Reliable estimation method for frequency modulation sharing proportion of multi-unit cluster in thermal power plant and related device

Abstract

The application discloses a kind of power plant multi-unit cluster frequency modulation sharing ratio reliable estimation method and related device, comprising: 1) obtaining frequency modulation information table;2) generate initial sharing ratio table;3) correction coefficient is generated based on correction sharing ratio table;4) under the operating limit condition of ramp rate limit and regulation margin limit, according to correction sharing ratio table, the frequency modulation demand is decomposed, and unit regulation target table is generated;5) according to unit regulation target table, each unit regulation target is issued, and the deviation of unit regulation target table and actual output change is calculated, and the following deviation record of this frequency modulation period is generated;6) output correction sharing ratio table, unit regulation target table, the following deviation record of this frequency modulation period and the updated correction coefficient, the method and related device can improve the reliability of sharing ratio in continuous period.

Description

Technical Field

[0001] This invention belongs to the field of frequency regulation control of multiple units in thermal power plants, and relates to a reliable estimation method and related device for the frequency regulation sharing ratio of multiple units in thermal power plants. Background Technology

[0002] In grid-connected operation, thermal power plants typically respond to grid AGC or frequency regulation commands at the plant level. This requires decomposing the "plant-level frequency regulation demand" into target output changes for multiple units within the frequency regulation cycle. Due to differences in unit capacity, ramp-up capability, adjustment margins (both upward and downward), start-up / shutdown status, and health / aging levels, the amount of regulation each unit can handle varies. Furthermore, parallel operation of multiple units creates coupling relationships due to electrical connections, shared auxiliary equipment and systems, control channels, and signal links. This can cause the regulation behavior of one unit to affect the adjustable resources, response process, and control stability of other units. In actual operation, valve characteristics, boiler combustion conditions, coal grinding, and feedwater processes can cause the following capability to drift over time, and this drift is often reflected in periodic records as a "deviation between target and actual output changes." To ensure the safe and compliant execution of frequency regulation, the decomposition process must also meet operational constraints such as ramp rate limits and adjustment margin limits. When constraints prevent some demands from being allocated, they must be redistributed to form unit regulation targets that can be issued and to record the following deviations in a closed loop.

[0003] The current methods for determining the frequency regulation sharing ratio of multiple generating units mostly adopt rules or simplified models. For example, the ratio is allocated according to the rated capacity of the units, available adjustment margin, ramp rate margin, unit priority, or fixed order, and the results are normalized before being used for demand decomposition. Some schemes write the decomposition process as a linear or quadratic programming problem to solve the target adjustment of each unit under ramp rate, upper and lower limits, and reserve constraints. Some engineering implementations introduce weights or compensation factors based on historical response performance. After updating the weights in each frequency regulation cycle, they participate in the allocation of the next cycle to adapt to changes in the operating status of the units.

[0004] The aforementioned schemes often equate the relationships between units to a "set of independent units" or a single similarity effect, making it difficult to simultaneously express the differences between various types of coupling, such as electrical coupling, resource competition, and control interference, and their propagation effects between strong local coupling and plant-wide coordination. This results in the difficulty of synchronously adjusting the sharing ratio with changes in operating modes. Furthermore, the allocation based on fixed rules or static weights lacks an interpretable recursive calibration channel for the cross-cycle drift of the actual following capability of the units. Deviation records are difficult to reliably convert into acceptable correction amounts, thus affecting the reliability of the sharing ratio over continuous periods. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a reliable estimation method and related device for the frequency regulation sharing ratio of multiple units in a thermal power plant. This method and related device can improve the reliability of the sharing ratio in a continuous period.

[0006] To achieve the above objectives, this invention discloses a reliable estimation method for the frequency regulation sharing ratio of multiple units in a thermal power plant cluster, comprising: 1) Obtain the frequency modulation information table; 2) Construct a unit coupling diagram based on the coupling relationship between units in the frequency regulation information table. The edges of the unit coupling diagram carry coupling type identifiers. Input the unit coupling diagram and the real-time operating status of each unit into a hierarchical graph convolutional network. Generate an initial sharing ratio table based on the output of the hierarchical graph convolutional network. 3) Input the initial sharing ratio table, following deviation record, and correction coefficient into the recursive calibration framework. The recursive calibration framework makes a judgment based on the following deviation record computer group following consistency amount and following judgment threshold. When the group following consistency amount is not less than the following judgment threshold, the correction coefficient is updated under the upper limit constraint of correction amplitude, and the updated correction coefficient is used to correct the initial sharing ratio table to obtain the correction sharing ratio table. When the group following consistency amount is less than the following judgment threshold, the correction sharing ratio table is generated based on the correction coefficient. 4) Under the operating constraints of ramp rate limit and adjustment margin limit, decompose the frequency regulation demand according to the correction sharing ratio table and generate the unit adjustment target table; 5) Issue adjustment targets to each unit according to the unit adjustment target table, collect the actual output changes of each unit, calculate the deviation between the unit adjustment target table and the actual output changes, and generate a frequency regulation cycle follow-up deviation record. 6) Output the correction sharing ratio table, the unit regulation target table, the following deviation record for this frequency regulation cycle, and the updated correction coefficient.

[0007] Furthermore, the specific operation of step 1) is as follows: Obtain the topology and control connection information of thermal power plant units, extract the coupling relationships between units, write a coupling type identifier for each coupling relationship, and organize them into a coupling type adjacency table according to the coupling type identifier; Collect the real-time operating status of each unit, and assemble the unit operating status feature matrix according to the unit index. The operating status features include the unit's current output, adjustment margin, ramp-up margin, start-up and shutdown status, and aging degree. Obtain the frequency regulation demand and operating constraints, and parse the operating constraints into ramp rate limits and regulation margin limits for each unit; Read the following deviation record and correction coefficient of the previous frequency regulation cycle, and merge the coupling type adjacency table, unit operating status characteristic matrix, frequency regulation demand, operating constraints, following deviation record of the previous frequency regulation cycle and correction coefficient to generate a frequency regulation information table.

[0008] Furthermore, the unit coupling graph and the unit operating state feature matrix are input into a hierarchical graph convolutional network. In the coupling type convolutional layer of the hierarchical graph convolutional network, the corresponding convolutional neuron group is called for each coupling type identifier. The unit operating state feature matrices of adjacent units are aggregated according to the coupling type adjacency list and mapped to a unified dimension through a fully connected neuron group to obtain the type influence matrix. In the type fusion layer of the hierarchical graph convolutional network, the various type influence matrices corresponding to the same unit are concatenated in a fixed order and input into a fully connected neuron group to output the unit local representation matrix. In the unit propagation layer, based on the unit coupling graph, the local representation matrix of the unit is propagated once by adjacency propagation and transformed by a fully connected neuron group to obtain the local propagation representation matrix. In the whole plant propagation layer, the whole plant summary vector is calculated on the local propagation representation matrix. The whole plant summary vector and the local propagation representation matrix are concatenated by the unit dimension and then input into the fully connected neuron group to output the unit coupling representation matrix. In the sharing ratio generation layer of the hierarchical graph convolutional network, the sharing weight vector is output for each unit by a fully connected neuron group. The sharing weight vector is then normalized to obtain the initial sharing ratio table.

[0009] Furthermore, type convergence vector Obtained by the following formula:

[0010] in, Indicates the unit In coupling type identifier The type convergence vector below; This indicates an adjacency list of coupled types. The given unit In coupling type identifier The set of adjacent unit indexes; and Indicates the unit index; Indicates the unit The start / stop state components; Indicates the unit In coupling type identifier The type degree and take The number of elements; Indicates the unit In coupling type identifier The type degree below; Indicates the unit The unit operating state feature vector is taken from the unit operating state feature matrix. The corresponding row vector; when When the value is zero, set the denominator to one. Converge the type vector. Input a fully connected neuron group and press and Mapping to a unified dimension yields a vector of type influence. Stack all the type influence vectors of all units according to their unit indices to form a type influence matrix. .

