Power plant distributed frequency modulation optimization method based on variational inference and related device

Abstract

The application discloses a thermal power plant distributed frequency modulation optimization method based on variational inference and a related device, belongs to the field of thermal power unit primary frequency modulation / AGC collaborative control and intelligent inference of electric power systems, and comprises the following steps: determining the weighting coefficient of the reference table correction amount of each unit based on a consistent working condition threshold, updating a shared frequency modulation response reference table, obtaining an updated shared frequency modulation response reference table, and sending the updated shared frequency modulation response reference table to each unit; each unit obtains a frequency modulation power increment prediction value according to the shared frequency modulation response reference table, unit difference parameters and frequency modulation power increment candidates, judges the frequency modulation power increment prediction value under the unit operation constraint, and generates a frequency modulation power increment instruction; and the control system of the unit adjusts the unit output according to the frequency modulation power increment instruction, so that the method and the related device can improve the distributed frequency modulation optimization effect of the thermal power plant.

Description

Technical Field

[0001] This invention belongs to the field of primary frequency regulation / AGC collaborative control and intelligent inference of thermal power units in power systems, and relates to a distributed frequency regulation optimization method and related devices for thermal power plants based on variational inference. Background Technology

[0002] In power system operation, frequency deviations need to be suppressed through primary frequency regulation and automatic generation control. Thermal power plants typically use multiple units participating in frequency regulation in parallel to provide adjustable active power. Due to the significant coupling between the turbine-speed control system and the boiler-thermal system, the frequency regulation response of the units varies under different loads, valve positions, main steam pressures, reheat temperatures, etc. Furthermore, the safety limitations of the primary equipment and the control adjustment ranges of each unit differ, resulting in variations in available regulation capacity and execution characteristics. The frequency regulation control link is also affected by measurement point noise, time delays, and operating condition switching; the same grid disturbance exhibits inconsistent measurement point characteristics on different unit sides. At the thermal power plant level, frequency regulation actions need to rapidly generate executable power increment commands within the control cycle and ensure compatibility with unit operating constraints. Simultaneously, it requires consistency in the meaning of commands under the same frequency regulation condition among multiple units to facilitate coordinated allocation and unified acceptance. In engineering practice, the stable data that can be obtained from the unit side is usually real-time operating measurement data and known operating constraint boundaries. However, the centralized aggregation of raw measurement data for unified modeling is subject to limitations such as network bandwidth, data interface, compliance and security. Therefore, it is necessary to form a frequency regulation response characterization mechanism that can align the meaning of operating conditions across units and can self-correct as operation progresses, based solely on the available data and constraints.

[0003] In existing technologies, frequency regulation of multiple units in thermal power plants generally adopts an allocation strategy based on factors such as frequency regulation participation status, available up and down adjustment margins, and ramp-up capabilities. The plant-level or dispatching side calculates the power increment of each unit and issues it for execution. In some scenarios, offline tuning response curves or rule bases are combined to provide empirical adjustment amounts under typical operating conditions. In recent years, some solutions have also used historical measurement points and execution results for data-driven modeling, training regression models on a centralized platform and then deploying them to the unit side to output frequency regulation power increments. Other solutions attempt to train local models on the unit side and aggregate parameters through a federation approach, aiming to obtain a unified model without aggregating raw data, for use in outputting frequency regulation power increments of each unit or for online correction.

[0004] The existing solutions described above are prone to misalignment in meaning when there are differences in operating conditions and inconsistent data distribution among multiple units, leading to difficulties in aligning the frequency regulation power increments output by each unit under the same disturbance. This affects coordinated allocation and unified judgment within the plant. Centralized modeling typically relies on aggregating local real-time operating measurement data, which is subject to data access and security boundary constraints during engineering implementation. Federated aggregation, which mainly uses model weight averaging, lacks explicit constraints on "consistent operating condition meaning" and evidence consistency judgments. This can easily cause semantic drift in shared knowledge during iteration, which can further propagate to the instruction generation and execution acceptance stages, resulting in poor distributed frequency regulation optimization performance in thermal power plants. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and related apparatus for distributed frequency regulation optimization of thermal power plants based on variational inference. This method and related apparatus can improve the effect of distributed frequency regulation optimization in thermal power plants.

[0006] To achieve the above objectives, this invention discloses a distributed frequency regulation optimization method for thermal power plants based on variational inference, comprising: 1) Obtain the set of units participating in frequency regulation in the thermal power plant, the unit operation constraints and operating condition consistency thresholds of each unit, initialize a unified shared frequency regulation response reference table across units in the parameter server, and send it to each unit; 2) Each unit acquires local real-time operating measurement data. Under the constraints of unit operation, the local real-time operating measurement data is matched with the operating conditions in conjunction with the shared frequency regulation response reference table to generate a characterization of the unit's frequency regulation operating condition mode. 3) Based on the frequency regulation operating mode characterization of the unit, each unit calls the operating mode-shared frequency regulation response reference table decomposition network to determine the frequency regulation operating mode identifier. Then, under the constraints of the shared frequency regulation response reference table, the local real-time operating measurement point data is decomposed into the reference response component and unit difference parameter corresponding to the frequency regulation operating mode identifier. The frequency regulation power increment candidate is calculated according to the reference response component and the unit difference parameter, and the shared frequency regulation response reference table correction information is generated based on the deviation between the reference response component and the shared frequency regulation response reference table. 4) The parameter server receives the correction information of the shared frequency regulation response reference table from each unit, and uses consensus-based distributed variational inference to merge the correction information of the shared frequency regulation response reference table into a reference table correction amount with consistent operating conditions; based on the operating condition consistency threshold, it determines the weighting coefficient of the reference table correction amount of each unit and updates the shared frequency regulation response reference table to obtain the updated shared frequency regulation response reference table, and sends the updated shared frequency regulation response reference table to each unit. 5) Each unit obtains the predicted value of frequency regulation power increment based on the shared frequency regulation response reference table, unit difference parameters and frequency regulation power increment candidates, and judges the predicted value of frequency regulation power increment under unit operation constraints to generate frequency regulation power increment command; 6) The unit's control system adjusts the unit's output according to the frequency regulation power increment command.

[0007] Furthermore, the specific operation of step 1) is as follows: Obtain the set of generating units participating in frequency regulation and assign unit identifiers to the set of generating units. Establish a consistent mapping between the unit identifiers and the 64-dimensional measurement point sequence of the feature vector of the local real-time operating measurement point data. Collect the primary equipment safety limits and controller adjustment ranges of each unit, encode the limits and ranges into a ten-dimensional unit operation constraint feature vector, and configure the same threshold for the same operating condition for the unit set. In the parameter server, a shared frequency modulation response reference table is constructed. The shared frequency modulation response reference table is set as an indexable and readable reference library, and eight types of frequency modulation mode identifiers are defined, each corresponding to a 16-dimensional reference response component. The values ​​and allowable correction ranges of each type of reference response component are initialized. The shared frequency regulation response reference table, along with the frequency regulation operating condition mode identifier, the unit operation constraint feature vector encoding rules, and the operating condition consistency threshold, are sent to each unit for subsequent operating condition matching and reference table correction.

[0008] Furthermore, the specific operation of step 2) is as follows: Each unit collects local real-time operation measurement point data at the unit side and assembles it into a 64-dimensional local real-time operation measurement point data feature vector according to the preset measurement point order; Read the ten-dimensional unit operation constraint feature vector, perform constraint screening on the local real-time operation measurement point data feature vector, adjust the components that exceed the allowable range of unit operation constraints to the allowable range, and obtain the local real-time operation measurement point data feature vector after constraint screening; Based on the feature vector of local real-time operating measurement point data after constraint screening, the reference response components corresponding to the eight types of frequency regulation mode identifiers in the shared frequency regulation response reference table are called in sequence, the matching degree with each reference response component is calculated, and an eight-dimensional unit frequency regulation mode representation is formed. Based on the frequency regulation mode characterization of the eight-dimensional unit, candidate frequency regulation mode identifiers are determined, and the frequency regulation mode characterization of the eight-dimensional unit and candidate frequency regulation mode identifiers are output.

