Source-load-storage modeling method, power grid voltage control method and system

Abstract

The present disclosure belongs to the technical field of urban power grid safety control, and particularly relates to a source-load-storage modeling method, a power grid voltage control method and system. The method comprises: carrying out fine modeling of photovoltaic, energy storage and type-specific load, collecting equipment operation state data and identifying parameters; building a double-layer control architecture of centralized coordination-distributed collaboration, deploying centralized coordinators and distributed collaborators and defining functions; under the control architecture, measuring power distribution network topology, source and load real-time and historical data, and performing dynamic cluster division; based on historical optimal solution database initialization, using ADMM distributed optimization algorithm to calculate cluster total power adjustment of each cluster; based on cluster total power adjustment, voltage deviation factor and resource margin factor, combining the model and model parameters to perform power distribution of photovoltaic reactive power adjustment and energy storage active power adjustment on the equipment connected to the node. Due to fine modeling, the voltage control is provided with a basis, and the accuracy and reliability of the control can be improved.

Description

Technical Field

[0001] This disclosure belongs to the field of urban power grid safety control technology, and in particular relates to source-load-storage modeling methods, power grid voltage control methods and systems. Background Technology

[0002] With the increasing penetration of distributed photovoltaic, energy storage, and other new energy equipment in urban power grids, the power grid structure is shifting from a traditional passive radial type to an active interactive type. "Source-load-storage" synergy has become the core architecture for the operation and control of new power systems. Currently, the power industry has formed a relatively mature power grid voltage stability control technology system, accumulating rich achievements in equipment modeling, control structures, and prevention and control strategies, as follows:

[0003] In terms of source-load-storage equipment modeling, existing technologies have achieved the characterization and model construction of various types of equipment: photovoltaics balances model accuracy and computational efficiency through techniques such as piecewise linearization, and introduces a prediction error reservation mechanism to cope with output uncertainty; energy storage modeling incorporates constraints such as charging and discharging power limits and upper and lower limits of state of charge (SOC), and some studies use binary state variables to identify the operating status to reduce charging and discharging switching losses; load modeling generally adopts the ZIP static model to describe power-voltage characteristics, and combines it with demand response mechanisms to achieve adjustable modeling of flexible loads, such as establishing a dynamic response model based on temperature constraints for commercial air conditioning loads, providing basic support for distribution network scheduling.

[0004] In terms of control structure design, both centralized and distributed control architectures have been implemented in engineering applications: centralized control aims at global optimization and coordinates controllable resources through the power grid energy management system, and is widely used in traditional power grid dispatching for scenarios such as line loss optimization and voltage regulation; distributed control decomposes system problems into sub-problems based on physical partitioning and achieves collaborative optimization through information interaction between adjacent areas, which can effectively reduce dependence on the central controller.

[0005] In terms of voltage control strategies, existing technologies for cluster division are mostly based on calculating electrical distances from the power grid topology and using clustering algorithms to achieve resource aggregation management, such as aggregating distributed resources into clusters by region for hierarchical regulation; in the optimization solution process, the ADMM algorithm can effectively reduce network active power loss by iteratively coordinating boundary variables; power allocation is mostly based on indicators such as voltage deviation to formulate adjustment strategies, and combining energy storage and demand response resources to achieve peak shaving and valley filling.

[0006] However, as the penetration rate of new energy sources increases, the frequency of grid voltage fluctuations rises, and fault control time needs to be compressed. Existing technologies show room for optimization in terms of model transient accuracy, architecture coordination timeliness, and source-load strategy adaptability. It is urgent to build source-load-storage modeling methods and grid voltage control methods to improve the accuracy and reliability of control. Summary of the Invention

[0007] To address the aforementioned issues, this disclosure provides a source-load-storage modeling method, a power grid voltage control method, and a system. Through refined modeling, the accuracy and reliability of control are improved.

[0008] This disclosure provides a source-load-storage modeling method based on a two-layer control architecture, which includes at least two distributed coordinators and one centralized coordinator, wherein one cluster corresponds to one distributed coordinator; the distributed coordinator connects to all nodes within its corresponding cluster, and the nodes include photovoltaic, energy storage and load; the centralized coordinator is deployed in the scheduling center;

[0009] Each distributed coordinator reports the collected running status data of each node in the corresponding cluster to the centralized coordinator.

[0010] The centralized coordinator uses operational status data to identify and update the model parameters of the pre-built source-load-storage model, and then distributes the updated model parameters to each distributed coordinator.

[0011] Each distributed coordinator receives model parameters and uses them to update the source-load-storage model.

[0012] The source-load-storage model includes: distributed photovoltaic model, load model and energy storage model;

[0013] The construction of a distributed photovoltaic model includes: characterizing the reactive power characteristics of photovoltaics;

[0014] The distributed photovoltaic (PV) model parameters include: steady-state parameters and fault ride-through parameters. The steady-state parameters include the internal resistance of a single diode; the fault ride-through parameters include the reactive power support factor in LVRT mode. Active power retention coefficient under HVRT mode Fault current limit With transient response time ;

[0015] The identification of distributed photovoltaic (PV) model parameters includes: minimizing the error between the PV characteristics calculated by the distributed PV model and the actual collected operating status data by using the least squares method based on fault weights, thereby identifying the PV model parameters.

[0016] Furthermore, the reactive power characteristics of photovoltaic (PV) power sources are characterized by the reactive power generated by distributed PV power sources and the rate of change constraint; the reactive power generated by distributed PV power sources is:

[0017]

[0018] in, This represents the reactive power generated by the photovoltaic power source at time t; Indicates the reactive power support coefficient; Indicates the rated voltage at the grid connection point; This represents the actual voltage at the photovoltaic grid connection point at time t; This is the active power retention coefficient under HVRT mode;

[0019] The current constraint of photovoltaic power sources is:

[0020]

[0021] in, This represents the output current of the photovoltaic power source at time t; For fault current limits; This refers to the transient response time of the photovoltaic power source. This indicates the maximum permissible change during the transient response period;

[0022] The identification of distributed photovoltaic (PV) model parameters specifically includes: using the following objective function to identify the PV model parameters:

[0023]

[0024] in, This indicates the error between the photovoltaic characteristics calculated by the distributed photovoltaic model and the actual collected operating status data; and These are the actual measured photovoltaic active power and the photovoltaic active power calculated using a distributed photovoltaic model, respectively. and S represents the actual measured photovoltaic reactive power and the reactive power calculated using the distributed photovoltaic model; S represents the time interval for steady-state data; F represents the time interval for fault data. and These are the weights for steady-state data and fault data, respectively; when the error When the minimum value is reached, the parameters of the distributed photovoltaic model achieve the optimal identification value.

[0025] Furthermore, the construction of the energy storage model includes: adopting an improved Thevenin model with dynamic SOC characteristics, constructing a model that includes the static characteristics of the energy storage battery terminal voltage, the dynamic change process of the polarization voltage, and the dynamic calculation of SOC. The dynamic change process of the polarization voltage is achieved by characterizing the decay and excitation process, and the dynamic calculation of SOC is determined by combining the charging and discharging efficiency and the rated capacity.

[0026] The energy storage model parameters include: the polarization resistance at time t under the SOC state and the internal resistance of the stored energy at time t under the SOC state;

[0027] The identification of energy storage model parameters includes: calculating using the recursive least squares method, introducing a forgetting factor to highlight the latest energy storage data, iteratively correcting the energy storage model parameters using the prediction error of the latest energy storage data, and periodically correcting the polarization resistance through HPPC testing.

[0028] Furthermore, the static characteristics of the energy storage battery terminal voltage are expressed as follows:

[0029] ;

[0030] in, This represents the terminal voltage of the energy storage system at time t; This represents the open-circuit voltage at time t under SOC (State of Charge) conditions. Let be the charging and discharging current at time t; Let be the internal resistance of the stored energy in the SOC state at time t; Polarization voltage;

[0031] The dynamic change process of polarization voltage is represented as follows:

[0032]

[0033] in, This represents the polarization resistance at time t under the SOC state; Indicates polarization capacitance;

[0034] The dynamic calculation of SOC is represented as follows:

[0035]

[0036] in, This represents the state of charge at time t; Represents the initial state of charge (SOC); Indicates the rated capacity of energy storage; express The charging and discharging efficiency at any given time;

[0037] The energy storage charge and discharge current constraint is expressed as:

[0038]

[0039] in, Indicates the time interval of energy storage current variation; This is the maximum allowable current. The maximum allowable change in current within a specified time period;

[0040] The regulation power model for energy storage is as follows:

[0041]

[0042] in, The power value that can be adjusted; The active power rated for energy storage.

[0043] Furthermore, the construction of the load model includes: for each type of load, constructing a composite model of the load's static power characteristic equation, dynamic response equation, and fault constraints, and introducing a first-order inertial element into the dynamic response equation to describe its transient process, representing the power ramp-up process when the grid voltage and current change abruptly.

[0044] The load model parameters include: static parameters: the active ZIP coefficient and reactive ZIP coefficient of the load; dynamic parameters: the active transient time constant and reactive transient time constant of the load; fault parameters: the surge in active power and the surge in reactive power during load faults.

[0045] The identification of load model parameters includes: collecting load voltage and power operation data; classifying load types based on average daily electricity consumption duration, peak-to-valley ratio, and reactive power-voltage sensitivity characteristics using clustering algorithms; and identifying load model parameters by setting weights according to the degree of load impact on the distribution network and using hierarchical weighted least squares method.

[0046] Furthermore, the hierarchical weighted least squares method has the following objective function:

[0047]

[0048] in, , These are the calculated values ​​of active power and reactive power of the load, obtained from the static power characteristic equation, the dynamic response equation, and the fault reference value, respectively. , These are the active power measurements of the load during normal operation and the reactive power measurements of the load during a fault. The weighting coefficient for load type m is defined based on the degree of impact of different types of loads on the power grid.

[0049] This disclosure also provides a power grid voltage control method, based on a source-load-storage model established by the above method, including:

[0050] The centralized coordinator divides the network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type.

[0051] The distributed coordinator reports the collected running status data of each node in the corresponding cluster to the centralized coordinator.

[0052] The centralized coordinator determines the global control target of each cluster based on the overall network topology and the received operating status data, and issues adjustment instructions to the corresponding distributed coordinators. The global control target includes the total power adjustment amount of the cluster.

[0053] Based on the received adjustment instructions and combined with the source-load-storage model, the distributed coordinator determines the power allocated to the devices connected to the nodes; it verifies the allocation results through the source-load-storage model and generates control instructions for each device.

[0054] Furthermore, the method also includes: the distributed coordinator determines the penetration rate of distributed photovoltaics in the cluster, the load type of the cluster, and the type of disturbance to the cluster based on the collected operating status data of the corresponding cluster, and reports it as the current scenario;

[0055] The centralized coordinator, based on the overall network topology and received operational status data, determines the global control objectives for each cluster and issues adjustment instructions to the corresponding distributed coordinators, including:

[0056] The centralized coordinator, based on the overall network topology, received operational status data, and current scenario data, initializes the ADMM algorithm using a pre-established ADMM historical best solution database, determines the global control objectives of each cluster, and issues adjustment instructions to the corresponding distributed coordinators.