[0011] Furthermore, the specific operation of step 3) is as follows: Input the initial sharing ratio table, the follow-up deviation record of the previous frequency modulation cycle and the correction coefficient into the recursive calibration framework, and obtain the follow-up judgment threshold and the upper limit of the correction amplitude. In the response consistency calculation unit, the deviation direction and deviation magnitude information between the unit adjustment target and the actual output change in the following deviation record are read according to the unit dimension, and the unit following consistency vector is generated. The unit following consistency vector is compared with the following judgment threshold for each unit to obtain the unit following judgment result. When the unit follow-up judgment result is that the unit follow-up consistency amount is not less than the follow-up judgment threshold, the correction coefficient update amount is generated according to the follow-up deviation record of the corresponding unit, and the correction coefficient update amount is limited to the upper limit of the correction range. The correction coefficient is updated unit by unit by using the correction coefficient update amount to obtain the updated correction coefficient. When the unit follow-up judgment result is that the unit follow-up consistency amount is less than the follow-up judgment threshold, the correction coefficient update amount of the corresponding unit is set to zero and the correction coefficient remains unchanged. The corrected sharing weight vector is obtained by multiplying the initial sharing ratio table element by element with the updated correction coefficients according to the unit dimension. The corrected weighting vector is normalized proportionally to obtain the corrected weighting ratio table.

[0012] Furthermore, the unit follows the consistency quantity for:

[0013] in, Indicates the unit The unit follows the consistency quantity; This represents the directional consistency weighting coefficient; Indicates the number of sampling intervals; Indicates the sampling interval index; Indicates the unit In the The directional component of the sampling interval deviation; Indicates the unit In the The amplitude component of the sampling interval deviation; This represents the sampling interval index used to calculate the mean of the deviation amplitude components; This indicates a positive constant that avoids a denominator of zero; Indicates exponentiation; This represents absolute value operations.

[0014] Furthermore, the specific operation of step 5) is as follows: Read the unit regulation target from the unit regulation target table one by one, and send the unit regulation target to the frequency regulation control interface of the corresponding unit; At the beginning of this frequency regulation cycle, the current output of each unit is collected, and at the end of this frequency regulation cycle, the final output of each unit is collected, and the actual output change of each unit is calculated. The unit regulation targets in the unit regulation target table are converted into target output changes. The target output changes are then aligned with the actual output changes according to the unit index, and the deviation is calculated for each unit to obtain the deviation direction information and deviation magnitude information. The deviation direction information and deviation magnitude information are assembled according to the unit dimension to generate cost frequency regulation cycle following deviation records.

[0015] This invention discloses a reliable estimation system for the frequency regulation sharing ratio of multiple units in a thermal power plant cluster, comprising: The acquisition module is used to acquire the frequency modulation information table; The generation module is used to construct a unit coupling diagram based on the coupling relationship between units in the frequency regulation information table. The edges of the unit coupling diagram carry coupling type identifiers. The unit coupling diagram and the real-time operating status of each unit are input into a hierarchical graph convolutional network, and an initial sharing ratio table is generated based on the output of the hierarchical graph convolutional network. The calibration module is used to input the initial sharing ratio table, the following deviation record, and the calibration coefficient into the recursive calibration framework. The recursive calibration framework makes a judgment based on the following deviation record computer group following consistency amount and the following judgment threshold. When the group following consistency amount is not less than the following judgment threshold, the calibration coefficient is updated under the upper limit constraint of the calibration range, and the updated calibration coefficient is used to correct the initial sharing ratio table to obtain the calibration sharing ratio table. When the group following consistency amount is less than the following judgment threshold, the calibration sharing ratio table is generated based on the calibration coefficient. The limiting module is used to decompose the frequency regulation demand based on the correction sharing ratio table and generate the unit regulation target table under the operating constraints of ramp rate limit and adjustment margin limit; The calculation module is used to issue the adjustment targets of each unit according to the unit adjustment target table, collect the actual output changes of each unit, calculate the deviation between the unit adjustment target table and the actual output changes, and generate the frequency regulation cycle follow-up deviation record. The output module is used to output the correction sharing ratio table, the unit regulation target table, the following deviation record for the current frequency regulation cycle, and the updated correction coefficient.

[0016] This invention discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant.

[0017] This invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant.

[0018] The present invention has the following beneficial effects: The reliable estimation method and related devices for frequency regulation sharing ratio of multi-unit clusters in thermal power plants described in this invention, in specific operation, acquire the unit coupling relationship, real-time operating status, frequency regulation demand, operating constraints, previous cycle following deviation, and correction coefficient to generate a frequency regulation information table; construct a unit coupling graph with coupling type identifiers on the edges; a hierarchical graph convolutional network first groups and converges according to coupling type and performs local and plant-wide two-level propagation fusion, outputting and normalizing the initial sharing ratio; a recursive calibration framework calculates the following consistency quantity based on the following deviation, updates the correction coefficient and corrects the ratio by combining the threshold and the upper limit of the correction amplitude; under the constraints of ramp rate and adjustment margin, the frequency regulation demand is decomposed to form the unit adjustment target and new deviations are recorded. This can ensure that the sharing ratio continuously reflects the coupling influence and changes in actual following capability while meeting operating constraints, thereby improving the reliability of the sharing ratio in continuous cycles. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the method of the present invention; Figure 2 The flowchart for step 1); Figure 3 This is a flowchart for step 2); Figure 4 The flowchart for step 3); Figure 5 This is a flowchart for step 4); Figure 6The flowchart for step 5); Figure 7 This is the flowchart for step 6). Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0025] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0026] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0028] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0029] Example 1 refer to Figure 1 The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant, as described in this invention, includes the following steps: 1) Obtain the coupling relationship between units, the real-time operating status of each unit, the frequency regulation demand, the operating constraints, the follow-up deviation record and correction coefficient of the previous frequency regulation cycle, and generate a frequency regulation information table; It should be noted that the reference Figure 2 The system acquires the topology and control connection information of thermal power plant units, extracts the coupling relationships between units, writes a coupling type identifier for each coupling relationship, and organizes them into a coupling type adjacency table according to the coupling type identifier; it collects the real-time operating status of each unit, assembles the unit operating status feature matrix according to the unit index, and the operating status features include the unit's current output, regulation margin, ramp margin, start-stop status, and aging degree; it acquires the frequency regulation demand and operating constraints, and parses the operating constraints into ramp rate limits and regulation margin limits for each unit; it reads the follow-up deviation record and correction coefficient of the previous frequency regulation cycle, and merges the coupling type adjacency table, the unit operating status feature matrix, the frequency regulation demand, the operating constraints, the follow-up deviation record of the previous frequency regulation cycle, and the correction coefficient to generate a frequency regulation information table.

[0030] The specific operation of step 1) is as follows: All generating units within a thermal power plant that participate in frequency regulation are uniformly numbered to form a set of unit indexes. The unit reads the topology and control connection information of the thermal power plant units, including parallel grid connection relationships, shared system connection relationships, and control channel connection relationships. The parsing unit extracts the inter-unit coupling relationships based on the thermal power plant unit topology and control connection information, representing each inter-unit coupling relationship as a directed association from the source unit index to the target unit index, and writes a coupling type identifier for this directed association. Coupling type identifier Taken from the set of coupling type identifiers Coupling type identifier set It should include at least electrical coupling identifiers, resource contention identifiers, and control channel interference identifiers. These are then identified by coupling type. Group directed associations and identify each coupling type. Generate a coupled type adjacency list Adjacency list of coupling type Using the unit index as the key, a set of adjacent unit indexes is given, enabling the hierarchical graph convolutional network to group and aggregate adjacency information according to the coupling type identifier.

