amount required by each distributed power source participating in the grid-type microgrid control based on the source load prediction value and the fault level when the judgment result is that the trigger condition is met, and to determine the target control strategy from a plurality of preset control strategies based on the virtual inertia compensation amount, and to execute the target control strategy to eliminate the abnormal situation that occurs in the preset grid-type microgrid during the preset second time period.

12. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; When the processor executes the computer execution instructions stored in the memory, it is used to implement the grid-type microgrid control method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the grid-type microgrid control method as described in any one of claims 1 to 10.

14. A computer program product, characterized in that, It includes a computer program, which, when executed by a processor, is used to implement the grid-type microgrid control method as described in any one of claims 1 to 10.

Images