[0057] The ADMM historical optimal solution database contains the correspondence between each historical scenario and the initial value. The historical scenarios include penetration rate, load type and disturbance type. The initial values ​​include: initial values ​​of Lagrange multipliers, initial control quantities of photovoltaic reactive power and initial control quantities of energy storage active power.

[0058] Furthermore, based on the overall network topology, received operational status data, and current scenario data, the centralized coordinator initializes the ADMM algorithm using a pre-established ADMM historical optimal solution database, determines the global control objectives for each cluster, and issues adjustment instructions to the corresponding distributed coordinators, including:

[0059] The centralized coordinator constructs an ADMM historical optimal solution database containing the correspondence between each historical scenario and the initial value;

[0060] For each cluster, the centralized coordinator calculates the similarity between the current scene of the cluster and each historical scene; and selects the initial value of the cluster based on the similarity value.

[0061] Substitute the initial values ​​and operating status data of each selected cluster into the global objective function, and use the ADMM algorithm for optimization to obtain the total active power adjustment and total reactive power adjustment for each cluster.

[0062] The total active power adjustment and total reactive power adjustment of each cluster are used as the global control targets for each cluster, and adjustment instructions are issued to the corresponding distributed coordinators.

[0063] Furthermore, based on the received adjustment instructions and in conjunction with the source-load-storage model, the distributed coordinator determines the power allocated to the devices connected to the nodes, including:

[0064] The distributed coordinator obtains the total active power adjustment and total reactive power adjustment of the cluster from the adjustment command;

[0065] The distributed coordinator uses the source-load-storage model and the local node operating status to determine the voltage deviation factor and resource margin factor of each node;

[0066] The distributed coordinator decomposes the total active power adjustment and total reactive power adjustment of the cluster according to the voltage deviation factor and resource margin factor of each node, and obtains the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value of each node.

[0067] The distributed coordinator uses the photovoltaic reactive power adjustment allocation value, the energy storage active power adjustment allocation value, and the source-load-storage model and model parameters of each node to determine the power allocated to the devices connected to the node.

[0068] Furthermore, the centralized coordinator divides the network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type, including:

[0069] The centralized coordinator determines the distribution network topology, node power, and equivalent impedance between nodes;

[0070] The centralized coordinator determines the net power correlation coefficient between nodes based on node power, and corrects the net power correlation coefficient between nodes based on whether the loads between nodes are of the same type.

[0071] The centralized coordinator determines the weighted electrical distance between nodes based on the equivalent impedance between nodes and the corrected net power correlation coefficient between nodes;

[0072] The centralized coordinator uses an improved K-means clustering algorithm to determine clusters in the entire distribution network based on load type proportion data, weighted electrical distance between nodes, and node voltage regulation resources.

[0073] Furthermore, the distributed coordinator determines the cluster load type based on the collected cluster's operational status data, including:

[0074] The distributed coordinator selects the dominant node in the cluster based on a comprehensive weighted score of weighted electrical distance centrality, power fluctuation coordination among nodes, and voltage regulation resource margin.

[0075] The distributed coordinator uses the load type of the dominant node in the cluster as the load type of the cluster.

[0076] Furthermore, the centralized coordinator performs similarity calculations between the current scene of each cluster and each historical scene, including:

[0077] Use the following formula to calculate the similarity between the current scene and historical scenes. :

[0078]

[0079] in, The weights for the impact of penetration rate, load type, and disturbance type on the optimization are: This represents the current penetration rate in various scenarios. For historical scene penetration rate; This refers to the degree of difference between the load type in the current scenario and the load type in historical scenarios. This is a perturbation type matching function; The type of disturbance in the current scene; This refers to the type of disturbance in the historical scene.

[0080] Furthermore, the centralized coordinator determines the weighted electrical distance between nodes based on the equivalent impedance between nodes and the corrected net power correlation coefficient between nodes, including:

[0081] The weighted electrical distance between nodes is determined using the following formula:

[0082]

[0083] in, Weighted electrical distance; For nodes i and nodes j The normalized value of the equivalent impedance; The power correlation coefficient, For synergy-weighted terms; , Let be the weighting coefficient, satisfying .

[0084] Furthermore, the distributed coordinator decomposes the total active power adjustment and total reactive power adjustment of the cluster according to the voltage deviation factor and resource margin factor of each node, obtaining the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value for each node, including:

[0085] The total active power adjustment of any cluster a can be decomposed using the following formula. :

[0086]

[0087] The total reactive power adjustment of any cluster a can be decomposed using the following formula. :

[0088]

[0089] in, This represents the active power adjustment of energy storage at node j. For standardized nodes j Voltage deviation factor; For standardized nodes j The active resource margin factor; The reactive power resource margin factor of node j after standardization; Let be the photovoltaic reactive power adjustment amount at node j; , The weight of the voltage deviation factor is set by the distributed coordinator based on the voltage state.

[0090] Furthermore, the global objective function is:

[0091]

[0092] ;

[0093]

[0094] Where L is the global objective function value; For adjacent clusters a and b Objective function between; For clusters a The objective function; For clusters b The objective function; and Clusters a , b The control variables all include photovoltaic reactive power control quantities and energy storage active power control quantities; These are the initial Lagrange multipliers; The penalty parameter is K; K is the number of clusters. Indicates the total number of nodes in the cluster; Represents a cluster a The reduction in photovoltaic power; Represents a cluster a The Middle i The voltage of each node; Indicates the node reference voltage; For clusters a Network loss; For three different items in the cluster a The weights in the objective function are defined by the centralized coordinator.

[0095] This disclosure also provides a source-load-storage modeling system based on a two-layer control architecture, which includes: at least two distributed coordinators and one centralized coordinator, wherein one cluster corresponds to one distributed coordinator; the distributed coordinator connects to all nodes within its corresponding cluster, and the nodes include: photovoltaic, energy storage and load; the centralized coordinator is deployed in the dispatch center;

[0096] Each distributed coordinator is used to report the collected operating status data of each node in the corresponding cluster to the centralized coordinator; and to receive model parameters and update the source-load-storage model using the model parameters; wherein, the source-load-storage model includes: distributed photovoltaic model, load model and energy storage model;

[0097] The centralized coordinator is used to identify and update the model parameters of the pre-built source-load-storage model using runtime status data, and then distributes the updated model parameters to each distributed coordinator.

[0098] The construction of a distributed photovoltaic model includes: characterizing the reactive power characteristics of photovoltaics;

[0099] The distributed photovoltaic (PV) model parameters include: steady-state parameters and fault ride-through parameters. The steady-state parameters include the internal resistance of a single diode; the fault ride-through parameters include the reactive power support factor in LVRT mode. Active power retention coefficient under HVRT mode Fault current limit With transient response time ;

[0100] The identification of distributed photovoltaic (PV) model parameters includes: minimizing the error between the PV characteristics calculated by the distributed PV model and the actual collected operating status data by using the least squares method based on fault weights, thereby identifying the PV model parameters.

[0101] This disclosure also provides a power grid voltage control system based on the above-described source-load-storage modeling system, comprising:

[0102] A centralized coordinator is used to divide the distribution network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type. Based on the overall distribution network topology and received operating status data, it determines the global control target of each cluster and issues adjustment instructions to the corresponding distributed coordinator. The global control target includes the total power adjustment of the cluster.

[0103] The distributed coordinator is used to report the collected operating status data of each node in the corresponding cluster to the centralized coordinator; based on the received adjustment instructions, combined with the source-load-storage model, it determines the power allocated to the devices connected to the nodes; it verifies the allocation results through the source-load-storage model and generates control instructions for each device.

[0104] Compared with the prior art, this disclosure has the following advantages:

[0105] 1. Significantly improved modeling accuracy: Through source-load-storage classification, full-characteristic refined modeling, and differentiated parameter identification strategies, the deviation between the model's calculated values ​​and the actual operating characteristics of the equipment is greatly reduced, providing accurate node characteristic support for voltage stability control and solving the problem of inaccurate characterization by traditional models;

[0106] 2. Faster control response: In the two-layer control architecture, the historical optimal solution is used to initialize ADMM, which reduces the number of optimization convergence times, meets the requirements of real-time control of distribution networks, and solves the problem of slow response in traditional control.

[0107] 3. Optimize resource utilization efficiency: Dynamic cluster partitioning ensures precise control of control targets, and dual-factor power allocation takes into account both the urgency and capacity of adjustment to avoid resource waste or insufficient control, while strictly adhering to equipment safety constraints to extend equipment life;

[0108] 4. Enhanced grid stability: The closed-loop control throughout the entire process covers all stages of "modeling-decision-execution-feedback", which can effectively cope with voltage fluctuations caused by the high penetration rate of new energy access, reduce the risk of voltage instability, and improve the safe operation level of urban power grid and the capacity for new energy consumption.

[0109] Other features and advantages of this disclosure will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0110] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0111] Figure 1 A schematic diagram of a power grid voltage stabilization control system according to an embodiment of the present disclosure is shown;

[0112] Figure 2 A flowchart of a parameter identification method for a load model according to an embodiment of the present disclosure is shown;

[0113] Figure 3 A flowchart of a power grid voltage control method according to an embodiment of the present disclosure is shown;

[0114] Figure 4A flowchart of a dynamic cluster partitioning method according to an embodiment of the present disclosure is shown;

[0115] Figure 5 A flowchart illustrating the process of a centralized coordinator determining the global control objectives of each cluster according to an embodiment of the present disclosure is shown.

[0116] Figure 6 A schematic flowchart illustrating the process of a distributed coordinator generating control commands for each device according to an embodiment of this disclosure is shown. Detailed Implementation

[0117] This disclosure involves conducting refined modeling of photovoltaic (PV), energy storage, and categorized loads, collecting equipment operating status data, and identifying parameters. A two-layer control architecture of centralized coordination and distributed collaboration is established, deploying a centralized coordinator and distributed coordinators and defining their functions. Under this two-layer control architecture, real-time and historical data of the distribution network topology and source loads are collected, and dynamic cluster partitioning is performed. Based on the initialization of the historical optimal solution database, the ADMM distributed optimization algorithm is adopted, aiming to minimize the weighted sum of the square of PV active power reduction, the square of the difference between node voltage and rated voltage, and the square of network loss, to calculate the total power adjustment of each cluster. Based on the total power adjustment of the cluster, voltage deviation factor, and resource margin factor, combined with the model and model parameters, the power allocation of PV reactive power adjustment and energy storage active power adjustment of the equipment under the node is performed. This refined modeling provides a foundation for voltage control, improving the accuracy and reliability of control.

[0118] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0119] The grid voltage stability control method that integrates photovoltaic, energy storage, and load fine modeling and dynamic collaborative control provided in this disclosure is applicable to grid voltage safety control in distributed generation access scenarios, and is especially applicable to grid voltage safety control in high-penetration distributed generation access scenarios.