[0031] Within the same frequency regulation cycle, the real-time operating status of each unit is collected, and specific data is collected for each unit. Generate unit operating status feature vector And assembled into a unit operating status feature matrix according to the unit index order. Unit operating status feature vector Characteristic components include the unit's current output. Increase the adjustment margin Lowering the adjustment margin Increase the ramp margin Reduce the ramp margin Start-stop status Aging degree Current output of the unit Data collected from unit output measuring points; adjustment margin increased. The maximum allowable output of the unit With the current output of the unit The difference is obtained, and the adjustment margin is lowered. Based on the current output of the unit With the unit's allowable lower limit of output The difference is obtained; the ramp margin is increased. Increase the ramp rate limit by the unit With frequency modulation cycle duration Multiply to obtain, and reduce the ramp margin. Lowering the ramp rate limit for the generator unit With frequency modulation cycle duration The frequency modulation cycle duration is obtained by multiplication. Provided by the frequency modulation strategy configuration; start / stop status The data is collected from the unit's operating permit status and used to identify whether the unit is ready for frequency regulation; aging level. The results are obtained from the unit health status assessment. The unit health status assessment results are calculated from three basic quantities: the unit's cumulative operating hours, the number of start-ups and shutdowns, and the deviation of key performance parameters. The deviation of key performance parameters is synthesized from the normalized deviations of heat consumption, main steam pressure deviation, and valve opening deviation.

[0032] The scheduling interface obtains the plant-level frequency regulation requirements. Plant-level frequency regulation demand This indicates that the total output change demand of each unit needs to be broken down into the current frequency regulation cycle. The analysis unit obtains the operating constraints. and running restrictions Analysis into the ramp rate limiting parameter set for each unit With adjustment margin limit parameter group Climb rate limiting parameter group Include and Adjusting margin limit parameter group Include and ,in From increasing the adjustment margin Obtain directly Reduce adjustment margin This allows operational constraints to be directly applied at the unit index granularity, constraining the upward and downward adjustment directions of unit regulation targets. The estimated characteristics of the unit sharing ratio are shown in Table 1. Table 1

[0033] The status interface reads the follow deviation record from the previous frequency modulation cycle. With correction factor Previous frequency modulation cycle follow-up deviation record Includes deviation direction component at the unit level. With deviation amplitude component Among them, the deviation direction component The deviation amplitude component is obtained by comparing the direction of change of the target output of the unit with the direction of change of the actual output of the unit. The correction coefficient is obtained from the absolute value of the difference between the target output change and the actual output change of the unit. Includes correction factor components at the unit level. This is used to perform cross-cycle correction of the initial sharing ratio table within a recursive calibration framework. Subsequently, the coupled type adjacency table is... Unit operating status characteristic matrix Plant-level frequency regulation demand Slope rate limiting parameter group Adjustment margin limit parameter group Previous frequency modulation cycle follow-up deviation record and correction factor Merge the data using structured fields to generate a frequency modulation information table. Among them, the frequency modulation information table The fields can be directly read as coupled adjacency lists by hierarchical graph convolutional networks. With unit operating status characteristic matrix It can be directly read as a follow-up deviation record by the recursive calibration framework. With correction factor The data fields for the frequency modulation system are shown in Table 2: Table 2

[0034] 2) Construct a unit coupling diagram based on the coupling relationship between units in the frequency regulation information table. The edges of the unit coupling diagram carry coupling type identifiers. Input the unit coupling diagram and the real-time operating status of each unit into a hierarchical graph convolutional network. The hierarchical graph convolutional network aggregates the influence of adjacent nodes according to the coupling type identifier to form type influence quantities. Then, it merges the type influence quantities according to the two-level fusion of unit local propagation and plant-wide global propagation to form unit coupling characterization quantities. The initial sharing ratio table is generated from the unit coupling characterization quantities. It should be noted that the reference Figure 3The coupling relationships between generating units are read from the frequency regulation information table. Generating a generating unit coupling graph with generating units as nodes, and writing the coupling relationship between each generating unit into the edge attributes of the generating unit coupling graph as a coupling type identifier, the generating unit coupling graph is converted into a coupling type adjacency list according to the coupling type identifier. The coupling type adjacency list provides a set of adjacent generating unit indices for each coupling type identifier. The real-time operating status of each generating unit is read from the frequency regulation information table and organized into a generating unit operating status feature matrix according to the generating unit index. The generating unit coupling graph and the generating unit operating status feature matrix are input into a hierarchical graph convolutional network. In the coupling type convolutional layer of the hierarchical graph convolutional network, the corresponding convolutional neuron group is called for each coupling type identifier. The generating unit operating status feature matrices of adjacent generating units are aggregated according to the coupling type adjacency list and mapped to a unified dimension through a fully connected neuron group to obtain the type. The layered graph convolutional network generates an influence matrix. In the type fusion layer, the influence quantities of various types corresponding to the same unit are concatenated in a fixed order and input into a fully connected neuron group, outputting a unit local representation matrix. In the local propagation layer, the unit local representation matrix is ​​propagated once based on the unit coupling graph and transformed by a fully connected neuron group to obtain a local propagation representation matrix. In the whole plant propagation layer, the whole plant summary vector is calculated on the local propagation representation matrix. The whole plant summary vector and the local propagation representation matrix are concatenated according to the unit dimension and input into a fully connected neuron group, outputting a unit coupling representation matrix. In the sharing ratio generation layer, the unit coupling representation matrix is ​​processed group by group to output a sharing weight vector. The sharing weight vector is normalized to obtain an initial sharing ratio table.

[0035] The specific operation of step 2) is as follows: Layered graph convolutional networks from frequency modulation information tables Read the unit index set The coupling relationship between generating units is established using a set of generating unit indexes. As a set of nodes, a unit coupling diagram is constructed. Unit coupling diagram containing edge set edge set Each edge is identified by the source unit index, the target unit index, and the coupling type identifier. The configuration makes the unit coupling diagram The edge attribute carries a coupling type identifier. Based on edge sets Identified by coupling type Grouping, generating a coupled adjacency list. Adjacency list of coupling type For each coupling type identifier, a set of adjacent unit indexes is provided for each unit. It includes at least electrical coupling identifiers, resource contention identifiers, and control channel interference identifiers, and serves as the source of the fixed splicing order for the type fusion layer.

[0036] Layered graph convolutional networks from frequency modulation information tables The system reads the real-time operating status of each unit and organizes it into a unit operating status feature matrix. Unit operating status characteristic matrix The unit operating state feature vector is formed by stacking the unit operating state feature vectors in order of unit index. The unit operating state feature vector is written as: ,in, For the current output of the unit, To increase the adjustment margin, To reduce the adjustment margin, To increase the ramp margin To reduce the ramp margin. In start / stop state. To determine the degree of aging, the unit coupling diagram is... With unit operating status characteristic matrix Input a layered graph convolutional network.

[0037] In coupled-type convolutional layers, each coupling type is identified. Configure the convolutional neuron group and read the type mapping parameters from the model parameter file. With bias For each unit Adjacency list based on coupling type Extract the adjacent unit index set and calculate the type convergence vector. Type convergence vector Used to express adjacent unit to unit while maintaining the difference in coupling source. Type influence, type convergence vector Obtained by the following formula:

[0038] in, Indicates the unit In coupling type identifier The type convergence vector below; This indicates an adjacency list of coupled types. The given unit In coupling type identifier The set of adjacent unit indexes; and Indicates the unit index; Indicates the unit The start / stop state components; Indicates the unit In coupling type identifier The type degree and take The number of elements; Indicates the unit In coupling type identifier The type degree below; Indicates the unit The unit operating state feature vector is taken from the unit operating state feature matrix. The corresponding row vector; when When the value is zero, set the denominator to one. Converge the type vector. Input a fully connected neuron group and press and Mapping to a unified dimension yields a vector of type influence. Stack all the type influence vectors of all units according to their unit indices to form a type influence matrix. .

[0039] In the type fusion layer, by The fixed order concatenates the type influence vectors of the same unit into Form of unit type fusion input vector And read the type fusion mapping parameters from the model parameter file. With bias Integrate unit type into input vector Inputting a fully connected group of neurons yields the local representation vector of the crew. The unit's local representation vectors are stacked to form the unit's local representation matrix. .