[0009] Furthermore, the process of fusing the shared frequency modulation response reference table correction information into a reference table correction quantity with consistent operating condition meaning using consensus-based distributed variational inference is as follows: Consensus-based distributed variational inference is used to fuse the correction information of the shared frequency modulation response reference table to obtain the correction amount of the candidate reference table. Correction amount for candidate reference table By performing boundary constraints and direction verification, a reference table correction amount with consistent working condition meaning is obtained. .

[0010] Furthermore, the candidate reference table correction amount for:

[0011] in, The frequency modulation operating mode is identified as follows: The revised evidence set; This is a sixteen-dimensional reference deviation vector; A sixteen-dimensional vector that allows for correction range; The sixteen positive direction vectors are given, with each dimension taking the values ​​of 1, 0, and negative 1. This is the sixteen-dimensional evidence mean vector obtained under the constraints of the allowed correction range and correction direction; This is a sixteen-dimensional evidence precision vector constructed from the allowable correction range vector; For operating condition consistency threshold The resulting weighting coefficients; It is a sixteen-dimensional prior precision vector and is a fixed configuration parameter of the parameter server; A sixteen-dimensional fusion precision vector; This is the correction amount for the sixteen-dimensional candidate reference table; The parameter is a positive lower bound. This is an element-wise indicator function; it returns 1 if the condition is true and 0 otherwise. and For the operators of taking the smallest element and taking the largest element; This is element-wise multiplication; It is the element-wise reciprocal vector.

[0012] Furthermore, the boundary constraints include a shared frequency modulation response reference table. Preset allowable correction range ,in, It is a sixteen-dimensional lower bound vector. It is a sixteen-dimensional upper bound vector.

[0013] Furthermore, frequency modulation power increment command for

[0014] in, This is a ten-dimensional characteristic vector representing the operational constraints of the generating unit. This represents the lower limit of the unit's output. This is the upper limit of the unit's output. The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. The lower limit of the main valve opening. The upper limit of the main valve opening. The lower limit of the main steam pressure. The upper limit of the main steam pressure. This is the lower limit of the reheat temperature. This is the upper limit of the reheat temperature; This represents the real-time value of the unit's current active power. This is the real-time value of the main steam valve opening. This is the real-time value of the main steam pressure. This is the real-time value of the reheat temperature; For the constraint transformation coefficient vector, , , These are the coefficients that convert the main steam valve opening margin, main steam pressure margin, and reheat temperature margin into the usable power increment limit, respectively. To control cycle duration; For one-dimensional frequency modulation power increment prediction; This refers to the frequency modulation power increment command issued in the previous control cycle; and These are the upper and lower bounds of the allowed interval, respectively. and These are the lower and upper bounds of the executable instruction range after intersecting the allowed range and the rate of change boundary, respectively; This is a frequency modulation power increment command; To find the minimum value operator, This is the operator for finding the maximum value.

[0015] This invention discloses a distributed frequency regulation optimization system for thermal power plants based on variational inference, comprising: The acquisition module is used to acquire the set of units participating in frequency regulation in the thermal power plant, the unit operation constraints and operating condition consistency thresholds of each unit, initialize a unified shared frequency regulation response reference table across units in the parameter server, and send it to each unit. The matching module is used by each unit to acquire local real-time operating measurement data. Under the constraints of unit operation, it performs operating condition matching on the local real-time operating measurement data in conjunction with the shared frequency regulation response reference table to generate a characterization of the unit's frequency regulation operating condition mode. The calculation module is used by each unit to decompose the network based on the unit frequency regulation operating mode characterization, call the operating mode-shared frequency regulation response reference table, determine the frequency regulation operating mode identifier, and then decompose the local real-time operating measurement point data into the reference response component and unit difference parameter corresponding to the frequency regulation operating mode identifier under the constraints of the shared frequency regulation response reference table; calculate the frequency regulation power increment candidate based on the reference response component and the unit difference parameter, and generate shared frequency regulation response reference table correction information based on the deviation between the reference response component and the shared frequency regulation response reference table; The update module is used by the parameter server to receive the correction information of the shared frequency regulation response reference table of each unit, and to use consensus-based distributed variational inference to merge the correction information of the shared frequency regulation response reference table into a reference table correction amount with consistent operating conditions; based on the operating condition consistency threshold, the weighting coefficient of the reference table correction amount of each unit is determined and the shared frequency regulation response reference table is updated to obtain the updated shared frequency regulation response reference table, and the updated shared frequency regulation response reference table is sent to each unit; The determination module is used for each unit to obtain the predicted value of frequency regulation power increment based on the shared frequency regulation response reference table, unit difference parameters and frequency regulation power increment candidates, and to determine the predicted value of frequency regulation power increment under the unit operation constraints, and generate frequency regulation power increment command. The control module is used by the unit's control system to adjust the unit's output according to the frequency regulation power increment command.

[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 distributed frequency regulation optimization method for thermal power plants based on variational inference.

[0017] This invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the variational inference-based distributed frequency regulation optimization method for thermal power plants.

[0018] The present invention has the following beneficial effects: In practical operation, the distributed frequency regulation optimization method and related devices for thermal power plants based on variational inference described in this invention establish a decomposable stochastic representation of operating condition mode identifier, reference response component, and unit difference parameters: On the unit side, reference response components with the same name and meaning across units are obtained by indexing the shared reference table according to the operating condition identifier, and the unit difference parameters are estimated to generate candidate frequency regulation power increments. Simultaneously, the reference deviation and its allowable correction range / direction are output as reference table correction information. The parameter server merges and corrects evidence according to the operating condition identifier, uses consensus-based distributed variational inference, and combines it with a weighted fusion based on the operating condition consistency threshold to update the shared reference table. On the unit side, predictions are recalculated based on the updated reference table, and executable instructions are generated under operating constraints. This maintains consistency in the meaning of operating conditions without uploading local measurement point data, improves the consistency and deliverability of frequency regulation instructions across multiple units, and ultimately improves the distributed frequency regulation optimization effect of thermal power plants. 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 6 The 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 distributed frequency regulation optimization method for thermal power plants based on variational inference described in this invention includes the following steps: 1) Obtain the set of units participating in frequency regulation in the thermal power plant, the unit operation constraints and operating condition consistency thresholds of each unit, initialize a unified shared frequency regulation response reference table across units in the parameter server, and send it to each unit; It should be noted that the reference Figure 2 The process involves: acquiring the set of generating units participating in frequency regulation and assigning unit identifiers to the set; establishing a consistent mapping between the unit identifiers and the 64-dimensional measurement point sequence of the feature vector of local real-time operating measurement points; collecting the primary equipment safety limits and controller adjustment ranges of each unit; encoding the limits and ranges into a 10-dimensional unit operation constraint feature vector; configuring a consistent threshold for the same operating condition for the set of generating units; constructing a shared frequency regulation response reference table in the parameter server; setting the shared frequency regulation response reference table as an indexable and readable reference library; defining eight types of frequency regulation operating condition mode identifiers and their corresponding 16-dimensional reference response components; initializing the values ​​and allowable correction ranges of each type of reference response component; and sending the shared frequency regulation response reference table, along with the frequency regulation operating condition mode identifiers, the defined unit operation constraint feature vector encoding rules, and the consistent threshold for the operating condition to each unit for subsequent operating condition matching and reference table correction.

[0030] The specific operation of step 1) is as follows: During the parameter server deployment phase, the parameter server obtains the set of generating units in the thermal power plant participating in frequency regulation. ,in, To determine the number of generating units participating in frequency regulation, the parameter server assigns a unique unit identifier to each unit. And establish a measurement point address vector for each unit. ,in, The first in the unit-side data acquisition system The address or point number of each measurement point. The parameter server also defines a unified semantic order table for measurement points. ,in, For the first The Chinese meaning of measurement points, and This applies consistently to all units; the unit side then adjusts the collected values ​​accordingly. The sequence is assembled into a 64-dimensional local real-time running measurement point data feature vector. The aforementioned The characteristics are grouped by function as follows: The grid and frequency regulation request characteristic group has eight dimensions, including frequency deviation, active power command deviation, frequency regulation request direction, frequency regulation request amplitude, frequency regulation request change rate, unit participation status, unit current ramp margin, and unit current downhill margin; the turbine regulation and execution characteristic group has twenty-four dimensions, including unit active power, unit active power change rate, governor setpoint, governor output, main steam valve opening, governor valve opening change rate, speed deviation, primary frequency regulation activation flag, and valve position. The measurement points include: limit amount, valve position tracking error, regulation dead zone, regulation hysteresis, reheat valve related measurement points, bypass related measurement points, and actuator feedback related measurement points; the boiler and thermal state characteristic group consists of 32 dimensions, including main steam pressure, main steam temperature, reheat pressure, reheat temperature, drum water level, feedwater flow rate, fuel quantity, forced draft volume, induced draft volume, oxygen quantity, furnace negative pressure, superheater metal temperature related measurement points, economizer outlet temperature related measurement points, condenser vacuum, circulating water temperature difference related measurement points, and plant power consumption related measurement points. and The combination ensures that different units express the same measurement point meaning at the same dimensional location, supporting the common representation layer of the operating condition mode—shared frequency regulation response reference table decomposition network. To ensure consistent input representation across different units.