[0120] like Figure 1 As shown, this is a power grid voltage stability control system in an embodiment of this disclosure, designed as a hierarchical collaborative architecture, including:

[0121] There are at least two distributed coordinators and one centralized coordinator, where one cluster corresponds to one distributed coordinator; the distributed coordinator connects to all nodes within its corresponding cluster and is deployed near the cluster, and the nodes include: photovoltaic equipment, energy storage equipment and load equipment; the centralized coordinator is deployed in the dispatch center;

[0122] The distributed coordinator connects to local photovoltaic, energy storage, and load nodes. It is responsible for collecting and reporting real-time operating status data, preprocessing and utilizing the source-load-storage model to calculate the equipment output capacity and load demand and report it. It strictly adheres to the execution of equipment safety constraint control commands and performs local emergency response.

[0123] The centralized coordinator receives data reported by the distributed coordinators and is responsible for global monitoring and parameter management such as model parameter identification and updating, global control target setting, issuance of adjustment instructions, cross-regional collaboration and emergency response.

[0124] The distributed coordinator collects the following three types of core data from local nodes, as well as local node anomaly information, such as equipment failure and current exceeding limits:

[0125] Photovoltaic node: grid connection point voltage, output current, real-time irradiance, and ambient temperature;

[0126] Energy storage node: charging and discharging current, terminal voltage, state of charge;

[0127] Load node: node voltage, active power, reactive power, and load type identifier.

[0128] The centralized coordinator and the distributed coordinator interact with each other through a communication line, forming a collaborative mechanism of global optimal decision-making and local precise execution, which takes into account both overall planning and real-time response.

[0129] In this context, "deployed near the cluster" means that the physical deployment location of the distributed coordinator meets any of the following conditions: the physical distance from any node in the cluster is less than or equal to 50 meters, or the one-way network transmission latency between the distributed coordinator and any node in the cluster is less than or equal to 1 ms.

[0130] The following sections describe the source-load-storage modeling method based on a two-layer control architecture and the grid voltage method based on a two-layer control architecture.

[0131] The source-load-storage modeling method based on the two-layer control architecture is as follows:

[0132] Each distributed coordinator reports the collected running status data of each node in the corresponding cluster to the centralized coordinator.

[0133] The centralized coordinator uses operational status data to identify and update the model parameters of the pre-built source-load-storage model, and then distributes the updated model parameters to each distributed coordinator.

[0134] Each distributed coordinator receives the updated model parameters and uses them to update the source-load-storage model.

[0135] The source-load-storage model includes: distributed photovoltaic model, load model and energy storage model.

[0136] The following sections will explain the construction of distributed photovoltaic models, load models, and energy storage models, as well as the corresponding parameter identification.

[0137] 1) Distributed photovoltaic model

[0138] 11) Construction of the distributed photovoltaic model:

[0139] For distributed photovoltaic (PV) systems on the load side of urban power grids, a mechanistic model and engineering algorithms are used to establish a distributed PV model and identify its parameters in urban areas.

[0140] Specifically, the reactive power characteristics of photovoltaic (PV) power are characterized by the reactive power generated by distributed PV power sources and the rate of change of these reactive power sources. The details are as follows:

[0141] The photovoltaic model adopts the Single Diode Model (SDM), and the formula for calculating the photocurrent of the photovoltaic array can be expressed as:

[0142] (1)

[0143] in, Represents the photocurrent at time t; This refers to the photocurrent under standard operating conditions. Represents real-time radiance; Represents standard radiance; Indicates the temperature coefficient of current; Indicates ambient temperature; Indicates the standard ambient temperature.

[0144] Active power output of photovoltaic It can be represented as:

[0145] (2)

[0146] in, This represents the DC voltage at which the array operates at its maximum. For inverter conversion efficiency, here It can also be written as This refers to the photovoltaic active power calculated using a distributed photovoltaic model.

[0147] The reactive power generated by distributed photovoltaic (PV) power sources can be expressed as:

[0148] (3)

[0149] in, This represents the reactive power generated by the photovoltaic power source at time t; This represents the actual voltage at the photovoltaic grid connection point at time t; when the grid voltage is lower than 0.9 times the rated value, the photovoltaic system enters the Low Voltage Ride Through (LVRT) mode. This represents the reactive power support factor, which determines the additional reactive power generated per unit voltage drop. Under steady-state conditions, the photovoltaic reactive power is adjusted according to dispatch instructions. This indicates the rated voltage at the grid connection point, under high voltage ride-through conditions; The High Voltage Ride Through (HVRT) active power retention factor determines the retention rate of photovoltaic active power when the voltage rises. When the grid voltage is higher than 1.1 times the rated value, the photovoltaic power source enters HVRT mode. When the grid voltage rises, it limits the sudden decrease of photovoltaic active power or the increase of reactive power generation to prevent the voltage from rising further.

[0150] To ensure that the photovoltaic current does not exceed the equipment's safe operating limit, prevent the photovoltaic system from tripping due to overcurrent, and reduce the impact of the photovoltaic system on the grid frequency and voltage, it is necessary to constrain the photovoltaic power supply current.

[0151] (4)

[0152] in, This represents the output current of the photovoltaic power source at time t; For fault current limits; This refers to the transient response time of the photovoltaic power source. This indicates the maximum permissible change during the transient response period.

[0153] Furthermore, the rated apparent capacity of the inverter in a photovoltaic device can be found on the device's nameplate. .

[0154] 12) Distributed photovoltaic model parameter identification

[0155] The model parameters that need to be identified in distributed photovoltaic (PV) models mainly include two categories: steady-state fundamental parameters and fault ride-through parameters. Steady-state fundamental parameters include the internal resistance of a single diode, while fault ride-through parameters include the reactive power support factor of the LVRT (Low Voltage Ride-Through Rate). HVRT active power retention coefficient Fault current limit With transient response time .

[0156] The process of model parameter identification and updating of distributed photovoltaic models consists of three steps: photovoltaic data preprocessing, identification and parameter updating based on the least squares method with fault weights.

[0157] Photovoltaic data comes from a high-speed power line carrier (HPLC) device on the grid side. Under steady-state conditions, the actual voltage at the photovoltaic grid-connected point can be collected at 15-minute intervals, and under transient conditions, at a sampling frequency of 1Hz. Output current Real-time irradiance and ambient temperature Kalman filtering is used to remove communication interference from the data, retaining 30 minutes of continuous steady-state data to reflect conventional power generation characteristics. Based on the LVRT and HVRT factory test reports provided by the photovoltaic inverter manufacturer, approximately 10 seconds of three-segment wavelength data—voltage surge, stabilization, and recovery—are extracted to identify fault ride-through parameters. For the current limit and series internal resistance values ​​during a fault, initial values ​​are set according to the set multiple of the photovoltaic rated current and the module nameplate values.

[0158] By minimizing the error between the photovoltaic characteristics calculated by the model and the actual collected data using the time-domain least squares method based on fault weights, the accurate parameters can be derived, i.e., the parameters can be identified.

[0159] Specifically, the objective function is set as follows:

[0160] (5)

[0161] in, This indicates the error between the photovoltaic characteristics calculated by the distributed photovoltaic model and the actual collected operating status data; and These are the actual measured photovoltaic active power and the photovoltaic active power calculated using a distributed photovoltaic model, respectively. and S represents the actual measured photovoltaic reactive power and the reactive power calculated using the distributed photovoltaic model; S represents the time interval for steady-state data; F represents the time interval for fault data. and These are the weights for steady-state data and the weights for fault data, respectively.

[0162] During the calculation, the least squares fitting tool built into MATLAB can be used to iterate and adjust the model parameters until the required identification is achieved. Once the minimum value is reached, ordinary edge computing can usually handle its computational load.

[0163] , For fault scenarios, the parameters are updated once when the power grid triggers LVRT or HVRT to ensure that the corresponding parameters are up-to-date for the next fault. Fault current limits are affected by temperature, so the update interval can be extended to once a month; for every 10-degree increase in ambient temperature, [the update interval is adjusted]. Reduce by 5% to ensure that the current does not exceed the actual withstand capacity of the component in the event of a fault.

[0164] Through data processing, parameter identification algorithms, and dynamic updates, the parameters of the distributed photovoltaic model can reflect the steady-state power generation characteristics while reproducing the transient response of fault circuits, thus providing a reliable photovoltaic model for frequency and voltage control.

[0165] 2) Energy storage model

[0166] 21) Construction of energy storage model

[0167] The energy storage model adopts an improved Thevenin model with SOC dynamic characteristics, which requires clarifying the static characteristics, transient characteristics, energy characteristics and safety constraints of the energy storage battery.

[0168] The static composition of the terminal voltage of an energy storage battery can be expressed as:

[0169] (6)

[0170] In equation (6), This represents the terminal voltage of the energy storage system at time t; This represents the open-circuit voltage at time t under SOC (State of Charge) conditions. Let be the charging and discharging current at time t; Let be the internal resistance of the stored energy in the SOC state at time t; This is the polarization voltage.

[0171] The terminal voltage is divided into ideal open-circuit voltage, internal resistance voltage drop, and polarization voltage, and the effects of charging and discharging current, SOC, and internal resistance on the terminal voltage are described respectively.

[0172] The dynamic change of polarization voltage can be expressed as:

[0173] (7)

[0174] in, This represents the polarization resistance at time t under the SOC state; Indicates polarization capacitance; used to represent the transient storage capability of polarization voltage.

[0175] Formula (7) above describes the decay and excitation process of polarization voltage, which is the key to simulating the transient characteristics of energy storage. The first term indicates that the polarization voltage decays naturally through the polarization branch. When there is no current, the polarization voltage gradually disappears. The second term indicates the excitation effect of the charging and discharging current on the polarization voltage. The current will charge the polarization capacitor, thereby increasing the polarization voltage.

[0176] The dynamic calculation process of SOC can be represented as follows:

[0177] (8)

[0178] in, This represents the state of charge at time t, i.e., the proportion of remaining capacity to rated capacity; Represents the initial state of charge (SOC); Indicates the rated capacity of energy storage; express The charging and discharging efficiency at any given time is approximately 0.05 to 0.1 during charging, and can be approximated as lossless during discharging.

[0179] To prevent excessive current from damaging the energy storage, there is an upper limit to the charging and discharging current. and the maximum allowable change in current within a certain time period The constraints are expressed as:

[0180] (9)

[0181] in, Indicates the time interval of energy storage current variation; This is the maximum allowable current. This is the maximum allowable change in current within a specified time period.

[0182] Furthermore, to support stable voltage regulation and active power control in urban power grids, it is necessary to clearly define the voltage and active power capacity boundaries of energy storage. The active power regulation capability of energy storage can be determined through the nameplate parameters of the energy storage converter. .

[0183] The regulation power model for energy storage is as follows:

[0184] (10)

[0185] in, The power value that can be adjusted; This refers to the rated active power regulation capacity of the energy storage. When the SOC of the energy storage is below 20% or above 80%, there is a risk of overcharging or over-discharging, and the energy storage will cease active power regulation. Within the range of 20% to 80%, the active power changes linearly with the SOC. The measured current actual power of the energy storage is... .

[0186] By decomposing static terminal voltage, dynamically calculating transient polarization and SOC, the operating characteristics of energy storage are comprehensively described, providing a refined energy storage model for urban power grid security and control.