[0040] In the local propagation layer, based on the unit coupling diagram Local representation matrix of the unit Performing one adjacency propagation yields the local propagation representation vector. And read the local propagation mapping parameters from the model parameter file. With bias Complete the fully connected transformation and stack them to form a local propagation representation matrix. In the plant-wide propagation layer, the local propagation representation matrix is... Summing along the unit dimension and normalizing by the number of units yields the plant-wide summary vector. ,Will Concatenate and read plant-wide propagation mapping parameters from model parameter file With bias The unit coupling characterization vector is obtained. Stacking forms the unit coupling characterization matrix The network layer processing flow is shown in Table 3: Table 3

[0041] In the sharing ratio generation layer, the unit coupling representation matrix is... Each unit receives a linear fully connected neuron group to obtain a weighted vector. Weight vector Write it according to the unit index. For the weight-sharing vector Perform nonnegation processing and normalize proportionally according to the sum of all shared weights to obtain the initial sharing ratio table. Initial sharing ratio table Write it according to the unit index. The coupling type identifiers are shown in Table 4.

[0042] Table 4

[0043] 3) Input the initial sharing ratio table, following deviation record, and correction coefficient into the recursive calibration framework. The recursive calibration framework makes a judgment based on the following deviation record computer group following consistency amount and following judgment threshold. When the group following consistency amount is not less than the following judgment threshold, the correction coefficient is updated under the upper limit constraint of correction amplitude, and the updated correction coefficient is used to correct the initial sharing ratio table to obtain the correction sharing ratio table. When the group following consistency amount is less than the following judgment threshold, the correction sharing ratio table is generated based on the correction coefficient. It should be noted that the reference Figure 4 The initial sharing ratio table, the following deviation record of the previous frequency regulation cycle, and the correction coefficient are input into the recursive calibration framework, and the following judgment threshold and the upper limit of the correction amplitude are obtained. In the response consistency calculation unit, the deviation direction and deviation amplitude information of the unit adjustment target and the actual output change in the following deviation record are read according to the unit dimension, and the unit following consistency vector is generated. The unit following consistency vector is judged against the following judgment threshold for each unit to obtain the unit following judgment result. When the unit following judgment result is that the unit following consistency is not less than the following judgment threshold, the correction coefficient update amount is generated according to the following deviation record of the corresponding unit, and the correction coefficient update amount is limited to the upper limit of the correction amplitude. The correction coefficient is updated for each unit using the correction coefficient update amount to obtain the updated correction coefficient. When the unit following judgment result is that the unit following consistency is less than the following judgment threshold, the correction coefficient update amount of the corresponding unit is set to zero and the correction coefficient remains unchanged. The initial sharing ratio table is multiplied element by element with the updated correction coefficient according to the unit dimension to obtain the corrected sharing weight vector. The corrected sharing weight vector is subjected to proportional normalization processing to obtain the corrected sharing ratio table.

[0044] The specific operation of step 3) is as follows: Initial sharing ratio table Previous frequency modulation cycle follow-up deviation record and correction factor Input the recursive calibration framework and read the follow-up decision threshold from the parameter configuration. With the upper limit of the correction amplitude The number of generating units is recorded as follows: The unit index is denoted as Initial sharing ratio table Written as Correction coefficient Written as .

[0045] Previous frequency modulation cycle follow-up deviation record Organized by unit, each unit The corresponding record item is Deviation direction component Written as Deviation amplitude component Written as The number of sampling intervals is denoted as The sampling interval index is denoted as For each sampling interval First, read the change in the unit's regulation target and the change in actual output within the sampling interval, and then compare the directions of the change in the unit's regulation target and the change in actual output to generate... When the change in the unit's regulation target is zero, take... When the change in the unit's adjustment target and the change in actual output are in the same direction, take... When the change in the unit's adjustment target is opposite to the change in the actual output, take... Then, the absolute value of the difference between the unit's target adjustment change and the actual output change is used to generate... ,make These are deviation amplitude samples that can be directly used for numerical calculations.

[0046] The response consistency calculation unit generates a unit-following consistency vector at the unit level. The unit follows the consistency vector. Written as Among them, the unit follows the consistency quantity Calculate using the following formula:

[0047] in, Indicates the unit The unit follows the consistency quantity; This represents the directional consistency weighting coefficient; Indicates the number of sampling intervals; Indicates the sampling interval index; Indicates the unit In the The directional component of the sampling interval deviation; Indicates the unit In the The amplitude component of the sampling interval deviation; This represents the sampling interval index used to calculate the mean of the deviation amplitude components; This indicates a positive constant that avoids a denominator of zero; Indicates exponentiation; This represents absolute value operations.

[0048] The recursive calibration framework will cause the unit to follow the consistency vector. Unit-by-unit and following judgment threshold The determination is performed to obtain the unit following determination result vector. , among which, when Not less than Time to take ,when Less than Time to take .exist Timely generation of correction coefficient update amount ,exist At that time, Set to zero. (This is the correct setting.) For the unit, first calculate the mean value of the deviation amplitude components. The calculation method is to... Summing and dividing by ; then calculate the direction penalty factor The calculation method is statistical. The value of the middle is The number of times and divided by Update the correction coefficients The amplitude is set to Gain coefficient The amount of correction coefficient update is given by the parameter configuration. The symbol is set to random. Increase and point to decrease the correction coefficient component The direction, and Limited to the range Internally, the correction factor update values ​​of all units are assembled into a correction factor update value vector. .

[0049] Update the vector using correction coefficients For correction coefficients Update the correction coefficients unit by unit to obtain the updated correction coefficients. ,in, Depend on and Add them together to get the result, and then... Apply a nonnegative lower bound ,when Less than At that time, Set as nonnegative lower bound Provided by parameter configuration. Initial sharing ratio table. By unit dimension and updated correction factor Element-wise multiplication yields the corrected weight distribution vector. ,in, Depend on and Multiply to obtain; for the corrected weight distribution vector Perform proportional normalization processing to obtain the correction sharing ratio table. The proportional normalization process adopts the following approach: For each unitary denominator Normalization yields the corresponding .

[0050] 4) Under the operating constraints of ramp rate limit and adjustment margin limit, decompose the frequency regulation demand according to the correction sharing ratio table and generate the unit adjustment target table; It should be noted that the reference Figure 5 The system reads the correction sharing ratio table, frequency regulation demand, current output of each unit, ramp margin, and regulation margin from the frequency regulation information table, and also reads the ramp rate limit and regulation margin limit from the operating constraints. It decomposes the frequency regulation demand into a unit-by-unit decomposition according to the correction sharing ratio table to obtain the unit target regulation vector. Operating constraints are applied to the unit target regulation vector for each unit, limiting the unit target regulation within the allowable range corresponding to the ramp rate limit and the allowable range corresponding to the regulation margin limit, thus obtaining the constrained unit target regulation vector and calculating the unallocated frequency regulation demand. When unallocated frequency regulation demand exists, the system performs proportional normalization based on the correction sharing ratio table corresponding to units that still have remaining ramp margin and regulation margin, and then decomposes the unallocated frequency regulation demand again. Finally, it combines the final unit target regulation with the current output of each unit to generate the unit regulation target table.

[0051] The specific operation of step 4) is as follows: From the frequency modulation information table Read the correction sharing ratio table With frequency modulation demand And read the current output of each unit according to the unit index. Increase the ramp margin Reduce the ramp margin Increase the adjustment margin Lowering the adjustment margin Simultaneously, read the ramp-up rate limit from the runtime constraints. Lowering the gradient limit Adjusting the adjustment margin limit Lowering the adjustment margin limit And read the frequency modulation cycle duration. For each unit Calculate the allowable change in the gradient rate. With the reduction of the allowable change in the gradient rate ,in Depend on and Multiplying them together, we get Depend on and Multiply them to get the result.

[0052] Frequency modulation demand According to the correction sharing ratio table Decompose the data unit by unit to obtain the target adjustment vector for each unit. For each unit ,Will Calculated as and The product of makes exist A positive value indicates an upward adjustment of the target adjustment amount. A negative value indicates a downward adjustment of the target adjustment amount.