[0031] The parameter server collects the primary equipment safety limits and controller adjustment ranges for each unit, and directly encodes the collected values ​​into a ten-dimensional unit operation constraint feature vector. And it is agreed that the meaning of each dimension is fixed and unchanging: This represents the lower limit of the unit's output. This is the upper limit of the unit's output. The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. The lower limit of the main valve opening. The upper limit of the main valve opening. The lower limit of the main steam pressure. The upper limit of the main steam pressure. This is the lower limit of the reheat temperature. This represents the upper limit of the reheat temperature. The parameter server configures consistent threshold values ​​for the same operating condition for the entire unit set. , The threshold benchmark is used to determine the consistency of the shared frequency regulation response reference table correction information uploaded by multiple units under the same frequency regulation mode identifier.

[0032] The parameter server constructs a unified shared frequency modulation response reference table across units. Shared frequency response reference table Set up an indexable and readable reference library and define eight types of frequency modulation mode identifiers. ,in, The value ranges from one to eight. The parameter server is for each class. Initialize a sixteen-dimensional reference response component And organize all reference response components into a reference response matrix. Reference response components The initial values ​​are provided by the reference template library built into the parameter server. This reference template library is offline configuration data, providing the initial shapes of the sixteen-dimensional reference response components according to eight frequency modulation mode identifiers. The parameter server also provides initial shapes for each type... Initialize allowed correction range ,Will Represented as a sixteen-dimensional lower bound vector With the sixteen-dimensional upper bound vector Boundary pairs Used to define the reference response component Adjustable range during fusion correction.

[0033] The parameter server will share the frequency modulation response reference table. Together with the frequency modulation mode identifier definition and the measurement point semantic sequence table Ten-dimensional unit operation constraint feature vector The coding rules are consistent with the threshold of the working conditions. Send to each unit; the units will then receive and... Loaded as an indexable and readable reference library, and with Generate 64-dimensional local real-time measurement point data feature vectors in a prescribed order. ,by The prescribed sequence generates a ten-dimensional unit operation constraint feature vector, providing a consistent data interface for generating the unit frequency regulation operating mode representation for operating condition matching, and for decomposing the operating mode-shared frequency regulation response reference table decomposition network according to the frequency regulation operating mode identifier index reference response components and generating shared frequency regulation response reference table correction information. Details of the frequency regulation response system data objects are shown in Table 1: Table 1

[0034] 2) Each unit acquires local real-time operating measurement data. Under the constraints of unit operation, the local real-time operating measurement data is matched with the operating conditions in conjunction with the shared frequency regulation response reference table to generate a characterization of the unit's frequency regulation operating condition mode. It should be noted that the reference Figure 3Each unit collects local real-time operation measurement point data at the unit side and assembles it into a 64-dimensional local real-time operation measurement point data feature vector according to a preset measurement point order. The 10-dimensional unit operation constraint feature vector is read, and constraint screening is performed on the local real-time operation measurement point data feature vector. Components exceeding the allowable range of unit operation constraints are adjusted to the allowable range, resulting in a constraint-screened local real-time operation measurement point data feature vector. Based on the constraint-screened local real-time operation measurement point data feature vector, the reference response components corresponding to the eight types of frequency regulation operating condition mode identifiers in the shared frequency regulation response reference table are sequentially called. The matching degree with each reference response component is calculated to form an 8-dimensional unit frequency regulation operating condition mode representation. Candidate frequency regulation operating condition mode identifiers are determined based on the 8-dimensional unit frequency regulation operating condition mode representation, and the 8-dimensional unit frequency regulation operating condition mode representation along with the candidate frequency regulation operating condition mode identifiers are output for subsequent operating condition mode—shared frequency regulation response reference table decomposition network operating condition entry control.

[0035] The specific operation of step 2) is as follows: In each control cycle, the unit reads local real-time operating measurement data from the acquisition system and performs the data according to the semantic order table of the measurement points. The collected values ​​are assembled into a 64-dimensional local real-time measurement point data feature vector according to fixed-dimensional positions. ,in, for The Middle Each measuring point represents a corresponding real-time value. To establish a definite correspondence between unit operation constraints and measuring point dimensions, a constraint association index vector is defined. and agreed The active power of the unit. The rate of change of the unit's active power. Main valve opening. Main steam pressure This is the reheat temperature.

[0036] Unit reading ten-dimensional unit operation constraint feature vector ,in, This represents the lower limit of the unit's output. This is the upper limit of the unit's output. The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. The lower limit of the main valve opening. The upper limit of the main valve opening. The lower limit of the main steam pressure. The upper limit of the main steam pressure. This is the lower limit of the reheat temperature. This is the upper limit of the reheat temperature. (For) When performing constraint screening, only for The corresponding measurement point components are adjusted at the boundary to form a feature vector of local real-time operating measurement point data after constraint screening. :Will Crop to range ,get ,Will Crop to range ,get ,Will Crop to range ,get ,Will Crop to range ,get ,Will Crop to range ,get The remaining dimensions satisfy This constraint screening ensures that the key measuring points entering the operating condition matching meet the safety limits of the primary equipment and the adjustment range of the controller.

[0037] For reference to the shared frequency response table Each type of frequency modulation operating mode is identified by its corresponding identifier. The corresponding sixteen-dimensional reference response components To create comparable objects, the unit side from Extracting aligned measurement point sub-vectors for working condition matching Alignment of measurement points is determined by the alignment index vector. Sure, according to Sequence from Values ​​are taken to ensure The Wei and The "Dimension" has the same meaning. To reduce the impact of dimensional differences on the matching degree, the unit side... Standardization is performed dimension by dimension, and the standardization parameters are composed of a sixteen-dimensional bias vector. With a sixteen-dimensional scale vector Given, Used to describe the reference center value of each measurement point in each dimension. The scaling factor used to describe the measurement points in each dimension, and and Reference table for shared frequency modulation response from parameter server The data is issued together and fixed on the unit side. Standardized operations involve subtracting the corresponding value from each dimension first. Get the centralized value, then multiply by the corresponding Obtain the scaling value and crop the scaling value to the range. This forms a standardized alignment subvector of measurement points. .

[0038] The generating units sequentially invoke the shared frequency regulation response reference table. Category 8 FM mode identification Corresponding reference response component For each category Calculate matching degree And form an eight-dimensional frequency regulation mode characterization of the unit. The matching degree calculation adopts a step-by-step bias weighted aggregation process: first calculate and The dimension-wise absolute deviations form a sixteen-dimensional deviation vector. Then introduce a sixteen-dimensional weight vector. For the deviation vector By performing a dimension-wise weighted summation, we obtain the weighted bias scalar. ,in, Reference table for shared frequency modulation response from parameter server The data was distributed and fixed on the unit side; subsequently, threshold mapping was used to... Convert to The threshold mapping consists of two scalar thresholds. and definition, and Parameters are issued by the parameter server and satisfy the following conditions: Less than ,when Not greater than season Take one, when Not less than season Take zero, when In and Press between Relative position within this interval The value is taken in a linearly decreasing manner. Using the above method, The document provides an expression for the attribution strength of the current frequency modulation (FM) condition to the eight types of FM mode identifiers, and... The calculation process only depends on , With the issued , , .