[0187] 22) Parameter identification of energy storage models

[0188] The parameter identification and updating of the energy storage model is divided into three parts: energy storage data preprocessing, recursive least squares method and parameter updating.

[0189] The converter in the energy storage module is the core of energy conversion between energy storage and the grid. It can monitor the operating status of the energy storage in real time. The data collected at the energy storage end mainly includes the charging and discharging current of the energy storage. Terminal voltage of energy storage system and energy storage state of charge The sampling frequency is the same as that of photovoltaics, using 1Hz to ensure the real-time dynamic operating characteristics of the energy storage are captured. On the other hand, energy storage systems typically come with a Hybrid Pulse Power Characteristic (HPPC) test report during the manufacturing process. As a standard test in the energy storage industry, HPPC obtains the characteristic parameters of the energy storage at different SOCs through pulsed charging and discharging. The 20%, 50%, and 80% points respectively correspond to... and , , are the initial internal resistance and initial polarization resistance, respectively, which reflect the purely resistive losses and the "polarization effect" within the energy storage system. Preprocessing the obtained data yields initial values ​​for subsequent parameter iterations, accelerating algorithm convergence.

[0190] The core of energy storage parameter identification is to update parameters using recursive least squares and introduce a forgetting factor to highlight the latest data. Since energy storage characteristics dynamically change with state of charge (SOC), charge-discharge cycle count, and temperature, a forgetting factor is introduced to make parameter identification more focused on the current state. The main function of the forgetting factor is to increase the weight of the latest data in the iteration process. The essence of recursive least squares is to iteratively correct the parameters using the prediction error of the new data; the core formulas include the gain matrix formula and the parameter update formula.

[0191] The gain matrix formula can be expressed as:

[0192] (11)

[0193] in, The covariance matrix, which is a 2×2 matrix in this case, reflects the uncertainty of parameter estimation. The initial value is generally set to a large positive definite matrix. The denominator is the forgetting factor, and the denominator is the normalization term to prevent excessive gain from causing unstable parameter updates. The numerator is related to the input vector. The higher the sensitivity of the input vector to the identification parameters, the greater the gain. The input vector can be represented as:

[0194] (12)

[0195] The parameter update formula is:

[0196] (13)

[0197] in, Let k be the parameter vector to be identified. In this embodiment of the disclosure, the identification parameter is selected as a two-dimensional vector. ; express k The parameter vector at time -1; The difference between the actual terminal voltage and the ideal open-circuit voltage is expressed as:

[0198] (14)

[0199] The output of the prediction model, obtained by combining the previous parameters with the input vector, is... Subtracting the actual values ​​from the model predictions yields the error between them. The larger the error, the more the parameters need to be corrected. Multiplying the prediction error by the following gain matrix corrects the previous parameters, resulting in the current parameters.

[0200] The correction of energy storage model parameters is based on daily slow charging and discharging data. Due to the polarization characteristics of energy storage, the polarization resistance is only significant under pulsed current, making it difficult to distinguish the source of the collected data. and Impact. Dynamic HPPC can effectively solve this problem; after the energy storage has been running for about a week, it was found that... When the identification is inaccurate (mainly manifested as a large difference between the calculated polarization voltage and the actual value), the energy storage can be periodically subjected to pulse charging and discharging. After experiencing a rapid impact current of 1.05 times the rated current, the current, voltage, and SOC during the pulse process are collected. This transient data is used for correction and substituted into the parameter update formula (13) to enable the energy storage model to better identify the polarization. and Furthermore, by specifying a power level and measuring the event at which the energy storage reaches its rated power during a pulsed experimental pulse, the response time constant of the energy storage can be determined. Response time constant Used to reflect the delay time from receiving instructions to actual power adjustment in energy storage.

[0201] The active power curve of energy storage can be obtained directly from the factory test report. Before the energy storage leaves the factory, the maximum charge and discharge power is tested at 20%, 30%, 40%, 50%, 60%, 70%, and 80% respectively. The active power curve can be obtained using a linear fitting tool. curve.

[0202] In summary, through the process of initial value - daily correction - periodic testing - cyclical iteration, a relatively accurate energy storage model can be obtained, providing a reliable foundation for energy storage to participate in grid frequency and voltage regulation.

[0203] 3) Load Model

[0204] 31) Construction of the load model:

[0205] To improve the coordinated control of the source-load-storage system, it is necessary to construct load models that are categorized and include static and dynamic fault characteristics, and obtain their core parameters through parameter identification to adapt to the photovoltaic energy storage system and support the subsequent control of grid voltage.

[0206] The load model is constructed using a composite model that integrates the static ZIP, dynamic first-order inertia, and fault constraints of different types.

[0207] Specifically, considering the differences between various types of loads, such as commercial, residential and industrial loads, the three types of loads are labeled with serial numbers m=1, 2 and 3 respectively, and load models covering steady state, transient state and fault state are constructed respectively, that is, composite models of static power characteristic equation, dynamic response equation and fault constraint of load.

[0208] Static power characteristic equation:

[0209] Both active and reactive power of the load exhibit ZIP characteristics (impedance-type Z, current-type I, power-type P) with respect to node voltages. The ratios of these three characteristics satisfy coefficient normalization constraints to accommodate reactive power flow calculations in multi-dimensional coordinated control. Active power static equations and reactive power The static equations are as follows:

[0210] (15)

[0211] in, , These are the rated active power and rated reactive power of the m-th type of load, respectively, both obtained from power grid user files and SCADA historical data statistics; Real-time voltage at the load node; The rated voltage of the power grid at the load connection point; , , Let be the active ZIP factor for the m-th type of load; , , Let m be the reactive power ZIP factor for the m-th type of load. The above factor satisfies:

[0212] (16)

[0213] Dynamic response equation:

[0214] Considering the response delay of load power to voltage changes, a first-order inertial element is introduced to describe its transient process, representing the power ramp-up process when the grid voltage and current change abruptly. Simultaneously adapting to the synchronous clustering requirements in subsequent dynamic cluster partitioning, the active power transient equation and reactive power transient equation can be described by the following equations:

[0215] (17)

[0216] in, , These are the active and reactive transient time constants of the m-th type of load, respectively, and the industrial load's... The largest load is in residential areas, followed by commercial areas, and the smallest load is in residential areas. and These represent the active and reactive reference values ​​for the m-th load, respectively. Under steady-state conditions, they are equivalent to the values ​​calculated by the static ZIP equation, as shown in static equation (15). However, under fault conditions, the increment caused by the fault needs to be considered, as shown in formulas (18) and (19) below:

[0217] (18)

[0218] (19)

[0219] in, This indicates the moment the fault was triggered, which is detected by the power grid's protection system. Indicates the duration of the fault; and For the surge in active power and reactive power during the m-th type of load fault.

[0220] 32) Parameter identification of the load model

[0221] The parameter identification method for the load model involves classifying the load at each node; for each load type, weighted least squares calculations are performed using two errors based on the load data to identify the load model parameters. Specifically, taking commercial, residential, and industrial loads as examples, the parameter identification of the load model consists of four steps: data preprocessing, load type clustering, and hierarchical weighted least squares identification. The specific parameter identification method for the load model is as follows: Figure 2 As shown, it includes:

[0222] Step 201: Collect voltage and power operation data for commercial, residential, and industrial loads in multiple time periods, including steady-state, transient, and fault states;

[0223] First, all load nodes in the target area are screened to determine the types and number of load nodes in the area. The node voltage, active power and reactive power of the nodes are collected for 30 consecutive days, with a 15-minute interval between each collection. The collected data is then put into the steady-state dataset.

[0224] Secondly, fault events are screened. For each event with a voltage change rate greater than 5%, the data is extracted starting 2 seconds before the event occurs. Each event includes 10 seconds of recording voltage, active power, and reactive power values ​​at a sampling frequency of 1Hz. At least 5 typical events are retained for each node.

[0225] In addition, it is necessary to extract user profiles containing user ID, electricity type, average daily electricity consumption and daily load peak-valley ratio, match user profiles with corresponding nodes, and include fault events in the fault dataset.

[0226] The collected voltage and power are plotted as time-series curves, and obvious outliers are removed. Then, Kalman filtering is applied to the data to correct the noise.

[0227] Step 202: Based on the average daily electricity consumption duration, peak-to-valley ratio, and reactive power-voltage sensitivity, the K-means clustering algorithm is used to classify the load type of the nodes;

[0228] This will include the average daily electricity consumption time of each node. Peak-to-valley ratio The dataset of electricity usage types is defined as an auxiliary feature set.

[0229] Extract from auxiliary feature set and The first two dimensions of clustering are used as the basis features. A linear regression method is used to calculate the slope of reactive power versus voltage for the steady-state data of each node, and the mean of all groups is taken to obtain the reactive power-voltage sensitivity of each node. This serves as the third dimension of clustering criteria.

[0230] K-means clustering is used to classify the nodes, with the number of clusters set to 3. Cluster centers are manually initialized based on the load characteristics of commercial, residential, and industrial loads, which can be set to [12, 3.5, 0.8], [7, 2.5, 0.5], and [20, 1.5, 0.3] respectively. The cluster centers are iteratively updated using the K-means clustering algorithm until the node affiliation no longer changes, and the cluster labels 1, 2, and 3 for each node are output.

[0231] The clustering results are compared with the electricity usage types in the archives to ensure an accuracy rate of over 95%, and exceptions are manually classified.

[0232] Step 203: Construct a multi-type load composite model consisting of a static ZIP model, a dynamic first-order inertial element, and a fault constraint element;

[0233] Step 204: Set weights according to the degree of impact of load on the distribution network, and identify model parameters using the hierarchical weighted least squares method.

[0234] The hierarchical weighted least squares method minimizes the power error values ​​for various load types. Power errors include the error between the calculated active power and the measured value of the load during normal operation, and the error between the calculated reactive power and the measured reactive power of the load during fault conditions. Specifically, it minimizes the value of the following objective function:

[0235] (20)

[0236] in, , These are the calculated values ​​of active power and reactive power of the load obtained from static and dynamic equations and fault reference values, that is, the calculated values ​​of active power and reactive power of the load calculated based on the load model. , These are the active power measurements of the load during normal operation and the reactive power measurements of the load during a fault. Weighting coefficients are defined based on the impact of different types of loads on the power grid. For example, commercial loads have the greatest impact on the grid voltage, so they have the highest weight to prioritize ensuring the accuracy of their parameter identification. Finally, constraints are set based on engineering experience.

[0237]

[0238] (twenty one)

[0239] Static parameters for three types of loads were obtained through parameter identification. Dynamic parameters With fault parameters Finally, the static parameters of the regional nodes are updated based on the voltage sensitivity change rate being greater than a set value (e.g., 10%), and the dynamic parameters are updated based on the load type proportion change rate being greater than a set value (e.g., 10%). The fault parameters are updated after each fault event in the power grid.

[0240] Here, this disclosure will connect the verified categorized load model to the distributed coordinator.