[0053] Target adjustment vector of the unit By applying operational constraints to each unit, the target adjustment vector of the constrained units is obtained. For each unit Calculate the allowable change amount. for , , , The minimum value of the four; calculate the allowable downward adjustment. for , , , The minimum value of the four. Execute range restrictions: when Greater than When Set as ;when Less than At that time, Set as Otherwise, let Calculate the unallocated frequency modulation demand. for minus The summation of each component makes This indicates the remaining frequency regulation demand that has not been allocated to the unit after applying ramp rate and adjustment margin limits.

[0054] When unallocated frequency modulation demand When the value is not zero, the unit with remaining ramp-up margin and remaining adjustment margin will be decomposed again. If If positive, then for each unit Calculate and adjust the allocated amount Increase the allocated amount Depend on Compare with zero and take the larger value; calculate and adjust the remaining available amount. Filter to meet Units with values ​​greater than zero form a set of redistributable units. If If it is negative, then for each unit Calculate and adjust the allocated amount Reduce the allocated amount Depend on The negative number is compared with zero, and the larger value is taken; the remaining available amount is then calculated and adjusted downwards. Filter to meet Units with a positive value of zero form a set of redistributable units. The corresponding corrected sharing ratio component is extracted from the set of redistributable units, and the normalized ratio is calculated. For each selected unit... The normalization ratio is calculated as follows: Divided by the selected units The sum of components. This includes the unallocated frequency modulation demand. The additional target adjustment is decomposed into a normalized proportion, and the remaining available quantity constraint is applied to each unit for the additional target adjustment. To make it positive, the additional target adjustment amount will be limited to... inside, if If it is negative, the additional target adjustment will be limited to [a certain value]. Within, the additional target adjustment amount that satisfies the constraints is accumulated to Obtain the final target adjustment of the unit Maintain when the set of redistributable units is empty. The process for allocating target adjustment quantities for the generating units is shown in Table 5: Table 5

[0055] Final unit target adjustment With the current output of each unit The output of each unit is summed up to obtain the target output of the unit. For each unit ,calculate Assemble the entries of the unit adjustment target table according to the unit index. Each entry contains... It is used to issue unit adjustment targets to the corresponding units and support deviation calculation based on the deviation records.

[0056] 5) Issue adjustment targets to each unit according to the unit adjustment target table, collect the actual output changes of each unit, calculate the deviation between the unit adjustment target table and the actual output changes, and generate a frequency regulation cycle follow-up deviation record. It should be noted that the reference Figure 6 The system reads the unit regulation targets from the unit regulation target table for each unit and sends them to the frequency control interface of the corresponding unit. At the beginning of the frequency regulation cycle, it collects the current output of each unit and at the end of the frequency regulation cycle, and calculates the actual output change of each unit. The unit regulation targets in the unit regulation target table are converted into target output changes, and the target output changes and actual output changes are aligned according to the unit index. The deviation is then calculated for each unit to obtain the deviation direction information and deviation magnitude information. The deviation direction information and deviation magnitude information are assembled according to the unit dimension to generate the following deviation record for the frequency regulation cycle.

[0057] The specific operation of step 5) is as follows: Read the unit index one by one from the unit adjustment target table. Final unit target adjustment With the target output of the unit The number of units is recorded as The unit index set is Control messages will be sent to each unit. Write the message to the frequency regulation control interface of the corresponding unit. The frequency regulation control interface receives the control message and sets the target output of the unit. Write to the target register of the unit frequency control controller, so that the unit frequency control controller... Frequency regulation control is executed as the target of unit regulation, while retaining Used for calculating the change in target output.

[0058] In the frequency modulation cycle triggering logic, the start time is denoted as... The end time is recorded as , and Triggered by the frequency modulation cycle duration configuration and written to the same time base. Start time Collect the current output of each unit and record it as the initial output vector. End time Collect the final output of each unit and record it as the final output vector. Calculate the actual output change for each unit according to the unit index. Actual output change From the end of the force Subtract initial output The actual output changes of each unit are obtained and assembled into an actual output change vector. To facilitate auditing and consistency checks, the data collected from each unit will be formatted... Archived by unit index.

[0059] The final unit target adjustment in the unit adjustment target table is converted into the target output change. For each unit, the target output change is defined as follows: The target output changes of each unit are then assembled into a target output change vector. The target output change vector Vector of actual output change After aligning with the unit index, calculate the deviation value for each unit. Deviation value Depend on minus Obtain and assemble into a deviation vector .

[0060] In the calculation of deviation direction information, a deviation direction component is generated for each unit. When the target output changes At that time, Set as And as a marker for zero-target change; when and At that time, Set as And serve as a marker of consistent direction; when and At that time, Set as And as a marker of opposite direction; when and At that time, Set as And as a marker of consistent direction; when and At that time, Set as And serve as a marker of opposite direction. Assemble the deviation direction components of each unit into deviation direction information. In the calculation of deviation amplitude information, a deviation amplitude component is generated for each unit. Deviation amplitude component Take the deviation value The absolute value is then assembled into deviation magnitude information. This makes the deviation magnitude information a numerical sequence that can be directly input into the recursive calibration framework.

[0061] Deviation direction information Information on deviation magnitude Assembly by unit level, cost frequency regulation cycle tracking deviation recording This frequency modulation cycle tracking deviation record. Form a sequence of record items according to the unit index. ,in, Pick , Pick and will An associated index is established with the unit record items, enabling the recursive calibration framework to read deviation direction and deviation magnitude information at the unit level and complete cross-cycle correction coefficient updates.

[0062] 6) Output the correction sharing ratio table, the unit regulation target table, the following deviation record for this frequency regulation cycle, and the updated correction coefficient.

[0063] It should be noted that the reference Figure 7 The system receives the correction sharing ratio table and updated correction coefficients output by the recursive calibration framework, and receives the unit regulation target table and the following deviation record for the current frequency regulation cycle. It aligns the correction sharing ratio table and the unit regulation target table according to the unit index to generate a frequency regulation execution instruction set. The frequency regulation execution instruction set is then sent to the frequency regulation control interface of each unit, enabling each unit to obtain the corresponding correction sharing ratio table components and unit regulation target. The following deviation record for the current frequency regulation cycle and the updated correction coefficients are written into the following deviation record field and correction coefficient field of the frequency regulation information table, so that they can be directly called when the frequency regulation information table is generated in the next frequency regulation cycle.

[0064] The specific operation of step 6) is as follows: Receive the correction sharing ratio table output by the recursive calibration framework With the updated correction coefficients It also receives the unit's regulation target table and the following deviation record for this frequency regulation cycle. The number of generating units is recorded as follows: The unit index is denoted as The unit-by-unit adjustment target table is provided. ,in, For the final target adjustment of the unit, Output to the unit's target. Frequency regulation cycle tracking deviation record. Provided per unit ,in, For the deviation direction component, This represents the deviation amplitude component.

[0065] Correction sharing ratio table The frequency regulation execution instruction set is generated by aligning the unit regulation target table with the unit index. Frequency modulation execution instruction set Form a sequence of instruction items according to the unit index. Each instruction item Written as The alignment process uses a unit index. As keys respectively from Read ,from Read Read from the unit regulation target table and And write the read results into the same instruction item. Record the following deviation for this frequency modulation cycle. Organized into a write-back entry sequence according to the unit index. Each write-back entry Written as This is used to assign values ​​to each generator unit when writing the follow-up deviation record field into the frequency regulation information table.

[0066] Frequency modulation execution instruction set The command is sent to the frequency control interface of each generating unit. For each generating unit, the command item is... Encapsulated as a control message and control messages Write the unit index The corresponding frequency modulation control interface transmit queue; the frequency modulation control interface parses control messages. And Write to the target register, Write to the target change register, Write to the sharing ratio register, Write the correction coefficient register so that the unit-side controller can read the correction sharing ratio table components and the unit regulation target and execute the control within the same frequency regulation cycle.