[0039] The unit side is characterized by the eight-dimensional unit frequency regulation mode. Determine the candidate frequency modulation mode identifier ,Will Defined as The frequency modulation mode identifier corresponding to the largest component; when multiple largest components exist, the frequency modulation mode identifier with the smallest value is selected as the maximum value. The unit side will and The operating condition mode—shared frequency regulation response reference table decomposition network—is provided as the operating condition input control variable, wherein... Reference table for shared frequency modulation response used for index reading The corresponding reference response component, This is used to trigger a consistency check on the frequency modulation mode identifier output by the operating condition identification branch.

[0040] 3) Based on the frequency regulation operating mode characterization of the unit, each unit calls the operating mode-shared frequency regulation response reference table decomposition network to determine the frequency regulation operating mode identifier. Then, under the constraints of the shared frequency regulation response reference table, the local real-time operating measurement point data is decomposed into the reference response component and unit difference parameter corresponding to the frequency regulation operating mode identifier. The frequency regulation power increment candidate is calculated according to the reference response component and the unit difference parameter, and the shared frequency regulation response reference table correction information is generated based on the deviation between the reference response component and the shared frequency regulation response reference table. It should be noted that the reference Figure 4The process involves using a 64-dimensional local real-time operating measurement point data feature vector and a 10-dimensional unit operation constraint feature vector as inputs to call the operating condition mode—shared frequency regulation response reference table decompose network, and using the shared frequency regulation response reference table as an indexable reference library; inputting the 64-dimensional local real-time operating measurement point data feature vector into the common representation layer, and mapping it through three fully connected layers to obtain a 32-dimensional measurement point representation vector, where the number of neurons in the three fully connected layers is 128, 64, and 32 respectively; inputting the 32-dimensional measurement point representation vector into the operating condition identification branch, and outputting an 8-dimensional unit frequency regulation operating condition mode representation through a fully connected layer with 32 neurons, and determining the frequency regulation operating condition mode identifier based on the 8-dimensional unit frequency regulation operating condition mode representation; using the frequency regulation operating condition mode identifier as an index to read the corresponding 16-dimensional reference response component from the shared frequency regulation response reference table, and inputting the reference response component as a cross-unit synonymous reference part; and finally, using the 32-dimensional measurement point representation vector and the 10-dimensional unit operation constraint feature vector as inputs to the shared frequency regulation response reference table. The row constraint feature vectors are concatenated to form a 42-dimensional constraint fusion vector, which is then input into the difference characterization branch. This vector is then processed through two fully connected layers to obtain an 8-dimensional unit difference parameter. The number of neurons in the two fully connected layers is 32 and 8 respectively. The unit difference parameter is limited to participating only in the generation of local frequency regulation power increment and not in the update of the shared frequency regulation response reference table. The 16-dimensional reference response component and the 8-dimensional unit difference parameter are input into the decomposition and synthesis output head. This output head is first processed through a fully connected layer with 16 neurons to form a 24-dimensional frequency regulation synthesis representation, and then through a fully connected layer with one neuron to output a 1-dimensional frequency regulation power increment candidate. The 32-dimensional measurement point representation vector and the 16-dimensional reference response component are input into the reference correction output head. This output head is processed through a fully connected layer with 16 neurons to output a 16-dimensional reference deviation vector. Based on the 10-dimensional unit operation constraint feature vector, the reference deviation vector is converted into an allowable correction range and correction direction. This vector, together with the frequency regulation mode identifier, forms the shared frequency regulation response reference table correction information.

[0041] The specific operation of step 3) is as follows: Unit deployment operating mode - shared frequency regulation response reference table decomposition network, and 64-dimensional local real-time running measurement point data feature vector in each control cycle. With the ten-dimensional unit operation constraint feature vector Perform forward computation on the input, where the subscript This is for unit identification. The unit side will share the frequency regulation response reference table. Loaded as an indexable reference library as a reference response matrix ,in Identified by frequency modulation mode The eight rows of reference response components are stored in sequence. Each row has sixteen dimensions. The operating mode-shared frequency regulation response reference table decomposition network consists of a common representation layer, an operating condition identification branch, a difference characterization branch, a decomposition and synthesis output head, and a reference correction output head. The parameters of each layer are fixed on the unit side in the form of weight matrices and bias vectors.

[0042] The public representation layer will generate a 64-dimensional local real-time running feature vector of measurement point data. The input is sequentially processed through three fully connected layers, performing three linear mappings and nonlinear activations. The first layer outputs a dimension of 128, the second layer outputs a dimension of 64, and the third layer outputs a dimension of 32, resulting in a 32-dimensional measurement point representation vector. Operating condition identification branch reception It also characterizes the unit's frequency regulation mode by outputting an eight-dimensional mode through a fully connected layer containing thirty-two neurons. ,Will The category index corresponding to the largest component is determined as the frequency modulation operating mode identifier. When the frequency regulation mode of the eight-dimensional unit is characterized With candidate frequency modulation mode identifier When available, the unit side calculates the consistency score. And when consistency is insufficient, the frequency modulation operating mode will be marked. Replace with candidate frequency modulation mode identifier Consistency score The calculation method is to use Dividing each component by the sum of its absolute values ​​yields the normalized vector. Normalize in the same way, and calculate the sum of the products of corresponding components of the two normalized vectors to obtain... and the consistency threshold As a fixed configuration parameter, when Less than The replacement will be performed at that time.

[0043] Frequency modulation operating mode indicator Once determined, the unit side will use For indexing from the reference response matrix Read the corresponding row vector to obtain the sixteen-dimensional reference response components. and will refer to the response components As a reference component with the same name and meaning across units, it is input to the decomposition and synthesis output head and the reference correction output head, so that the generation of frequency modulation power increment must go through the unified interface of the reference response component.

[0044] The difference characterization branch represents the 32-dimensional measurement points as vectors. With the ten-dimensional unit operation constraint feature vector Forty-two-dimensional constraint fusion vectors are formed by splicing them together in a fixed order. The eight-dimensional unit difference parameters are obtained sequentially through two fully connected layers. The unit side will display the unit's differential parameters. Limited to use only for generating local frequency regulation power increments, and not included in the unit-side correction information for generating shared frequency regulation response reference tables. The key intermediate quantities / output information are shown in Table 2: Table 2

[0045] The decomposition and synthesis output head receives sixteen-dimensional reference response components. Difference parameters between the eight-dimensional unit and the unit And concatenate them into a 24-dimensional composite input vector. First, a fully connected layer containing sixteen neurons is used to form a twenty-four-dimensional frequency-modulated synthesis representation. Then, a fully connected layer containing one neuron outputs a one-dimensional frequency-modulated power increment candidate. .

[0046] Referencing the corrected output head, the thirty-two-dimensional measurement point representation vector is received. With sixteen-dimensional reference response components And concatenate them into forty-eight positive input vectors for maintenance. The output is a 16-dimensional reference bias vector after passing through a fully connected layer with 16 neurons. The unit side utilizes the current key measuring point values ​​of the unit. , , , , With the ten-dimensional unit operation constraint feature vector The constraint margin is calculated and then converted into an allowable correction range and correction direction using a margin mapping matrix. This conversion is performed using the following formula:

[0047] in, This is a 64-dimensional local real-time running measurement point data feature vector. The active power of the unit. The rate of change of the unit's active power. Main valve opening. Main steam pressure This is the reheat temperature; This is a ten-dimensional characteristic vector representing the operational constraints of the generating unit. This represents the lower limit of the unit's output. This is the upper limit of the unit's output. The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. The lower limit of the main valve opening. The upper limit of the main valve opening. The lower limit of the main steam pressure. The upper limit of the main steam pressure. This is the lower limit of the reheat temperature. This is the upper limit of the reheat temperature; This is a five-dimensional upward constraint margin vector. This is a five-dimensional downside constraint margin vector; It is a 16-row, 5-column uprow margin mapping matrix. It is a 16x5 downshift margin mapping matrix, and and The matrix elements are preset coefficients used to map the five-dimensional margin row by row to the allowable correction range of each dimension of the sixteen-dimensional reference response component; This is a sixteen-dimensional reference deviation vector; A sixteen-dimensional vector that allows for correction range; The sixteen positive direction vectors are given, with each dimension taking the values ​​of 1, 0, and negative 1. This is an element-wise indicator function; it returns 1 if the condition is true and 0 otherwise. This is element-wise multiplication; The lower bound truncation operator is used to truncate negative values ​​to zero; T is the vector transpose operator.