[0241] Based on the source-load-storage model method described above, this disclosure also provides a grid voltage control method. This voltage control method is also based on... Figure 1 The voltage control method for a two-layer control architecture power grid includes the following steps: Figure 3 As shown:

[0242] Step 301: The centralized coordinator divides the network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type;

[0243] Step 302: The distributed coordinator reports the collected running status data of each node in the corresponding cluster to the centralized coordinator.

[0244] Step 303: Based on the overall network distribution topology and operating status data, the centralized coordinator determines the global control target of each cluster and issues adjustment instructions to the corresponding distributed coordinators. The global control target includes the total power adjustment amount of the cluster.

[0245] Step 304: Based on the received adjustment instructions and combined with the source-load-storage model, the distributed coordinator determines the power allocated to the devices connected to the nodes; it verifies the allocation results through the source-load-storage model to ensure compliance with the constraints in the source-load-storage model, and generates control instructions for each device.

[0246] The dynamic cluster partitioning method in step 301 above is as follows: Figure 4 As shown, it includes the following steps:

[0247] Step 401: The centralized coordinator determines the distribution network topology, node power, and equivalent impedance between nodes;

[0248] Specifically, the centralized coordinator collects basic data from the urban power grid, organizes and inputs data related to the grid topology, line impedance, photovoltaics, energy storage, and real-time load power, acquires historical data at the 7-day 15-minute level, and obtains node data through the distributed coordinator. i-j Equivalent impedance between .

[0249] Step 402: The centralized coordinator determines the net power correlation coefficient between nodes based on the net power of the nodes, and corrects the net power correlation coefficient between nodes based on whether the loads between the nodes are of the same type.

[0250] Specifically, the time-series correlation coefficient is calculated using the weighted Pearson correlation coefficient.

[0251] All correlation calculations can be based on time-series data at a set level, such as 7-day 15-minute time-series data. The basic form of the Pearson correlation coefficient is:

[0252] (twenty two)

[0253] The net power of each node, measured by the distributed coordinator, is:

[0254] (twenty three)

[0255] in, Represents a node i Net power at time t; Represents a node i Power at time t; Represents a node i Energy storage power at time t.

[0256] Then node i and j The net power correlation can be expressed as:

[0257] (twenty four)

[0258] in, Represents a node i Net power average over 7 days; Relationship coefficient The closer the correlation coefficient is to 1, the more synchronized the net power fluctuations of the two nodes are. After obtaining the net power correlation coefficient between the two nodes, differences in load type can easily lead to spurious correlations, requiring the addition of a correction factor. Corrected net power correlation coefficient The calculation method is shown in formula (25), and the correction coefficient is... The values ​​of are shown in Table (1) below:

[0259] (25)

[0260] Table (1)

[0261]

[0262] Step 403: The centralized coordinator determines the weighted electrical distance between nodes based on the equivalent impedance between nodes and the corrected net power correlation coefficient between nodes;

[0263] The centralized coordinator uses a normalized weighting method to normalize the correlation coefficients and the equivalent impedances between nodes to the same order of magnitude, and then performs weighted summation:

[0264] (26)

[0265] in, Weighted electrical distance; For nodes i-j The normalized value of the equivalent impedance, ranging from 0 to 1; This is a weighted factor based on collaboration, meaning that stronger collaboration between nodes results in shorter distances. Let be the weighting coefficient, satisfying .

[0266] The equivalent impedance between nodes is the original electrical distance. In this embodiment of the disclosure, the improved electrical distance calculation can ensure that the photovoltaic output fluctuation, energy storage regulation capability, and load power demand of the same cluster are correlated, so that the subsequently selected dominant node has the advantages of a wide voltage influence range and a lot of adjustable resources, and can efficiently execute global control commands.

[0267] Step 404: The centralized coordinator uses an improved K-means clustering algorithm to determine the clusters in the entire distribution network based on the load percentage data of different types, the weighted electrical distance between nodes, and the amount of voltage regulation resources at each node.

[0268] Specifically, the first step is to determine the number of clusters K in the entire model. Calculate the sum of squared intra-cluster errors corresponding to different K values. Use the elbow rule to select the K value corresponding to the inflection point of the decreasing sum of squared intra-cluster errors. If the industrial load in a certain geographical area accounts for ≥40%, then K=K 基础 +1, because industrial nodes have special electrical characteristics and require independent clusters. Urban power grids typically have K=4~6, but the actual value depends on the situation.

[0269] Step 2: Using the calculated electrical distance Construct a similarity matrix.

[0270] Assuming there are N nodes, construct an N×N dimensional matrix with zeros on the diagonal and the rest represented as follows:

[0271] (27)

[0272] Step 3: Select K nodes with sufficient voltage regulation resources and dispersed electrical distances. As the initial centroids for clustering, it is essential to ensure that the centroid nodes possess controllability. In practice, this can be referenced to the average net output load of each node and the remaining controllable capacity of the energy storage. The specific selection process first selects candidate nodes that meet the criteria, and the weighted electrical distance between them is considered. The initial centroids should be as large as possible to avoid overlapping. Then, select the K nodes that best meet the above objective from all nodes that meet the conditions as the initial centroids.

[0273] Step 4: For all N nodes within the control range, calculate their distance to the initial centroid node. Find the initial centroid corresponding to the minimum distance, and use the K-means clustering algorithm to cluster the nodes within the range to obtain K clusters. If a node is equidistant from two or more centroids, it is classified as a net power similarity cluster. Clusters containing smaller particles.

[0274] Step 5: Calculate the electrical center index within the cluster for each cluster. k within Calculate each node, and perform calculations on each node. :

[0275] (28)

[0276] Represents a node To cluster k The sum of the electrical distances of all other nodes within the cluster; k Inside, select If there are two or more smallest nodes, the control capabilities of the two nodes are compared, and the node with the stronger control capability is selected as the new centroid, resulting in K new centroids. Calculate each centroid With the new center of mass Weighted electrical distance :

[0277] (29)

[0278] The iteration ends when the convergence condition is met or the specified number of convergences is reached, resulting in K centroids and K clusters. After convergence, the coupling degree, cluster size, and load type consistency within each cluster are verified to ensure the clusters are reasonable and to avoid fluctuations in the clustering results.

[0279] Furthermore, within the clustered clusters, dominant nodes are identified. Based on weighted electrical distance centrality and voltage regulation resource margin indicators, the selection of overall dominant nodes is refined to ensure that the dominant nodes have broad voltage influence, sufficient resource margin, synchronized net power, and suitable electrical locations. The dominant node is selected based on the comprehensive weighted score of multiple indicators.

[0280] (30)

[0281] in The weights assigned to each indicator are set by the centralized coordinator. It represents the voltage regulation resource margin, which measures the remaining regulation capacity of photovoltaic and energy storage nodes. It includes two sub-items: photovoltaic reactive power margin and energy storage active power margin, which are obtained based on the proportion of the actual power output of photovoltaic and energy storage in the controllable capacity. It reflects the coordination of power fluctuations between a node and other nodes in the cluster, and is the mean of the Pearson correlation coefficient with other nodes calculated based on net power data; The weighted electrical distance centrality of a node within the cluster is evaluated by the average distance from the node to other nodes in the cluster.

[0282] Step 303 above specifically includes: the centralized coordinator initializes the ADMM algorithm based on the overall network distribution topology, received operating status data and current scenario data, using a pre-established ADMM historical optimal solution database, determines the global control target of each cluster and issues adjustment instructions to the corresponding distributed coordinators.

[0283] Step 303 above more specifically includes, for example: Figure 5 The following steps are shown:

[0284] Step 501: The centralized coordinator receives the current scenario data from the distributed coordinator based on the collected running status data of the corresponding cluster;

[0285] The current scenario data includes: the penetration rate of distributed photovoltaics within the cluster, the load type of the cluster, and the type of disturbance experienced by the cluster;

[0286] Step 502: The centralized coordinator constructs an ADMM historical optimal solution database containing the correspondence between each historical scenario and the initial value;

[0287] The ADMM (Alternating Direction Method of Multipliers) historical optimal solution database contains the correspondence between each historical scenario and the initial value. The historical scenario includes three features: penetration rate, load type, and disturbance type. The initial values ​​include: initial values ​​of Lagrange multipliers, initial control quantities of photovoltaic reactive power and initial control quantities of energy storage active power. The penetration rate is the ratio of total photovoltaic active power to total active power consumption within the cluster.

[0288] Step 503: For each cluster, the centralized coordinator calculates the similarity between the current scene of the cluster and each historical scene; and selects the initial value for the cluster based on the similarity value.

[0289] Step 504: Substitute the initial values ​​and operating status data of each selected cluster into the global objective function, and use the ADMM algorithm for optimization to obtain the total active power adjustment and total reactive power adjustment for each cluster.

[0290] Step 505: Take the total active power adjustment and total reactive power adjustment of each cluster as the global control target of each cluster and issue adjustment instructions to the corresponding distributed coordinator.

[0291] Step 506: Based on the overall network distribution topology, received operational status data, and current scenario data, the centralized coordinator initializes the ADMM algorithm using the pre-established ADMM historical best solution database, determines the global control objective of each cluster, and issues adjustment instructions to the corresponding distributed coordinators.

[0292] The above process is that the centralized coordinator, based on the overall network topology and received operational status data, uses a pre-established ADMM historical optimal solution database to determine the global control objectives of each cluster and issue adjustment instructions to the corresponding distributed coordinators.

[0293] The ADMM distributed optimization method based on iterative initialization of historical best solutions is explained in detail below:

[0294] ADMM is an effective tool for solving distributed optimization problems. However, its convergence speed is heavily dependent on the selection of initial values. Traditional startup methods (setting all initial values ​​to 0) or random initialization methods often require a large number of iterations to converge when dealing with the complex and variable operating conditions of the power grid under high-penetration photovoltaic access, making it difficult to meet the stringent computational speed requirements of real-time control. This disclosure proposes an ADMM initialization method based on a historical optimal solution database. By constructing a historical optimization record library categorized by typical operating conditions and performing rapid scenario matching during real-time optimization, a set of high-quality initial values ​​is provided to the ADMM solver, thereby significantly reducing the number of iterations.

[0295] (1) ADMM iterative solution

[0296] The real-time total load within the cluster, collected by the distributed coordinator, can be expressed as:

[0297] (31)

[0298] in, This represents the load power on all nodes within cluster a; Load power at each node; The total number of nodes in the cluster; the control density per unit load can be expressed as:

[0299] (32)

[0300] in, These represent the unit photovoltaic reactive power density and the unit energy storage active power density, respectively, and represent the photovoltaic reactive power and energy storage active power required to be allocated per unit load. , Clusters a The adjustment amounts for total photovoltaic reactive power and total energy storage active power; the control variables are set as follows. By replacing absolute values ​​with density values, a unified collaborative benchmark can be achieved across clusters.