[0067] Write back to the entry sequence With the updated correction coefficients Write to frequency modulation information table ,in, This represents the frequency modulation information table. The field for recording deviation in the middle is denoted as The correction coefficient field is denoted as For each unit, an entry will be written back. Write Medium-sized unit index Corresponding position, make Include ,in, Indicates the field for following deviation records In the unit index The storage unit at the location; will update the correction coefficient components. Write Medium-sized unit index Corresponding position, make Include ,in Represents the correction coefficient field In the unit index After completing the writing of all the fields for all units in the storage unit, the frequency regulation information table is then processed. The write commit was executed, making the frequency modulation information table... The written follower deviation record field can be directly obtained during the next frequency modulation cycle reading. With correction coefficient field .

[0068] This invention has the following characteristics: When constructing the unit coupling diagram based on the inter-unit coupling relationships in the frequency regulation information table, the unit coupling diagram uses coupling type identifiers to classify edges, enabling it to simultaneously express the influence propagation paths under multiple physical coupling mechanisms. The hierarchical graph convolutional network does not perform a one-time mixing and aggregation of the unit coupling diagram. Instead, it first groups the adjacency information according to the coupling type identifier, and calculates the type influence quantity for each group of adjacency information separately. This preserves the differences in coupling sources in the type influence quantity, supporting the differentiation of different propagation semantics, such as electrical coupling caused by parallel operation, resource competition caused by shared systems, and mutual interference caused by control channels. The hierarchical graph convolutional network then performs two-level propagation on the type influence quantity: the first level of propagation aggregates the type influence quantity into local coupling contributions based on the nearest neighbor structure of the unit coupling diagram, explicitly incorporating the influence of locally strongly coupled units on the sharing ratio; the second level of propagation propagates and merges the local coupling contributions across the entire plant based on the cross-regional connections of the unit coupling diagram, forming a unit coupling representation quantity, allowing the remote linkage effects caused by changes in operating modes to enter the same representation space. After the unit coupling characteristics and the real-time operating status of each unit are fused within a hierarchical graph convolutional network, an initial sharing ratio table is output. This initial sharing ratio table is then normalized, enabling it to be directly used for the allocation of frequency regulation demand. Unlike existing technologies that obtain allocation results through equal distribution, fixed order, or single similarity weighting, the hierarchical graph convolutional network passes coupling type identifiers, type influence quantities, and unit coupling characteristics as essential intermediate objects level by level when generating the initial sharing ratio table. This ensures that the source of the sharing ratio corresponds to a specific propagation link in the unit coupling graph, forming a structured generation path that can be used for online frequency regulation sharing ratio estimation.

[0069] When the recursive calibration framework receives the initial sharing ratio table, following deviation records, and correction coefficients, it does not directly recalculate the unit coupling diagram or replace the generation logic of the layered graph convolutional network. Instead, it establishes targeted correction channels based on the time drift of the unit's actual following capability. First, the recursive calibration framework calculates the unit following consistency value based on the following deviation records. This value characterizes the consistency in the direction and magnitude of the deviation between the unit's regulation target and the actual output change over time, ensuring that the source of the deviation corresponds to the change in the unit's regulation execution capability. Next, the framework determines the following consistency value against a following judgment threshold. The result is used to determine the correction coefficient update method: when the following consistency value is not less than the following judgment threshold, the framework updates the correction coefficients under the upper limit constraint of the correction magnitude, ensuring that the adjustment amount of the correction coefficients is consistent with the magnitude of the following deviation records; when the following consistency value is less than the following judgment threshold, the framework keeps the correction coefficient update amount at zero. Finally, the framework corrects the initial sharing ratio table based on the correction coefficients and performs ratio normalization, outputting a corrected sharing ratio table. Unlike existing technologies that reallocate or use fixed compensation factors in each cycle, the recursive calibration framework transforms the following deviation record into a unit following consistency quantity. It also limits the updating of the correction coefficient to an interpretable range by using a following judgment threshold and a correction amplitude upper limit. The correction coefficient is output as a cross-cycle carrying quantity and re-entered into the frequency regulation information table, so that the correction sharing ratio table continuously reflects the impact of changes in the unit's actual following capability on the sharing ratio.

[0070] The input to the layered graph convolutional network consists of a unit coupling graph and the real-time operating status of each unit. The unit coupling graph uses units as nodes, with edges carrying coupling type identifiers. The unit coupling graph is represented as a coupling type adjacency list, which provides a set of adjacent unit indices based on the coupling type identifier. The real-time operating status of each unit is organized into a unit operating status feature matrix. The number of rows in the unit operating status feature matrix equals the number of units, and the columns represent the operating status feature dimensions. Operating status features include at least the unit's current output, regulation margin, ramp-up margin, start / stop status, and aging degree. The output of the layered graph convolutional network is an initial load-sharing ratio table, which is a one-dimensional ratio vector consistent with the number of units.

[0071] The intermediate layers of the hierarchical graph convolutional network are organized into two levels: coupling type grouping and two-level propagation from local to plant-wide. The first layer is a coupling type convolutional layer, which assigns a group of convolutional neurons to each coupling type identifier. These neurons first aggregate the unit operating state feature matrices of adjacent units according to the coupling type adjacency list, and then map the aggregated results to a unified dimension through a fully connected neuron layer, obtaining the type influence matrix. The second layer is a type fusion layer, which concatenates the various type influence quantities of the same unit in a fixed order, inputs them to a fully connected neuron layer, and outputs the unit's local representation matrix, enabling each unit to obtain information representations containing multiple coupling sources. The third layer is a local propagation layer, which performs adjacency propagation on the unit's local representation matrix based on the unit coupling graph and transforms it through a fully connected neuron layer, outputting a local propagation representation matrix. The fourth layer is the plant-wide propagation layer. This layer calculates the plant-wide summary vector from the local propagation representation matrix, concatenates it with the local propagation representation matrix according to the unit dimension, and inputs it into a fully connected neuron group. The output is a unit coupling representation matrix, ensuring that remote linkage information enters each unit's representation in the form of a plant-wide summary vector. The output layer is the sharing ratio generation layer. This layer outputs a sharing weight vector from the unit coupling representation matrix for each unit through a fully connected neuron group, and then normalizes the sharing weight vector proportionally to form an initial sharing ratio table. In the layered graph convolutional network, the fully connected neuron group consists of several neurons, with the number of neurons equal to the corresponding layer's output dimension. Non-linear activation is used between layers, while the sharing ratio generation layer uses linear output followed by ratio normalization.

[0072] The recursive calibration framework does not alter the generation path of the unit coupling characterization matrix in the hierarchical graph convolutional network. It receives the initial sharing ratio table, following deviation records, and correction coefficients, and outputs a correction sharing ratio table. The following deviation records, organized by unit dimension, document the deviation between the unit's regulation target and actual output changes. The recursive calibration framework includes a response consistency calculation unit: The response consistency calculation unit calculates the unit's following consistency vector based on the following deviation records; the dimension of the unit's following consistency vector is consistent with the number of units. It then uses the unit's following consistency vector to determine the following consistency with a following decision threshold and updates the correction coefficients under the constraint of the upper limit of the correction amplitude, obtaining the updated correction coefficients. The recursive calibration framework multiplies the initial sharing ratio table element-wise with the updated correction coefficients and performs proportional normalization to generate the correction sharing ratio table, whose dimension is consistent with the number of units.

[0073] This invention proposes an improved method for estimating the frequency regulation sharing ratio of multiple generating units. It replaces existing allocation logic based on equal distribution, fixed order, or single-index weighting by introducing coupling type identifiers on the edges of the unit coupling graph and then performing two-level fusion (local and plant-wide propagation) in a hierarchical graph convolutional network, grouping and converging according to coupling type. This design allows the network to explicitly retain the propagation semantics of different coupling sources when generating the initial sharing ratio, enabling it to separately express the effects of electrical coupling caused by parallel grid connection, resource competition caused by shared systems, and mutual interference of control channels. Furthermore, it incorporates strong local coupling and remote linkage into the same unit representation space through a nearest-neighbor-first-global propagation structure. Simultaneously, it performs proportional normalization on the output sharing weights, allowing the initial sharing ratio to be directly used for the share decomposition of plant-level frequency regulation demand and forming a structured generation path corresponding to the propagation links in the coupling graph, facilitating consistent alignment in online applications.