[0048] The unit side will indicate the frequency regulation mode. Sixteen-dimensional reference deviation vector Sixteen-dimensional allowable correction range vector With the sixteenth maintenance positive direction vector Combined with shared frequency response reference table correction information It is then uploaded to the parameter server, which uses the parameter server to perform consensus-based distributed variational inference fusion update on the shared frequency modulation response reference table.

[0049] 4) The parameter server receives the correction information of the shared frequency regulation response reference table from each unit, and uses consensus-based distributed variational inference to merge the correction information of the shared frequency regulation response reference table into a reference table correction amount with consistent operating conditions; based on the operating condition consistency threshold, it determines the weighting coefficient of the reference table correction amount of each unit and updates the shared frequency regulation response reference table to obtain the updated shared frequency regulation response reference table, and sends the updated shared frequency regulation response reference table to each unit. It should be noted that the reference Figure 5The parameter server receives the shared frequency regulation response reference table correction information uploaded by each unit and parses it into frequency regulation mode identifier, reference deviation vector, allowable correction range, and correction direction. Based on the frequency regulation mode identifier, the shared frequency regulation response reference table correction information is merged to form eight sets of correction evidence corresponding to different frequency regulation mode identifiers. Each set of correction evidence is limited to reference table entries corresponding to the frequency regulation mode identifier in the shared frequency regulation response reference table. For each set of correction evidence, the sixteen-dimensional reference response component of the corresponding reference table entry is read as the prior reference for variational inference, and the sixteen-dimensional reference table correction amount of the reference table entry is used as the variable to be estimated to establish the variational parameters for consensus-based distributed variational inference. Iterative fusion of the same set of correction evidence is performed using consensus-based distributed variational inference, and the reference deviation vector of each unit is adjusted under the constraints of the allowable correction range and correction direction. The process is transformed into a probabilistic consistency contribution to the reference table correction amount, and the candidate reference table correction amount corresponding to the frequency modulation operating condition mode identifier is output. Consistency judgment is performed on the same type of correction evidence set based on the operating condition consistency threshold. The directional consistency between reference deviation vectors is used as the consistency judgment criterion. When the consistency meets the operating condition consistency threshold, the weighting coefficient of the corresponding shared frequency modulation response reference table correction information is increased; when the consistency does not meet the operating condition consistency threshold, the weighting coefficient of the corresponding shared frequency modulation response reference table correction information is decreased. The candidate reference table correction amounts are weighted according to the weighting coefficients and boundary constraints are applied. The weighted reference table correction amounts are then trimmed to the allowable correction range, and the correction direction is verified to conform to the correction direction constraint, resulting in reference table correction amounts with consistent operating condition meaning. The reference table correction amounts with consistent operating condition meaning are written into the corresponding reference table entries in the shared frequency modulation response reference table, and the sixteen-dimensional reference response components are updated to obtain the updated shared frequency modulation response reference table, which is then sent to each unit.

[0050] The specific operation of step 4) is as follows: The parameter server receives the correction information for the shared frequency regulation response reference table uploaded by each unit in each shared update cycle, denoted as... ,in, As the unit identifier, the parameter server will This is interpreted as an FM mode identifier. Sixteen-dimensional reference deviation vector Sixteen-dimensional allowable correction range vector With the sixteenth maintenance positive direction vector And apply validity constraints to each dimension: when When Set to zero, when When it is a non-positive number, the corresponding dimension will be... Set to zero and Set to zero to eliminate evidence contributions from uncorrectable dimensions.

[0051] The parameter server identifies the frequency modulation operating mode. For all The frequency modulation mode, after being merged and forming eight categories of corrected evidence sets, is identified as follows: For each category The parameter server will stack the reference deviation vectors row by row to obtain the deviation evidence matrix. This will allow the range evidence matrix to be obtained by stacking the modified range vectors row by row. The direction evidence matrix is ​​obtained by stacking the corrected direction vectors row by row. and will Limited to working only on the shared frequency response reference table The medium frequency modulation operating mode is identified as follows: Reference table entries.

[0052] For each category The parameter server references the shared frequency modulation response table. Read the sixteen-dimensional reference response components corresponding to the frequency modulation mode identifier. As a priori reference for variational inference, the sixteen-dimensional reference table correction for this reference table entry is defined as... As the variable to be estimated, the parameter server's direction evidence matrix. The sixteen-dimensional majority direction vector is obtained by statistically analyzing the majority direction dimension by dimension. ,in, for The value that appears most frequently in this dimension.

[0053] Parameter server based on operating condition consistency threshold Calculate the weighting coefficients First calculate the score for consistent direction. ,Will Defined as satisfying The ratio is obtained by dividing the number of dimensions by sixteen; then two levels of weight scalars are set. and and satisfy Greater than and When it is a positive number, Not less than season ,when Less than season .

[0054] The parameter server uses consensus-based distributed variational inference for the same class The fusion is performed using variational parameters represented by a mean vector and a diagonal precision vector, and corrected using a sixteen-dimensional candidate reference table. As a result of the fusion, Calculate using the following formula:

[0055] in, The frequency modulation operating mode is identified as follows: The revised evidence set; This is a sixteen-dimensional reference deviation vector; A sixteen-dimensional vector that allows for correction range; The sixteen positive direction vectors are given, with each dimension taking the values ​​of 1, 0, and negative 1. This is the sixteen-dimensional evidence mean vector obtained under the constraints of the allowed correction range and correction direction; This is a sixteen-dimensional evidence precision vector constructed from the allowable correction range vector; For operating condition consistency threshold The resulting weighting coefficients; It is a sixteen-dimensional prior precision vector and is a fixed configuration parameter of the parameter server; A sixteen-dimensional fusion precision vector; This is the correction amount for the sixteen-dimensional candidate reference table; The parameter is a positive lower bound. This is an element-wise indicator function; it returns 1 if the condition is true and 0 otherwise. and For the operators of taking the smallest element and taking the largest element; This is element-wise multiplication; This is the element-wise reciprocal vector. The interaction direction and load details are shown in Table 3. Table 3

[0056] The parameter server performs boundary constraints and direction checks on the candidate reference table correction values ​​to obtain reference table correction values ​​with consistent working condition meanings. Boundary constraints include a shared frequency modulation response reference table. Preset allowable correction range ,in, It is a sixteen-dimensional lower bound vector. It is a 16-dimensional upper bound vector; the parameter server will Dimensional clipping to interval The correction amount within the boundary is then calculated back. The parameter server also uses the range evidence matrix. Calculate the evidence range threshold for each dimension. ,Will Defined as Sorting by value and then accumulating the weights. The weighted median corresponding to half of the total weight is reached, and the positive repair quantity is clipped to the interval. Direction verification uses the majority of direction vectors. Based on, when When the positive quantity of the maintenance is limited to non-negative, When the positive quantity of this maintenance is restricted to non-positive, when The maintenance quantity should be set to zero at that time.

[0057] The parameter server will Write to the shared frequency modulation response reference table The medium frequency modulation operating mode is identified as follows: The reference table entries, according to The sixteen-dimensional reference response components are updated, and the identifiers for each of the eight frequency modulation mode types are updated accordingly, resulting in the updated shared frequency modulation response reference table. And send it to each unit.

[0058] 5) Each unit receives the updated shared frequency regulation response reference table, obtains the predicted value of frequency regulation power increment based on the unit difference parameters and frequency regulation power increment candidates, and determines the predicted value of frequency regulation power increment under the unit operation constraints to generate frequency regulation power increment command. It should be noted that the reference Figure 6 Each unit receives the updated shared frequency modulation response reference table sent by the parameter server and loads the updated shared frequency modulation response reference table as an indexable reference library; it reads the frequency modulation operating condition mode identifier and the eight-dimensional unit difference parameter corresponding to the frequency modulation power increment candidate from the network output of the operating condition mode-shared frequency modulation response reference table, while keeping the eight-dimensional unit difference parameter from being written into the updated shared frequency modulation response reference table; it reads the sixteen-dimensional reference response component corresponding to the frequency modulation operating condition mode identifier from the updated shared frequency modulation response reference table using the frequency modulation operating condition mode identifier as an index, and uses the sixteen-dimensional reference response component as the unified reference input for frequency modulation power increment prediction; it decomposes and synthesizes the sixteen-dimensional reference response component and the eight-dimensional unit difference parameter input into an output head, which is first passed through a fully connected layer with sixteen neurons to form a twenty-four-dimensional frequency modulation response reference. The frequency modulation power increment prediction is generated by passing the frequency synthesis representation through a fully connected layer containing one neuron. The frequency modulation power increment candidates are then recalculated using the updated shared frequency modulation response reference table to complete the frequency modulation power increment prediction generation. Under the constraints of the ten-dimensional unit operation constraint feature vector, the executable boundary determination of the frequency modulation power increment prediction is performed, converting the primary equipment safety limits and controller adjustment range into allowable intervals and rate of change boundaries. When the frequency modulation power increment prediction simultaneously meets the allowable interval and rate of change boundaries, the frequency modulation power increment command is set as the frequency modulation power increment prediction. When the frequency modulation power increment prediction does not meet the allowable interval or rate of change boundaries, the frequency modulation power increment command is set as the result after adjusting the frequency modulation power increment prediction within the boundary. The output frequency modulation power increment command is used by the unit control system to execute the frequency modulation closed-loop delivery.