[0301] The centralized coordinator controls each cluster individually. To optimize the objective, for each successful ADMM optimization calculation, the initial optimization values ​​are recorded, which are also the controllable variables in the ADMM solution: the initial values ​​of the Lagrange multipliers. Initial control quantity of photovoltaic reactive power and initial control quantity of active power for energy storage It also records the number of optimization iterations required to reach the iteration termination condition each time. Each cluster has its own optimization objective, categorized by cluster size. a objective function For example:

[0302] (33)

[0303] in, For photovoltaic reduction; For network loss; Represents a cluster a The Middle i The voltage of each node; Indicates the node reference voltage; This is the second term of the voltage deviation; For different target items in the cluster a The weights in the objective function are defined by the centralized coordinator.

[0304] The above describes an optimization problem within a cluster. A global optimization problem, however, requires consideration of coordination between different clusters and the synergy of boundary conditions. Therefore, a method based on... The ADMM iterative formula ensures power grid stability, and a global objective function is established:

[0305] (34)

[0306] (35)

[0307] in, For clusters a and b The objective function between; L The objective function is the global objective function. K Number of clusters; For clusters a The objective function; For clusters b The objective function; For penalty parameters; These are the initial Lagrange multipliers; and Clusters a , b Control variables and constraints This allows the control variables among clusters to satisfy the collaborative consistency deviation, thereby achieving a balance between distributed optimization and global constraints and ensuring power grid stability.

[0308] Subsequently, updates were made during the iterative process. and The values ​​are used to obtain the optimal control variables for each cluster, and the Lagrange multipliers are updated based on... a, b Density deviation adjustment guides the next iteration to reduce the difference:

[0309] (36)

[0310] In the k~k+1 iterations, repeat the steps of the first iteration, using the results from the previous iteration. To replace the next one This continues until the convergence condition is met. The convergence condition is:

[0311] (37)

[0312] Where K represents a different cluster number; the above a, b These are all different cluster numbers; k Indicates the first k This iteration uses the optimal control variables obtained for each region. This is the power allocation density value. By comparing it with the total load within the cluster, the optimal allocation values ​​of active and reactive power within each cluster can be obtained. .

[0313] (2) Construction of historical optimal solution database

[0314] Based on the ADMM optimal solution selection rules, the global objective function value obtained from multiple iterations of ADMM is compared with the number of iterations. Under the premise of satisfying convergence accuracy, the number of iterations obtained from different iterations is compared. With global objective function value Initialization data with fewer iterations and smaller global objective function values ​​within each cluster is retained and recorded in the database. Historical data in the database is periodically evaluated to ensure that the optimal data quality of the database is continuously optimized over time. The update conditions should meet the following conditions:

[0315] (38)

[0316] in, , These represent the convergence count and the global objective function value of the optimal solution for this type of scenario within the cluster, respectively.

[0317] Then, considering that the power allocation varies for different types of clusters, a multi-dimensional operating condition system is constructed for the multiple clusters, including penetration rate, load type and disturbance type.

[0318] Among them, the penetration rate of distributed power sources will be classified. Discretized into multiple intervals, the penetration level is based on the ratio of total photovoltaic active power to total load active power within the cluster, with low penetration ( <15%), medium penetration rate (15%≤) <25%), high penetration rate ( ≥25%); Load types are further subdivided according to the load type of the dominant node, and can be divided into commercial, residential and industrial loads. Disturbance classification is also refined, including disturbance types such as no disturbance, photovoltaic slump, load surge, and DC blocking.

[0319] A scenario is constructed by combining specific values ​​from the three dimensions mentioned above. For example, low penetration rate, industrial load, and sudden drop in photovoltaic power constitute a scenario.

[0320] (3) Extraction of optimal initial values ​​for each class

[0321] First, the real-time penetration rate of each cluster is determined in real time based on the real-time data collected by the distributed coordinator. In load modeling, the load type determines the node type to which a node belongs. Then, a pre-defined disturbance detection and classification method is used to determine the type of disturbance experienced by each cluster. Real-time penetration rate within a single cluster is also considered. The calculation method is as follows:

[0322] (39)

[0323] in, This represents the total active power of photovoltaic systems within the cluster. This represents the total active power consumption within the cluster.

[0324] Secondly, multi-factor weighted scenario similarity matching is performed, and a similarity index Sim is defined. Combining grid penetration rate and disturbance type, the similarity index Sim should satisfy:

[0325] (40)

[0326] in, Weights for the impact of penetration rate and disturbance type on optimization; For historical scene penetration rate; The degree of difference between the actual node and the load type in the database; This is a perturbation type matching function.

[0327] Finally, the optimal initial value for a class can be selected based on the Sim value. The historical scenario with the highest Sim value can be chosen to extract the optimal initial value for the class, which should satisfy the following:

[0328] (41)

[0329] in, These are the class-optimal initial values ​​learned within this cluster for this scenario. When iterating during ADMM, these class-optimal initial values ​​are selected to replace the Lagrange multiplier initial values. Initial control quantity of photovoltaic reactive power and initial control quantity of active power for energy storage Since the optimal initial value of the class is close to the global optimum, the iteration can converge quickly.

[0330] Step 304 above specifically includes, for example: Figure 6 The following steps are shown:

[0331] Step 601: The distributed coordinator obtains the total active power adjustment and total reactive power adjustment of the cluster from the adjustment command;

[0332] Step 602: The distributed coordinator uses the source-load-storage model and the local node operating status to determine the voltage deviation factor and resource margin factor of each node;

[0333] Step 603: The distributed coordinator decomposes the total active power adjustment and total reactive power adjustment of the cluster according to the voltage deviation factor and resource margin factor of each node, and obtains the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value of each node.

[0334] Step 604: The distributed coordinator uses the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value of each node, as well as the established source-load-storage model and model parameters, to determine the power allocated to the devices connected to the nodes.

[0335] The above process will be explained in detail below:

[0336] First, the distributed coordinator, based on the total active power adjustment and total reactive power adjustment of the cluster, uses a voltage deviation-resource margin dual-factor composite weighted power allocation method to decompose the total active power adjustment and total reactive power adjustment of the cluster, obtaining the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value for each node. Addressing the shortcomings of traditional power allocation schemes that rely solely on voltage deviation, this disclosure proposes a dual-factor composite weighted allocation method that integrates voltage deviation and resource margin. Based on the ADMM optimization results (total cluster adjustment) and the adjustable resource parameters of source, load, and storage for each cluster, the method integrates voltage deviation and resource margin to ensure that the allocation of node resources conforms to equipment safety constraints. The voltage deviation factor quantifies the urgency of node adjustments, and the resource margin factor quantifies the adjustment capability of nodes.

[0337] (1) Voltage deviation and resource margin dual-factor index

[0338] Voltage deviation factor characterizes the node j The greater the deviation of the voltage from the rated value, the higher the urgency of adjustment, and the higher the weight should be allocated. The calculation formula is:

[0339] (42)

[0340] in, For nodes j Voltage deviation factor; For nodes j Real-time voltage; Node reference voltage.

[0341] To facilitate proportional allocation, the voltage deviation factor of each node in the cluster is standardized using the following formula:

[0342] (43)

[0343] in, This represents the total number of nodes in the cluster. For nodes i Voltage deviation factor; For standardized nodes j Voltage deviation factor, The larger the value, the more likely it is to be a node. j The more severe the voltage problem, the higher the weighting percentage should be.

[0344] Resource margin factor is used to represent nodes j Adjustment capability, characterizing nodes jThe remaining adjustable capacity of the controllable devices connected to the node, such as photovoltaic inverters and energy storage systems. In this disclosure, the photovoltaic reactive power margin and energy storage active power margin of each device connected to the node are integrated to obtain the calculation formulas for the reactive power reserve margin of the node as shown in formula (44) and the calculation formulas for the active power reserve margin of the node as shown in formula (45), as follows:

[0345] (44)

[0346] (45)

[0347] in, Represents a node j Above-ground reactive power reserve margin; This indicates that node j has an active power reserve margin; Indicates at node j The total number of photovoltaic devices connected to the above. For the first The rated apparent capacity of the inverter of a photovoltaic system can be obtained from the equipment nameplate. For the first The current active power output of the photovoltaic equipment; Representative node j The total number of energy storage devices connected to it. For the first The rated active power regulation capacity of the energy storage unit; the coefficient 0.3 is the limit value of the energy storage power. For the first The current actual power of the energy storage system (discharge is positive, charging is negative); the active power regulation range when the energy storage SOC is between 20% and 80%, but when the SOC is less than 20% or greater than 80%. .

[0348] If each node contains both photovoltaic and energy storage components, then the standardized active and reactive resource margin factors of the node can be expressed as:

[0349] (46)

[0350] (47)

[0351] in, Represents the standardized node j The active resource margin factor, Represents the standardized node j The reactive power resource margin factor; the larger the two resource margin factors, the more likely the node is to be affected. j The greater the adjustment potential, the more adjustment tasks it should undertake.

[0352] (2) Determine the power of the two-factor weighted allocation

[0353] The distributed coordinator calculates the active and reactive power allocation of each node based on the ADMM optimization results (total reactive and active power adjustments required to stabilize cluster voltage) of the centralized coordinator, combined with the dual-factor indicators of voltage deviation and resource margin.

[0354] With cluster a For example, the allocation of reactive power adjustment for photovoltaic systems at each node:

[0355] (48)

[0356] Allocation of active power adjustment for energy storage at each node:

[0357] (49)

[0358] in, Let J be the photovoltaic reactive power adjustment value at node j. This represents the active power adjustment of energy storage at node j. , The weight of the voltage deviation factor is set by the distributed coordinator based on the voltage state; , This represents the total active and reactive power adjustment within cluster a.

[0359] Next, the distributed coordinator uses the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value of each node, as well as the established source-load-storage model and parameters, to determine the power allocated to the devices connected to the nodes.

[0360] Specifically as follows:

[0361] Verification and control of power allocation based on the source-load-storage model:

[0362] The distributed coordinator, based on the active and reactive power allocation of each node and combined with the source-load-storage model, completes device-level power allocation and constraint verification to ensure the rationality of instructions and generate control commands to issue to the devices. Specifically, this includes the following processes:

[0363] (1) Model parameter call

[0364] The distributed coordinator retrieves the current node's real-time source-load-storage parameters and operational data from the local model database, including rated apparent capacity. Current active power output measurement value LVRT coefficient HVRT coefficient Fault current limit , and transient response time Dynamic calculation of energy storage deployment Rated capacity of energy storage Charge and discharge efficiency Current limit Maximum allowable variation in current Energy storage regulation power The load requires the load type label m, the load's active and reactive power, and the static ZIP factor. , , , , , The active and reactive transient time constants of the m-th type of load , .

[0365] (2) Calculate the power distribution of each device.

[0366] Similar to the resource margin mentioned above, the photovoltaic reactive power resource margin factor within the node is defined based on the established model and model parameters. Quantify the adjustment capability of each photovoltaic device, then the first node j... The margin factor for a photovoltaic system can be calculated as follows:

[0367] (50)

[0368] in, Indicates the first The reactive power resource margin factor of the photovoltaic equipment.

[0369] Assigned to the reactive power adjustment of Taiwan solar power It can be calculated as follows:

[0370] (51)

[0371] The more ample the equipment margin, the more adjustment amount will be allocated. The allocation method for energy storage is similar. After calculation, model verification is required according to the following steps.