[0074] This invention proposes a novel recursive calibration framework as a cross-cycle correction channel that does not alter the generation path of the hierarchical graph convolutional network. It transforms the following deviation record of the previous frequency modulation cycle into a unit following consistency quantity, and after updating the correction coefficients by combining the following judgment threshold and the upper limit of the correction amplitude, it corrects and re-normalizes the initial sharing ratio. This framework characterizes the change in the actual following capability of the unit by the degree of consistency between the deviation direction and the deviation amplitude over time. Correction coefficients are updated only within the upper limit of the amplitude constraint when the consistency meets the threshold; otherwise, the update amount remains zero, thus avoiding the direct introduction of short-term noise or occasional deviations into the sharing ratio. The correction coefficients are output as a cross-cycle carry-over quantity and written back to the frequency modulation information table of the next cycle, enabling the sharing ratio to be recursively adjusted according to the drift of the unit's execution capability without recalculating the coupling graph or replacing the network output every cycle. This enhances the usability and interpretability boundary of the sharing ratio within continuous frequency modulation cycles.

[0075] This invention proposes a cluster frequency regulation decomposition and closed-loop recording method oriented towards actual operational constraints. Under the conditions of ramp rate and regulation margin constraints, the plant-level frequency regulation demand is decomposed according to the correction sharing ratio, generating a deployable unit regulation target table. When constraints result in some demand remaining unallocated, the unallocated amount is re-decomposed based on the set of units with remaining ramp rate and regulation margins, ensuring that the final target meets the unit adjustable resource boundary. After the regulation target is issued, the output at the start and end of the cycle is collected to form the actual output change and align it with the target change to calculate the deviation. The deviation record for this cycle is generated and written back along with the updated correction coefficient, providing direct input for the recursive calibration of the next cycle. This overall closed loop connects "coupled sensing ratio generation - recursive calibration based on follow-consistency - constraint-feasible target decomposition - deviation record writing back" into one, making the sharing ratio estimation and field execution results a sustainable online process.

[0076] Example 2 The reliable estimation system for frequency regulation sharing ratio of multi-unit clusters in thermal power plants according to the present invention includes: The acquisition module is used to acquire the frequency modulation information table; The generation module is used to construct a unit coupling diagram based on the coupling relationship between units in the frequency regulation information table. The edges of the unit coupling diagram carry coupling type identifiers. The unit coupling diagram and the real-time operating status of each unit are input into a hierarchical graph convolutional network, and an initial sharing ratio table is generated based on the output of the hierarchical graph convolutional network. The calibration module is used to input the initial sharing ratio table, the following deviation record, and the calibration coefficient into the recursive calibration framework. The recursive calibration framework makes a judgment based on the following deviation record computer group following consistency amount and the following judgment threshold. When the group following consistency amount is not less than the following judgment threshold, the calibration coefficient is updated under the upper limit constraint of the calibration range, and the updated calibration coefficient is used to correct the initial sharing ratio table to obtain the calibration sharing ratio table. When the group following consistency amount is less than the following judgment threshold, the calibration sharing ratio table is generated based on the calibration coefficient. The limiting module is used to decompose the frequency regulation demand based on the correction sharing ratio table and generate the unit regulation target table under the operating constraints of ramp rate limit and adjustment margin limit; The calculation module is used to issue the adjustment targets of each unit according to the unit adjustment target table, collect the actual output changes of each unit, calculate the deviation between the unit adjustment target table and the actual output changes, and generate the frequency regulation cycle follow-up deviation record. The output module is used to output the correction sharing ratio table, the unit regulation target table, the following deviation record for the current frequency regulation cycle, and the updated correction coefficient.

[0077] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in each embodiment of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0078] Example 3 A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of a reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant. For example, the method includes: acquiring a frequency regulation information table; constructing a unit coupling graph based on the inter-unit coupling relationships in the frequency regulation information table, wherein the edges of the unit coupling graph carry coupling type identifiers; inputting the unit coupling graph and the real-time operating status of each unit into a hierarchical graph convolutional network, and generating an initial sharing ratio table based on the output of the hierarchical graph convolutional network; inputting the initial sharing ratio table, follow-up deviation records, and correction coefficients into a recursive calibration framework, wherein the recursive calibration framework makes a determination based on the follow-up consistency quantity of the computer group according to the follow-up deviation records and a follow-up determination threshold. When the unit's following consistency is not less than the following judgment threshold, the correction coefficient is updated under the upper limit constraint of the correction amplitude, and the initial sharing ratio table is corrected using the updated correction coefficient to obtain the correction sharing ratio table; when the unit's following consistency is less than the following judgment threshold, the correction sharing ratio table is generated based on the correction coefficient; under the operating constraints of ramp rate limit and adjustment margin limit, the frequency regulation demand is decomposed according to the correction sharing ratio table to generate the unit regulation target table; the regulation targets of each unit are issued according to the unit regulation target table, and the actual output changes of each unit are collected. The deviation between the unit regulation target table and the actual output changes is calculated to generate the following deviation record for this frequency regulation cycle; the correction sharing ratio table, the unit regulation target table, the following deviation record for this frequency regulation cycle, and the updated correction coefficient are output. The memory may include main memory, such as high-speed random access memory (RAM), or non-volatile memory, such as at least one disk storage device. The processor, network interface, and memory are interconnected via an internal bus, which may be an industry-standard architecture bus, a peripheral component interconnection standard bus, or an extended industry-standard architecture bus. The bus can be categorized as an address bus, data bus, or control bus. The memory stores programs; specifically, the program may include program code, which includes computer operation instructions. The memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0079] Example 4 A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant. For example, the method includes: acquiring a frequency regulation information table; constructing a unit coupling graph based on the inter-unit coupling relationships in the frequency regulation information table, wherein the edges of the unit coupling graph carry coupling type identifiers; inputting the unit coupling graph and the real-time operating status of each unit into a hierarchical graph convolutional network, and generating an initial sharing ratio table based on the output of the hierarchical graph convolutional network; inputting the initial sharing ratio table, following deviation records, and correction coefficients into a recursive calibration framework, wherein the recursive calibration framework determines the unit following consistency based on the following deviation records and a following determination threshold; when the unit follows consistently... When the unit's consistency is not less than the following determination threshold, the correction coefficient is updated under the upper limit constraint of the correction amplitude, and the initial sharing ratio table is corrected using the updated correction coefficient to obtain the correction sharing ratio table. When the unit's consistency is less than the following determination threshold, the correction sharing ratio table is generated based on the correction coefficient. Under the operating constraints of ramp rate limit and adjustment margin limit, the frequency regulation demand is decomposed according to the correction sharing ratio table to generate the unit adjustment target table. The adjustment targets of each unit are issued according to the unit adjustment target table, and the actual output changes of each unit are collected. The deviation between the unit adjustment target table and the actual output changes is calculated to generate the following deviation record for the current frequency regulation cycle. The correction sharing ratio table, the unit adjustment target table, the following deviation record for the current frequency regulation cycle, and the updated correction coefficient are output. Specifically, the computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0080] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and disclosure of the invention. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0085] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

[0086] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant, characterized in that, include: 1) Obtain the frequency modulation information table; 2) Construct a unit coupling diagram based on the coupling relationship between units in the frequency regulation information table. The edges of the unit coupling diagram carry coupling type identifiers. Input the unit coupling diagram and the real-time operating status of each unit into a hierarchical graph convolutional network. Generate an initial sharing ratio table based on the output of the hierarchical graph convolutional network. 3) Input the initial sharing ratio table, following deviation record, and correction coefficient into the recursive calibration framework. The recursive calibration framework makes a judgment based on the following deviation record computer group following consistency amount and following judgment threshold. When the group following consistency amount is not less than the following judgment threshold, the correction coefficient is updated under the upper limit constraint of correction amplitude, and the updated correction coefficient is used to correct the initial sharing ratio table to obtain the correction sharing ratio table. When the group following consistency amount is less than the following judgment threshold, the correction sharing ratio table is generated based on the correction coefficient. 4) Under the operating constraints of ramp rate limit and adjustment margin limit, decompose the frequency regulation demand according to the correction sharing ratio table and generate the unit adjustment target table; 5) Issue adjustment targets to each unit according to the unit adjustment target table, collect the actual output changes of each unit, calculate the deviation between the unit adjustment target table and the actual output changes, and generate a frequency regulation cycle follow-up deviation record. 6) Output the correction sharing ratio table, the unit regulation target table, the following deviation record for this frequency regulation cycle, and the updated correction coefficient.