[0059] The specific operation of step 5) is as follows: The unit receives the updated shared frequency modulation response reference table sent by the parameter server, denoted as... and will Load it as an indexable and readable reference library, and maintain the reference response matrix on the unit side. ,in, Frequency modulation operating mode identifier The corresponding sixteen-dimensional reference response components. Preset response node vectors on the unit side. And store in a fixed location. For reference response component number The corresponding response time scalar, The frequency modulation operating mode is identified as follows: At the moment of response The reference response value.

[0060] Unit Reading Operating Mode - Shared Frequency Regulation Response Reference Table Decomposition Network Output Frequency Regulation Operating Mode Identifier Difference parameters between the eight-dimensional unit and the unit ,in, For unit identification, to The eight numerical components of the branch output are used to characterize the differences in available adjustability and execution characteristics of the unit under the same reference response component, arranged in a fixed order. The unit side will use the eight-dimensional unit difference parameters. The differences are retained as local variations; the unit side does not refer to the updated shared frequency regulation response table. Write the difference parameters of the eight-dimensional unit .

[0061] The unit side is marked with frequency regulation mode. For indexing from the reference response matrix Reading the 16-dimensional reference response components and the sixteen-dimensional reference response components This serves as a unified reference input for predicting frequency regulation power increments. The unit side will... and Concatenate them into a 24-dimensional composite input vector in a fixed dimensional order. ,Will The input decomposes and synthesizes the output head, first passing it through a fully connected layer with sixteen neurons to obtain a twenty-four-dimensional frequency-modulated synthesis representation. Then, a one-dimensional frequency-modulated power increment prediction is output through a fully connected layer containing one neuron. Thus, with the updated shared frequency response reference table The candidate frequency modulation power increments are recalculated and a frequency modulation power increment prediction is generated. .

[0062] Reading the ten-dimensional unit operation constraint feature vector from the unit side And synchronously read the values ​​of the currently executing measurement points on the local machine, and define... This represents the current active power of the unit. Main valve opening. Main steam pressure This is the reheat temperature. The unit reads the frequency regulation power increment command issued in the previous control cycle. And read the control cycle duration The unit side introduces a constraint transformation coefficient vector with fixed configuration. The primary equipment safety limits and controller adjustment range are simultaneously converted into the allowable range and rate of change boundary of frequency modulation power increment. The intersection of the allowable range and the rate of change boundary is used to obtain the executable command range. Then, the frequency modulation power increment prediction is performed. Projecting this onto the executable instruction range yields the frequency modulation power increment instruction. for

[0063] in, This is a ten-dimensional characteristic vector representing the operational constraints of the generating unit. This represents the lower limit of the unit's output. This is the upper limit of the unit's output. The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. The lower limit of the main valve opening. The upper limit of the main valve opening. The lower limit of the main steam pressure. The upper limit of the main steam pressure. This is the lower limit of the reheat temperature. This is the upper limit of the reheat temperature; This represents the real-time value of the unit's current active power. This is the real-time value of the main steam valve opening. This is the real-time value of the main steam pressure. This is the real-time value of the reheat temperature; For the constraint transformation coefficient vector, , , These are the coefficients that convert the main steam valve opening margin, main steam pressure margin, and reheat temperature margin into the usable power increment limit, respectively. To control cycle duration; For one-dimensional frequency modulation power increment prediction; This refers to the frequency modulation power increment command issued in the previous control cycle; and These are the upper and lower bounds of the allowed interval, respectively. and These are the lower and upper bounds of the executable instruction range after intersecting the allowed range and the rate of change boundary, respectively; This is a frequency modulation power increment command; To find the minimum value operator, This is the operator for finding the maximum value.

[0064] The unit side will issue frequency regulation power increment commands The instructions are sent to the unit control system for frequency regulation closed-loop delivery.

[0065] 6) The unit's control system adjusts the unit's output according to the frequency regulation power increment command, generates frequency regulation execution results with time stamps, and forms the unit's frequency regulation record.

[0066] It should be noted that the reference Figure 7 The unit's control system receives the frequency modulation power increment command, parses it into an execution cycle corresponding to the unit output setting increment and the preset sampling period, and inputs the unit output setting increment into the output adjustment channel of the unit control system. Within the allowable range and rate of change boundary of the unit's operating constraints, the unit control system drives the actuator to complete the unit output adjustment. During the output adjustment process, it collects local real-time operating measurement point data according to the preset sampling period to form a frequency modulation execution measurement point sequence. Based on the time of the frequency modulation power increment command issuance, it extracts verifiable frequency modulation response segments corresponding to the unit output change from the frequency modulation execution measurement point sequence according to the preset response window length, and associates these verifiable frequency modulation response segments with the frequency modulation operating mode identifier. The frequency modulation power increment command, the frequency modulation operating mode identifier, the verifiable frequency modulation response segments, and the corresponding time identifier are combined to generate a frequency modulation execution result with a time identifier, forming the unit's frequency modulation record.

[0067] The specific operation of step 6) is as follows: The unit control system receives frequency modulation power increment commands. ,Will Analysis of the unit output setting increment With preset sampling period Corresponding execution clock vector ,in Pick The value, To execute the beat length, For the first The percentage of the set increment applied within each sampling period and satisfying the following conditions: and The unit control system is based on the ten-dimensional unit operation constraint feature vector. Reading the rate of change boundary and ,in, The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. This serves as the unit's identifier. The unit's control system uses... Convert the allowable setting variation range for a single sampling period: when When the value is non-negative, the allowable amplitude per cycle is... ;when When the value is negative, the allowable amplitude for a single cycle is... The unit control system is determined. To meet The allowable amplitude of a single cycle is not less than The smallest positive integer, and Decomposed into Segment setting increment sequence :forward Segment symbol and Consistent and with an amplitude equal to the allowable amplitude for a single cycle, the first Take the remaining amplitude of the segment to make Established. The unit control system will execute the cycle vector. The components are calculated as follows And set the unit output increment With execution clock vector The output regulation channel of the unit control system is input, and the time of issuance of the frequency regulation power increment command is recorded as follows: .

[0068] The unit control system drives the actuators to adjust the unit output within the allowable range and rate of change boundaries of the unit's operating constraints, and operates at a preset sampling period. Local real-time operating measurement point data is collected to form a frequency modulation execution measurement point sequence. The frequency modulation execution measurement point sequence uses a time stamp vector. With measurement point matrix It means that, among them, For the first scalar at each sampling time point The number of sampling points covered in this adjustment will be reduced by one. This is the 64-dimensional local real-time operating measurement point data feature vector at this sampling moment. The 64-dimensional local real-time operating measurement point data feature vector is named according to the preset 64-dimensional measurement point names in the unit control system. The data was collected in an array, among which... The name of the active power measurement point of the unit and its corresponding , The name of the measurement point for the rate of change of active power of the unit and its corresponding , Main valve opening measurement point name and corresponding , Main steam pressure measuring point name and corresponding , The name of the reheat temperature measurement point and its corresponding ,the remaining The name of the local real-time operating measurement point is fixed on the unit side and consistent with the network input of the shared frequency regulation response reference table in the operating mode. To ensure traceability of changes in the output regulation channel settings, the unit control system synchronously generates the set output sequence. ,in, , for The reference output setting value of the output adjustment channel at all times.