[0372] Similar to photovoltaics, the active power resource margin factor for energy storage within a node is defined. :

[0373] (52)

[0374] The final result is Taiwan's active power adjustment :

[0375] (53)

[0376] The power allocation in this embodiment of the disclosure involves three allocations: the first allocation is that the centralized coordinator determines the total power adjustment amount of each cluster based on the overall network topology and the received operating status data, and sends it to the corresponding distributed coordinator; the second allocation is that the distributed coordinator allocates the total power adjustment amount of the corresponding cluster to each node in the cluster; and the third allocation is that the distributed coordinator allocates the power adjustment amount of the node to each device connected to the node.

[0377] Specifically, the first allocation is performed using the ADMM distributed optimization algorithm, which is iteratively initialized with the historical best solution.

[0378] Based on the history in the second allocation, the active power and reactive power reserves of all devices contained in the distributed coordinator computing node are obtained to obtain the active power and reactive power reserves of the node. Formulas (46) and (47) above consider the margin of each node in the cluster in the voltage difference and margin ratio in the cluster, so as to allocate active power and reactive power to each node.

[0379] The third allocation mainly considers the proportion of each device's margin in the total margin of the node, and allocates the corresponding active and reactive power to each device in combination with model parameters. Essentially, it considers the proportion of individual margin in the collective margin, and realizes the layer-by-layer allocation of cluster-node-device.

[0380] (3) Equipment model power verification

[0381] Based on the characteristics of the source-load-storage model, the allocation values ​​at the device level are verified one by one to ensure compliance with the model's safety constraints.

[0382] Photovoltaic equipment verification includes fault ride-through constraints and current safety constraints. If the grid is in LVRT mode, it must be ensured that the photovoltaic reactive power adjustment is not lower than the minimum value required by the LVRT reactive power support coefficient; if it is in HVRT mode, it must be ensured that the photovoltaic active power output is not lower than the minimum value required by the HVRT active power retention coefficient. Secondly, the current safety constraint calculation includes active and reactive components. The total photovoltaic current must be less than the fault current limit, and the current change between adjacent times must not exceed the maximum allowable rate of change.

[0383] Energy storage device verification includes dynamic SOC constraints and current safety constraints. Dynamic SOC constraints predict that the SOC 5 seconds after the active power adjustment is allocated must remain within the range of 20% to 80% to avoid overcharging or over-discharging. Current safety constraints calculate the energy storage charging and discharging current, which must be less than the current limit, and the current change between adjacent times must not exceed the maximum allowable rate of change.

[0384] Load equipment verification includes dynamic transient constraints and fault state constraints. The rate of change of load power must not exceed a limit to avoid aggravating voltage fluctuations due to sudden power changes. Secondly, fault state constraints require that if the power grid is in a fault condition, the load power adjustment must not exceed the sudden increase in fault power in the model to prevent load instability.

[0385] If the above constraints are not met or the adjustable capacity is insufficient, the distributed coordinator will count the total reactive power gap and active power gap within the node, and prioritize calling the equipment with the largest adjustable margin within the same node to fill the gap. The reactive power gap will be prioritized to equipment with sufficient photovoltaic reactive power margin, and the active power gap will be prioritized to energy storage with margin or industrial loads with a large adjustable range.

[0386] (4) Generation of equipment control commands

[0387] The distributed coordinator converts verified device-level adjustments into control commands conforming to the device interface protocol. These commands must carry model constraint parameters. Commands delivered to each photovoltaic device include the final reactive power adjustment and fault mode identifier; energy storage commands include the final active power adjustment; and load commands include both final active and reactive power adjustments. If the distributed coordinator detects insufficient or unexecutable scheduling resources within the cluster, such as due to device failure, it immediately reports back to the centralized coordinator, initiating a secondary allocation of the remaining adjustments to ensure the cluster voltage stability target is achieved.

[0388] Based on the same inventive concept as the above-mentioned source-load-storage modeling method, this disclosure also provides a source-load-storage modeling system, based on a two-layer control architecture, which includes: at least two distributed coordinators and a centralized coordinator, wherein one cluster corresponds to one distributed coordinator; the distributed coordinator is connected to all nodes in its corresponding cluster, and the nodes include: photovoltaic, energy storage and load; the centralized coordinator is deployed in the dispatch center;

[0389] Each distributed coordinator is used to report the collected operating status data of each node in the corresponding cluster to the centralized coordinator; and to receive model parameters and update the source-load-storage model using the model parameters; wherein, the source-load-storage model includes: distributed photovoltaic model, load model and energy storage model;

[0390] The centralized coordinator is used to identify and update the model parameters of the pre-built source-load-storage model using runtime status data, and then distributes the updated model parameters to each distributed coordinator.

[0391] The construction of a distributed photovoltaic model includes: characterizing the reactive power characteristics of photovoltaics;

[0392] The distributed photovoltaic (PV) model parameters include: steady-state parameters and fault ride-through parameters. The steady-state parameters include the internal resistance of a single diode; the fault ride-through parameters include the reactive power support factor in LVRT mode. Active power retention coefficient under HVRT mode Fault current limit With transient response time ;

[0393] The identification of distributed photovoltaic (PV) model parameters includes: minimizing the error between the PV characteristics calculated by the distributed PV model and the actual collected operating status data by using the least squares method based on fault weights, thereby identifying the PV model parameters.

[0394] Based on the same inventive concept as the above-mentioned power grid voltage control method, this disclosure also provides a corresponding power grid voltage control system, including:

[0395] A centralized coordinator is used to divide the distribution network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type. Based on the overall distribution network topology and received operating status data, it determines the global control target of each cluster and issues adjustment instructions to the corresponding distributed coordinator. The global control target includes the total power adjustment of the cluster.

[0396] The distributed coordinator is used to report the collected operating status data of each node in the corresponding cluster to the centralized coordinator; based on the received adjustment instructions, combined with the source-load-storage model, it determines the power allocated to the devices connected to the nodes; the allocation results are verified through the source-load-storage model to ensure compliance with the constraints in the source-load-storage model, and control instructions for each device are generated.

[0397] In the scheme of this embodiment, by using source-load-storage classification, full-characteristic refined modeling, and differentiated parameter identification strategies, the deviation between the model calculation values ​​and the actual operating characteristics of the equipment is significantly reduced, providing precise node characteristic support for voltage stability control. The ADMM algorithm is initialized based on historical optimal solutions, which reduces the number of optimization convergence times, meets the needs of real-time control of the distribution network, and accelerates the control response speed. Dynamic cluster partitioning ensures the accuracy of the control objects, and the dual-factor power allocation takes into account both the urgency and adjustment capacity of the adjustment, avoiding resource waste or insufficient control. At the same time, it strictly adheres to equipment safety constraints, extends equipment life, and achieves optimized resource utilization efficiency. The full-process closed-loop control covers all stages of "modeling-decision-execution-feedback", which can effectively cope with voltage fluctuations caused by the high penetration rate of new energy access, reduce the risk of voltage instability, and enhance grid stability, thereby improving the safe operation level of the urban power grid and the capacity for new energy absorption.

[0398] Although the present disclosure 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 of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A source-load-storage modeling method, characterized in that, Based on a two-layer control architecture, the two-layer control architecture includes: at least two distributed coordinators and one centralized coordinator, wherein one cluster corresponds to one distributed coordinator; the distributed coordinator connects to all nodes within its corresponding cluster, and the nodes include: photovoltaic, energy storage and load; the centralized coordinator is deployed in the dispatch center; Each distributed coordinator reports the collected running status data of each node in the corresponding cluster to the centralized coordinator. The centralized coordinator uses operational status data to identify and update the model parameters of the pre-built source-load-storage model, and then distributes the updated model parameters to each distributed coordinator. Each distributed coordinator receives model parameters and uses them to update the source-load-storage model. The source-load-storage model includes: distributed photovoltaic model, load model and energy storage model; The construction of a distributed photovoltaic model includes: characterizing the reactive power characteristics of photovoltaics; The parameters of the distributed photovoltaic model include: steady-state parameters and fault ride-through parameters. The basic steady-state parameters include the internal resistance of a single diode; the fault ride-through parameters include the reactive power support factor in LVRT mode, the active power retention factor in HVRT mode, the fault current limit and transient response time. The identification of distributed photovoltaic model parameters includes: minimizing the error between the photovoltaic characteristics calculated by the distributed photovoltaic model and the actual collected operating status data by using the least squares method based on fault weights, thereby identifying the photovoltaic model parameters; The identification of distributed photovoltaic (PV) model parameters specifically includes: using the following objective function to identify the PV model parameters: in, This represents the error between the photovoltaic characteristics calculated by the distributed photovoltaic model and the actual collected operating status data; t represents time. and These are the actual measured photovoltaic active power and the photovoltaic active power calculated using a distributed photovoltaic model, respectively. and S represents the actual measured photovoltaic reactive power and the reactive power calculated using the distributed photovoltaic model; S represents the time interval for steady-state data; F represents the time interval for fault data. and These are the weights for steady-state data and fault data, respectively; when the error When the minimum value is reached, the parameters of the distributed photovoltaic model achieve the optimal identification value.

2. The method according to claim 1, characterized in that, The reactive power characteristics of photovoltaic (PV) power generation are characterized by the reactive power generated by distributed photovoltaic (PV) power sources and the rate of change constraint; the reactive power generated by distributed PV power sources is: in, This represents the reactive power generated by the photovoltaic power source at time t; Indicates the reactive power support coefficient; Indicates the rated voltage at the grid connection point; This represents the actual voltage at the photovoltaic grid connection point at time t; This represents the active power retention coefficient under HVRT mode; The current constraint of photovoltaic power sources is: in, This represents the output current of the photovoltaic power source at time t; For fault current limits; This refers to the transient response time of the photovoltaic power source. This indicates the maximum permissible change during the transient response period.

3. The method according to claim 1, characterized in that, The construction of the energy storage model includes: adopting an improved Thevenin model with dynamic SOC characteristics, constructing a model that includes the static characteristics of the energy storage battery terminal voltage, the dynamic change process of the polarization voltage, and the dynamic calculation of SOC. The dynamic change process of the polarization voltage is achieved by characterizing the decay and excitation process, and the dynamic calculation of SOC is determined by combining the charge and discharge efficiency and the rated capacity. The energy storage model parameters include: the polarization resistance at time t under the SOC state and the internal resistance of the stored energy at time t under the SOC state; The identification of energy storage model parameters includes: calculating using the recursive least squares method, introducing a forgetting factor to highlight the latest energy storage data, iteratively correcting the energy storage model parameters using the prediction error of the latest energy storage data, and periodically correcting the polarization resistance through HPPC testing.