2. The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant according to claim 1, characterized in that, The specific operation of step 1) is as follows: Obtain the topology and control connection information of thermal power plant units, extract the coupling relationships between units, write a coupling type identifier for each coupling relationship, and organize them into a coupling type adjacency table according to the coupling type identifier; Collect the real-time operating status of each unit, and assemble the unit operating status feature matrix according to the unit index. The operating status features include the unit's current output, adjustment margin, ramp-up margin, start-up and shutdown status, and aging degree. Obtain the frequency regulation demand and operating constraints, and parse the operating constraints into ramp rate limits and regulation margin limits for each unit; Read the following deviation record and correction coefficient of the previous frequency regulation cycle, and merge the coupling type adjacency table, unit operating status characteristic matrix, frequency regulation demand, operating constraints, following deviation record of the previous frequency regulation cycle and correction coefficient to generate a frequency regulation information table.

3. The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant according to claim 1, characterized in that, The unit coupling graph and the unit operating state feature matrix are input into a hierarchical graph convolutional network. In the coupling type convolutional layer of the hierarchical graph convolutional network, for each coupling type identifier, the corresponding convolutional neuron group is called. The unit operating state feature matrices of adjacent units are aggregated according to the coupling type adjacency list and mapped to a unified dimension via fully connected neurons to obtain the type influence matrix. In the type fusion layer of the hierarchical graph convolutional network, the influence matrices of various types corresponding to the same unit are concatenated in a fixed order and input into fully connected neurons, outputting the unit local representation matrix. Local propagation in the hierarchical graph convolutional network... In the layer, based on the unit coupling graph, the local representation matrix of the unit is subjected to adjacency propagation and transformed by a fully connected neuron group to obtain the local propagation representation matrix. In the whole plant propagation layer, the whole plant summary vector is calculated on the local propagation representation matrix. The whole plant summary vector and the local propagation representation matrix are concatenated according to the unit dimension and then input into the fully connected neuron group to output the unit coupling representation matrix. In the sharing ratio generation layer of the layered graph convolutional network, the sharing weight vector is output for each unit through the fully connected neuron group. The sharing weight vector is then normalized to obtain the initial sharing ratio table.

4. The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant according to claim 3, characterized in that, Type convergence vector Obtained by the following formula: in, Indicates the unit In coupling type identifier The type convergence vector below; This indicates an adjacency list of coupling types. The given unit In coupling type identifier The set of adjacent unit indexes; and Indicates the unit index; Indicates the unit The start / stop state components; Indicates the unit In coupling type identifier The type degree and take The number of elements; Indicates the unit In coupling type identifier The type degree below; Indicates the unit The unit operating state feature vector is taken from the unit operating state feature matrix. The corresponding row vector; when When the value is zero, set the denominator to one and merge the type vector. Input a fully connected neuron group and press and Mapping to a unified dimension yields a vector of type influence. Stack all the type influence vectors of all units according to their unit indices to form a type influence matrix. .

5. The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant according to claim 1, characterized in that, The specific operation of step 3) is as follows: Input the initial sharing ratio table, the follow-up deviation record of the previous frequency modulation cycle and the correction coefficient into the recursive calibration framework, and obtain the follow-up judgment threshold and the upper limit of the correction amplitude. In the response consistency calculation unit, the deviation direction and deviation magnitude information between the unit adjustment target and the actual output change in the following deviation record are read according to the unit dimension, and the unit following consistency vector is generated. The unit following consistency vector is compared with the following judgment threshold for each unit to obtain the unit following judgment result. When the unit follow-up judgment result is that the unit follow-up consistency amount is not less than the follow-up judgment threshold, the correction coefficient update amount is generated according to the follow-up deviation record of the corresponding unit, and the correction coefficient update amount is limited to the upper limit of the correction range. The correction coefficient is updated unit by unit by using the correction coefficient update amount to obtain the updated correction coefficient. When the unit follow-up judgment result is that the unit follow-up consistency amount is less than the follow-up judgment threshold, the correction coefficient update amount of the corresponding unit is set to zero and the correction coefficient remains unchanged. The corrected sharing weight vector is obtained by multiplying the initial sharing ratio table element by element with the updated correction coefficients according to the unit dimension. The corrected weighting vector is normalized proportionally to obtain the corrected weighting ratio table.

6. The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant according to claim 5, characterized in that, Unit following consistency quantity for: in, Indicates the unit The unit follows the consistency quantity; This represents the directional consistency weighting coefficient; Indicates the number of sampling intervals; Indicates the sampling interval index; Indicates the unit In the The directional component of the sampling interval deviation; Indicates the unit In the The amplitude component of the sampling interval deviation; This represents the sampling interval index used to calculate the mean of the deviation amplitude components; This indicates a positive constant that avoids a denominator of zero; Indicates exponentiation; This represents absolute value operations.

7. The reliable estimation method for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant according to claim 1, characterized in that, The specific operation of step 5) is as follows: Read the unit regulation target from the unit regulation target table one by one, and send the unit regulation target to the frequency regulation control interface of the corresponding unit; At the beginning of this frequency regulation cycle, the current output of each unit is collected, and at the end of this frequency regulation cycle, the final output of each unit is collected, and the actual output change of each unit is calculated. The unit regulation targets in the unit regulation target table are converted into target output changes. The target output changes are then aligned with the actual output changes according to the unit index, and the deviation is calculated for each unit to obtain the deviation direction information and deviation magnitude information. The deviation direction information and deviation magnitude information are assembled according to the unit dimension to generate cost frequency regulation cycle following deviation records.

8. A reliable estimation system for the frequency regulation sharing ratio of a multi-unit cluster in a thermal power plant, characterized in that, include: The acquisition module is used to acquire the frequency modulation information table; The generation module is used to construct a unit coupling diagram based on the coupling relationship between units in the frequency regulation information table. The edges of the unit coupling diagram carry coupling type identifiers. The unit coupling diagram and the real-time operating status of each unit are input into a hierarchical graph convolutional network, and an initial sharing ratio table is generated based on the output of the hierarchical graph convolutional network. The calibration module is used to input the initial sharing ratio table, the following deviation record, and the calibration coefficient into the recursive calibration framework. The recursive calibration framework makes a judgment based on the following deviation record computer group following consistency amount and the following judgment threshold. When the group following consistency amount is not less than the following judgment threshold, the calibration coefficient is updated under the upper limit constraint of the calibration range, and the updated calibration coefficient is used to correct the initial sharing ratio table to obtain the calibration sharing ratio table. When the group following consistency amount is less than the following judgment threshold, the calibration sharing ratio table is generated based on the calibration coefficient. The limiting module is used to decompose the frequency regulation demand based on the correction sharing ratio table and generate the unit regulation target table under the operating constraints of ramp rate limit and adjustment margin limit; The calculation module is used to issue the adjustment targets of each unit according to the unit adjustment target table, collect the actual output changes of each unit, calculate the deviation between the unit adjustment target table and the actual output changes, and generate the frequency regulation cycle follow-up deviation record. The output module is used to output the correction sharing ratio table, the unit regulation target table, the following deviation record for the current frequency regulation cycle, and the updated correction coefficient.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the reliable estimation method for frequency regulation sharing ratio of multi-unit clusters in thermal power plants as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the reliable estimation method for frequency regulation sharing ratio of multi-unit clusters in thermal power plants as described in any one of claims 1-7.

Images