[0069] The unit control system is Based on the preset response window length The verifiable frequency regulation response segment corresponding to the unit output change is extracted from the frequency regulation execution measurement point sequence. The unit control system will then increase the number of sampling points in the window. Determined to be no greater than The largest integer, and from Extracting row index range Forming a segment measurement point matrix ,from Extract the corresponding segment time identifier vector from the middle And satisfy The unit control system from The Active power sequence of the extraction unit And calculate the force change sequence. The output response serves as a verifiable frequency regulation response segment. The unit control system reads the frequency regulation mode identifier. And identify the verifiable frequency modulation response segment and the frequency modulation operating mode. To perform association, the association key uses a tuple. .

[0070] The unit control system will issue frequency regulation power increment commands. Frequency modulation operating mode indicator Verifiable frequency regulation response segments are combined with corresponding time markers to generate time-marked frequency regulation execution results and form the unit's frequency regulation record. The time-marked frequency regulation execution results are processed using a matrix. express, Align by line And concatenate them column by column ,in, It is a column vector. It is a column vector. It is a column vector. The unit frequency regulation record uses a structured data group. Storage enables the combination of frequency modulation power increment commands, execution beats, measurement point segments, and time markers into numerical vectors and numerical matrices that can be directly indexed and read.

[0071] This invention has the following characteristics: When each generating unit invokes the operating mode-shared frequency regulation response reference table decomposition network based on its frequency regulation operating mode characterization, the operating mode-shared frequency regulation response reference table decomposition network does not directly learn the overall black-box mapping from local real-time operating measurement data to frequency regulation power increments. Instead, it decomposes the generation link of frequency regulation power increments into cross-unit synonymous reference parts and unit-specific differences. The operating mode-shared frequency regulation response reference table decomposition network first calculates the frequency regulation operating mode identifier based on local real-time operating measurement data. The frequency regulation operating mode identifier is used to point to the reference table entries in the shared frequency regulation response reference table. The shared frequency regulation response reference table defines the meaning of the reference response component corresponding to the same frequency regulation operating mode identifier for each reference table entry, enabling different generating units to use the same reference response component expression method under the same frequency regulation operating mode identifier. The operating condition mode-shared frequency regulation response reference table decomposition network then calculates the unit difference parameters based on local real-time operating measurement data and unit operating constraints. These unit difference parameters describe the differences in available regulation capacity and execution characteristics of the unit under the same reference response component. These unit difference parameters do not participate in the updating of the shared frequency regulation response reference table. The operating condition mode-shared frequency regulation response reference table decomposition network calculates frequency regulation power increment candidates based on the reference response components and unit difference parameters. Unlike existing federated training that uploads and aggregates the weights of the entire model, the operating condition mode-shared frequency regulation response reference table decomposition network simultaneously generates shared frequency regulation response reference table correction information when generating frequency regulation power increment candidates. This correction information consists of a frequency regulation operating condition mode identifier and a description of the deviation between the reference response component and verifiable frequency regulation response segments in the local real-time operating measurement data. Under unit operating constraints, the deviation description forms an allowable correction range and direction, enabling the parameter server to perform semantic correction on the shared frequency regulation response reference table without receiving local real-time operating measurement data. In the multi-unit frequency regulation scenario of thermal power plants, this decomposition structure solidifies cross-unit consistency in the shared frequency regulation response reference table and reference response components, and solidifies cross-unit differences in the unit difference parameters. The candidate frequency regulation power increment and the correction information of the shared frequency regulation response reference table are derived from the same decomposition result, forming a unified interface that can be integrated by the parameter server.

[0072] After receiving the correction information from the shared frequency regulation response reference table of each unit, the parameter server does not perform a weighted average of the weights of the unit's local model. Instead, it treats the correction information as evidence for correcting the reference table entries in the shared frequency regulation response reference table, and uses consensus-based distributed variational inference to fuse the correction evidence, forming a reference table correction quantity for each reference table entry. The fusion object of the consensus-based distributed variational inference is the reference table entry corresponding to the frequency regulation operating condition mode identifier. The parameter server merges the correction information of the shared frequency regulation response reference table according to the frequency regulation operating condition mode identifier, and fuses the reference table correction quantity separately for each type of frequency regulation operating condition mode identifier, ensuring that different frequency regulation operating condition mode identifiers do not interfere with each other. The parameter server determines the weighting coefficient of the reference table correction based on the operating condition consistency threshold: The parameter server performs a consistency judgment on the deviation descriptions of the shared frequency regulation response reference table correction information for each unit under the same frequency regulation operating condition mode identifier. When the consistency meets the operating condition consistency threshold, the weighting coefficient of the corresponding shared frequency regulation response reference table correction information is increased; when the consistency does not meet the operating condition consistency threshold, the weighting coefficient of the corresponding shared frequency regulation response reference table correction information is decreased while maintaining the reference table correction within the allowable correction range of unit operating constraints. This weighting coefficient is driven by the operating condition consistency threshold and does not depend on the unit data volume or unit number, ensuring that the meaning of the frequency regulation operating condition mode identifier is still aligned even when data distribution across units is inconsistent. The parameter server updates the shared frequency regulation response reference table according to the reference table correction, obtains the updated shared frequency regulation response reference table, and sends it to each unit. Unlike existing technologies that use centralized aggregation of local real-time measurement data for unified modeling, the parameter server only processes and updates the shared frequency modulation response reference table correction information. The operating condition consistency threshold forms a clear judgment boundary in the fusion link, ensuring that the shared frequency modulation response reference table maintains the same frequency modulation operating condition mode identifier corresponding to the same reference response component meaning across multiple units. The coordinated execution of frequency modulation power increment commands is provided by a unified reference from the shared frequency modulation response reference table.

[0073] After receiving the updated shared frequency regulation response reference table, each unit's control link does not directly use a single local output as the frequency regulation power increment command. Instead, it uses the updated shared frequency regulation response reference table as a unified reference for the frequency regulation operating mode identifier, recalculating the frequency regulation power increment candidates to obtain the frequency regulation power increment prediction. The calculation process for the frequency regulation power increment prediction remains consistent with the operating mode—shared frequency regulation response reference table decomposition network: the frequency regulation power increment candidate is composed of reference response components and unit difference parameters. The updated shared frequency regulation response reference table changes the meaning of the reference response component values, while the unit difference parameters maintain the unit's individual difference expression. The unit difference parameters and the updated shared frequency regulation response reference table jointly determine the frequency regulation power increment prediction. Each generating unit, under operating constraints, determines the frequency modulation power increment prediction and generates a frequency modulation power increment command. The determination process uses executable boundary judgment rather than training index judgment. Unit operating constraints include primary equipment safety limits and controller adjustment ranges, providing allowable intervals and rate-of-change boundaries for the frequency modulation power increment prediction. When the frequency modulation power increment prediction meets the unit operating constraints, the frequency modulation power increment command is based on the predicted frequency modulation power increment. When the predicted frequency modulation power increment does not meet the unit operating constraints, the frequency modulation power increment command is obtained by adjusting the predicted frequency modulation power increment within the boundaries according to the unit operating constraints. This step differs from existing technologies that generate commands independently at the unit level, leading to inconsistent command meanings across multiple units under the same disturbance. The meaning of the frequency modulation power increment command is constrained by the updated shared frequency modulation response reference table, and the executability of the frequency modulation power increment command is determined by the unit operating constraints. At the cross-unit coordination level, the frequency modulation power increment command has the same frequency modulation operating mode identifier corresponding to the same reference response component meaning, forming a command semantic closure consistent with the shared frequency modulation response reference table of the parameter server.