4. The method according to claim 3, characterized in that, Static characteristics of energy storage battery terminal voltage: ； in, This represents the terminal voltage of the energy storage system at time t; This represents the open-circuit voltage at time t under SOC (State of Charge) conditions. Let be the charging and discharging current at time t; Let be the internal resistance of the stored energy in the SOC state at time t; Polarization voltage; The dynamic change process of polarization voltage is represented as follows: in, This represents the polarization resistance at time t under the SOC state; Indicates polarization capacitance; The dynamic calculation of SOC is represented as follows: in, This represents the state of charge at time t; Represents the initial state of charge (SOC); Indicates the rated capacity of energy storage; express The charging and discharging efficiency at any given time; The energy storage charge and discharge current constraint is expressed as: in, Indicates the time interval of energy storage current variation; This is the maximum allowable current. The maximum allowable change in current within a specified time period; The regulation power model for energy storage is as follows: in, The power value that can be adjusted; The active power rated for energy storage.

5. The method according to claim 1, characterized in that, The construction of the load model includes: for each type of load, constructing a composite model of the load's static power characteristic equation, dynamic response equation, and fault constraints, and introducing a first-order inertial element into the dynamic response equation to describe its transient process, representing the power ramp-up process when the grid voltage and current change abruptly. The load model parameters include: static parameters: the active ZIP coefficient and reactive ZIP coefficient of the load; dynamic parameters: the active transient time constant and reactive transient time constant of the load; fault parameters: the surge in active power and the surge in reactive power during load faults. The identification of load model parameters includes: collecting load voltage and power operation data; classifying load types based on average daily electricity consumption duration, peak-to-valley ratio, and reactive power-voltage sensitivity characteristics using clustering algorithms; and identifying load model parameters by setting weights according to the degree of load impact on the distribution network and using hierarchical weighted least squares method.

6. The method according to claim 5, characterized in that, The hierarchical weighted least squares method has the following objective function: in, , These are the calculated values ​​of active power and reactive power of the load, obtained from the static power characteristic equation, the dynamic response equation, and the fault reference value, respectively. , These are the active power measurements of the load during normal operation and the reactive power measurements of the load during a fault. The weighting coefficient for load type m is defined based on the degree of impact of different types of loads on the power grid.

7. A power grid voltage control method, characterized in that, The source-load-storage model established based on the method described in any one of claims 1-6 includes: The centralized coordinator divides the network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type. The distributed coordinator reports the collected running status data of each node in the corresponding cluster to the centralized coordinator. The centralized coordinator determines the global control target of each cluster based on the overall network topology and the received operating status data, and issues adjustment instructions to the corresponding distributed coordinators. The global control target includes the total power adjustment amount of the cluster. Based on the received adjustment instructions and combined with the source-load-storage model, the distributed coordinator determines the power allocated to the devices connected to the nodes; it verifies the allocation results through the source-load-storage model and generates control instructions for each device.

8. The method according to claim 7, characterized in that, The method further includes: the distributed coordinator determines the penetration rate of distributed photovoltaics in the cluster, the load type of the cluster, and the type of disturbance to the cluster based on the collected operating status data of the corresponding cluster, and reports them as the current scenario; The centralized coordinator, based on the overall network topology and received operational status data, determines the global control objectives for each cluster and issues adjustment instructions to the corresponding distributed coordinators, including: The centralized coordinator, based on the overall network topology, received operational status data, and current scenario data, initializes the ADMM algorithm using a pre-established ADMM historical best solution database, determines the global control objectives of each cluster, and issues adjustment instructions to the corresponding distributed coordinators. The ADMM historical optimal solution database contains the correspondence between each historical scenario and the initial value. The historical scenarios include penetration rate, load type and disturbance type. The initial values ​​include: initial values ​​of Lagrange multipliers, initial control quantities of photovoltaic reactive power and initial control quantities of energy storage active power.

9. The method according to claim 8, characterized in that, The centralized coordinator, based on the overall network topology, received operational status data, and current scenario data, initializes the ADMM algorithm using a pre-established ADMM historical best solution database, determines the global control objectives for each cluster, and issues adjustment instructions to the corresponding distributed coordinators, including: The centralized coordinator constructs an ADMM historical optimal solution database containing the correspondence between each historical scenario and the initial value; For each cluster, the centralized coordinator calculates the similarity between the current scene of the cluster and each historical scene; and selects the initial value of the cluster based on the similarity value. Substitute the initial values ​​and operating status data of each selected cluster into the global objective function, and use the ADMM algorithm for optimization to obtain the total active power adjustment and total reactive power adjustment for each cluster. The total active power adjustment and total reactive power adjustment of each cluster are used as the global control targets for each cluster, and adjustment instructions are issued to the corresponding distributed coordinators.

10. The method according to claim 7, characterized in that, Based on received adjustment instructions and in conjunction with the source-load-storage model, the distributed coordinator determines the power allocated to the devices connected to the nodes, including: The distributed coordinator obtains the total active power adjustment and total reactive power adjustment of the cluster from the adjustment command; The distributed coordinator uses the source-load-storage model and the local node operating status to determine the voltage deviation factor and resource margin factor of each node; The distributed coordinator decomposes the total active power adjustment and total reactive power adjustment of the cluster according to the voltage deviation factor and resource margin factor of each node, and obtains the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value of each node. The distributed coordinator uses the photovoltaic reactive power adjustment allocation value, the energy storage active power adjustment allocation value, and the source-load-storage model and model parameters of each node to determine the power allocated to the devices connected to the node.

11. The method according to claim 7, characterized in that, The centralized coordinator divides the network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type, including: The centralized coordinator determines the distribution network topology, node power, and equivalent impedance between nodes; The centralized coordinator determines the net power correlation coefficient between nodes based on node power, and corrects the net power correlation coefficient between nodes based on whether the loads between nodes are of the same type. The centralized coordinator determines the weighted electrical distance between nodes based on the equivalent impedance between nodes and the corrected net power correlation coefficient between nodes; The centralized coordinator uses an improved K-means clustering algorithm to determine clusters in the entire distribution network based on load type proportion data, weighted electrical distance between nodes, and node voltage regulation resources.

12. The method according to claim 8, characterized in that, The distributed coordinator determines the cluster load type based on collected cluster runtime status data, including: The distributed coordinator selects the dominant node in the cluster based on a comprehensive weighted score of weighted electrical distance centrality, power fluctuation coordination among nodes, and voltage regulation resource margin. The distributed coordinator uses the load type of the dominant node in the cluster as the load type of the cluster.

13. The method according to claim 9, characterized in that, The centralized coordinator performs similarity calculations between the current scene of each cluster and each historical scene, including: Use the following formula to calculate the similarity between the current scene and historical scenes. : in, The weights for the impact of penetration rate, load type, and disturbance type on the optimization are: This represents the current penetration rate in various scenarios. For historical scene penetration rate; This refers to the degree of difference between the load type in the current scenario and the load type in historical scenarios. This is a perturbation type matching function; The type of disturbance in the current scene; This refers to the type of disturbance in the historical scene.

14. The method according to claim 10, characterized in that, The distributed coordinator decomposes the total active power adjustment and total reactive power adjustment of the cluster according to the voltage deviation factor and resource margin factor of each node, obtaining the photovoltaic reactive power adjustment allocation value and energy storage active power adjustment allocation value for each node, including: Decompose any cluster using the following formula a Total active power adjustment : Decompose any cluster using the following formula a Total reactive power adjustment : in, This represents the active power adjustment of energy storage at node j. For standardized nodes j Voltage deviation factor; For standardized nodes j The active resource margin factor; The reactive power resource margin factor of node j after standardization; Let be the photovoltaic reactive power adjustment amount at node j; , The weight of the voltage deviation factor is set by the distributed coordinator based on the voltage state.

15. The method according to claim 11, characterized in that, The centralized coordinator determines the weighted electrical distance between nodes based on the equivalent impedance between nodes and the corrected net power correlation coefficient between nodes, including: The weighted electrical distance between nodes is determined using the following formula: in, Weighted electrical distance; For nodes i and nodes j The normalized value of the equivalent impedance; Power correlation coefficient; For synergy-weighted terms; , Let be the weighting coefficient, satisfying .

16. The method according to claim 9, characterized in that, The global objective function is: ； Where L is the global objective function value; For adjacent clusters a and b Objective function between; For clusters a The objective function; For clusters b The objective function; and Clusters a , b The control variables all include photovoltaic reactive power control quantities and energy storage active power control quantities; These are the initial Lagrange multipliers; The penalty parameter is K; K is the number of clusters. Indicates the total number of nodes in the cluster; Represents a cluster a The reduction in photovoltaic power; Represents a cluster a The Middle i The voltage of each node; Indicates the node reference voltage; For clusters a Network loss; For three different items in the cluster a The weights in the objective function are defined by the centralized coordinator.

17. A source-load-storage modeling system, characterized in that, Based on a two-layer control architecture, the two-layer control architecture includes: at least two distributed coordinators and one centralized coordinator, wherein one cluster corresponds to one distributed coordinator; the distributed coordinator connects to all nodes within its corresponding cluster, and the nodes include: photovoltaic, energy storage and load; the centralized coordinator is deployed in the dispatch center; Each distributed coordinator is used to report the collected operating status data of each node in the corresponding cluster to the centralized coordinator; and to receive model parameters and update the source-load-storage model using the model parameters; wherein, the source-load-storage model includes: distributed photovoltaic model, load model and energy storage model; The centralized coordinator is used to identify and update the model parameters of the pre-built source-load-storage model using runtime status data, and then distributes the updated model parameters to each distributed coordinator. The construction of a distributed photovoltaic model includes: characterizing the reactive power characteristics of photovoltaics; The distributed photovoltaic (PV) model parameters include: steady-state parameters and fault ride-through parameters. The steady-state parameters include the internal resistance of a single diode; the fault ride-through parameters include the reactive power support factor in LVRT mode. Active power retention coefficient under HVRT mode Fault current limit With transient response time ; The identification of distributed photovoltaic model parameters includes: minimizing the error between the photovoltaic characteristics calculated by the distributed photovoltaic model and the actual collected operating status data by using the least squares method based on fault weights, thereby identifying the photovoltaic model parameters; The identification of distributed photovoltaic (PV) model parameters specifically includes: using the following objective function to identify the PV model parameters: in, This represents the error between the photovoltaic characteristics calculated by the distributed photovoltaic model and the actual collected operating status data; t represents time. and These are the actual measured photovoltaic active power and the photovoltaic active power calculated using a distributed photovoltaic model, respectively. and S represents the actual measured photovoltaic reactive power and the reactive power calculated using the distributed photovoltaic model; S represents the time interval for steady-state data; F represents the time interval for fault data. and These are the weights for steady-state data and fault data, respectively; when the error When the minimum value is reached, the parameters of the distributed photovoltaic model achieve the optimal identification value.

18. A power grid voltage control system, characterized in that, The system based on claim 17 includes: A centralized coordinator is used to divide the distribution network into clusters based on the distribution network topology, node power, equivalent impedance between nodes, and load type. Based on the overall distribution network topology and received operating status data, it determines the global control target of each cluster and issues adjustment instructions to the corresponding distributed coordinator. The global control target includes the total power adjustment of the cluster. The distributed coordinator is used to report the collected operating status data of each node in the corresponding cluster to the centralized coordinator; based on the received adjustment instructions, combined with the source-load-storage model, it determines the power allocated to the devices connected to the nodes; it verifies the allocation results through the source-load-storage model and generates control instructions for each device.

Images