[0074] Example 2 The distributed frequency regulation optimization system for thermal power plants based on variational inference described in this invention includes: The acquisition module is used to acquire the set of units participating in frequency regulation in the thermal power plant, the unit operation constraints and operating condition consistency thresholds of each unit, initialize a unified shared frequency regulation response reference table across units in the parameter server, and send it to each unit. The matching module is used by each unit to acquire local real-time operating measurement data. Under the constraints of unit operation, it performs operating condition matching on the local real-time operating measurement data in conjunction with the shared frequency regulation response reference table to generate a characterization of the unit's frequency regulation operating condition mode. The calculation module is used by each unit to decompose the network based on the unit frequency regulation operating mode characterization, call the operating mode-shared frequency regulation response reference table, determine the frequency regulation operating mode identifier, and then decompose the local real-time operating measurement point data into the reference response component and unit difference parameter corresponding to the frequency regulation operating mode identifier under the constraints of the shared frequency regulation response reference table; calculate the frequency regulation power increment candidate based on the reference response component and the unit difference parameter, and generate shared frequency regulation response reference table correction information based on the deviation between the reference response component and the shared frequency regulation response reference table; The update module is used by the parameter server to receive the correction information of the shared frequency regulation response reference table of each unit, and to use consensus-based distributed variational inference to merge the correction information of the shared frequency regulation response reference table into a reference table correction amount with consistent operating conditions; based on the operating condition consistency threshold, the weighting coefficient of the reference table correction amount of each unit is determined and the shared frequency regulation response reference table is updated to obtain the updated shared frequency regulation response reference table, and the updated shared frequency regulation response reference table is sent to each unit; The determination module is used for each unit to obtain the predicted value of frequency regulation power increment based on the shared frequency regulation response reference table, unit difference parameters and frequency regulation power increment candidates, and to determine the predicted value of frequency regulation power increment under the unit operation constraints, and generate frequency regulation power increment command. The control module is used by the unit's control system to adjust the unit's output according to the frequency regulation power increment command.

[0075] 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.

[0076] 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 the variational inference-based distributed frequency regulation optimization method for thermal power plants. For example, the steps include: acquiring the set of units participating in frequency regulation in the thermal power plant, the unit operating constraints and operating condition consistency thresholds for each unit; initializing a unified shared frequency regulation response reference table across units in a parameter server and sending it to each unit; each unit acquiring local real-time operating measurement data, and under unit operating constraints, performing operating condition matching on the local real-time operating measurement data in conjunction with the shared frequency regulation response reference table to generate a unit frequency regulation operating condition mode representation; each unit, based on the unit frequency regulation operating condition mode representation, calling the operating condition mode-shared frequency regulation response reference table decomposition network to determine the frequency regulation operating condition mode identifier, and then, under the constraints of the shared frequency regulation response reference table, decomposing the local real-time operating measurement data into reference responses corresponding to the frequency regulation operating condition mode identifier. The system receives the shared frequency regulation response reference table correction information from each unit, and uses consensus-based distributed variational inference to fuse the correction information into a reference table correction amount with consistent operating conditions. Based on the operating condition consistency threshold, the system determines the weighting coefficients of the reference table correction amounts for each unit and updates the shared frequency regulation response reference table to obtain the updated reference table. The updated reference table is then sent to each unit. Each unit obtains a predicted value for the frequency regulation power increment based on the shared frequency regulation response reference table, unit difference parameters, and frequency regulation power increment candidates. Under unit operating constraints, the predicted value is judged, and a frequency regulation power increment command is generated. The unit's control system adjusts the unit output according to the frequency regulation power increment command. 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.

[0077] Example 4 A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the variational inference-based distributed frequency regulation optimization method for thermal power plants. For example, the steps include: acquiring the set of units participating in frequency regulation in the thermal power plant, the unit operating constraints and operating condition consistency thresholds for each unit; initializing a unified shared frequency regulation response reference table across units in a parameter server and sending it to each unit; each unit acquiring local real-time operating measurement data, and under unit operating constraints, performing operating condition matching on the local real-time operating measurement data in conjunction with the shared frequency regulation response reference table to generate a unit frequency regulation operating condition mode representation; each unit, based on the unit frequency regulation operating condition mode representation, calling the operating condition mode-shared frequency regulation response reference table decomposition network to determine the frequency regulation operating condition mode identifier, and then, under the constraints of the shared frequency regulation response reference table, decomposing the local real-time operating measurement data into reference response components corresponding to the frequency regulation operating condition mode identifier and unit differences. The parameters are as follows: The frequency modulation power increment candidate is calculated based on the reference response component and the unit difference parameter; and a shared frequency modulation response reference table correction information is generated based on the deviation between the reference response component and the shared frequency modulation response reference table. The parameter server receives the shared frequency modulation response reference table correction information from each unit, and uses consensus-based distributed variational inference to fuse the shared frequency modulation response reference table correction information into a reference table correction amount with consistent operating conditions. The weighting coefficient of the reference table correction amount for each unit is determined based on the operating condition consistency threshold, and the shared frequency modulation response reference table is updated to obtain the updated shared frequency modulation response reference table, which is then sent to each unit. Each unit obtains a predicted value of the frequency modulation power increment based on the shared frequency modulation response reference table, the unit difference parameter, and the frequency modulation power increment candidate, and judges the predicted value of the frequency modulation power increment under unit operating constraints to generate a frequency modulation power increment command. The unit's control system adjusts the unit output according to the frequency modulation power increment command. 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.

[0078] 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.

[0079] 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.

[0080] 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.

[0081] 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.

[0082] 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.

[0083] 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.

[0084] 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 method for optimizing distributed frequency regulation in thermal power plants based on variational inference, characterized in that, include: 1) Obtain the set of units participating in frequency regulation in the thermal power plant, the unit operation constraints and operating condition consistency thresholds of each unit, initialize a unified shared frequency regulation response reference table across units in the parameter server, and send it to each unit; 2) Each unit acquires local real-time operating measurement data. Under the constraints of unit operation, the local real-time operating measurement data is matched with the operating conditions in conjunction with the shared frequency regulation response reference table to generate a characterization of the unit's frequency regulation operating condition mode. 3) Based on the frequency regulation operating mode characterization of the unit, each unit calls the operating mode-shared frequency regulation response reference table decomposition network to determine the frequency regulation operating mode identifier. Then, under the constraints of the shared frequency regulation response reference table, the local real-time operating measurement point data is decomposed into the reference response component and unit difference parameter corresponding to the frequency regulation operating mode identifier. The frequency regulation power increment candidate is calculated according to the reference response component and the unit difference parameter, and the shared frequency regulation response reference table correction information is generated based on the deviation between the reference response component and the shared frequency regulation response reference table. 4) The parameter server receives the correction information of the shared frequency regulation response reference table from each unit, and uses consensus-based distributed variational inference to merge the correction information of the shared frequency regulation response reference table into a reference table correction amount with consistent operating conditions; based on the operating condition consistency threshold, it determines the weighting coefficient of the reference table correction amount of each unit and updates the shared frequency regulation response reference table to obtain the updated shared frequency regulation response reference table, and sends the updated shared frequency regulation response reference table to each unit. 5) Each unit obtains the predicted value of frequency regulation power increment based on the shared frequency regulation response reference table, unit difference parameters and frequency regulation power increment candidates, and judges the predicted value of frequency regulation power increment under unit operation constraints to generate frequency regulation power increment command; 6) The unit's control system adjusts the unit's output according to the frequency regulation power increment command.

2. The method for distributed frequency regulation optimization of thermal power plants based on variational inference according to claim 1, characterized in that, The specific operation of step 1) is as follows: Obtain the set of generating units participating in frequency regulation and assign unit identifiers to the set of generating units. Establish a consistent mapping between the unit identifiers and the 64-dimensional measurement point sequence of the feature vector of the local real-time operating measurement point data. Collect the primary equipment safety limits and controller adjustment ranges of each unit, encode the limits and ranges into a ten-dimensional unit operation constraint feature vector, and configure the same threshold for the same operating condition for the set of units. In the parameter server, a shared frequency modulation response reference table is constructed. The shared frequency modulation response reference table is set as an indexable and readable reference library, and eight types of frequency modulation mode identifiers are defined, each corresponding to a 16-dimensional reference response component. The values ​​and allowable correction ranges of each type of reference response component are initialized. The shared frequency regulation response reference table, along with the frequency regulation operating condition mode identifier, the unit operation constraint feature vector encoding rules, and the operating condition consistency threshold, are sent to each unit for subsequent operating condition matching and reference table correction.

3. The method for distributed frequency regulation optimization of thermal power plants based on variational inference according to claim 1, characterized in that, The specific operation of step 2) is as follows: Each unit collects local real-time operation measurement point data at the unit side and assembles it into a 64-dimensional local real-time operation measurement point data feature vector according to the preset measurement point order; Read the ten-dimensional unit operation constraint feature vector, perform constraint screening on the local real-time operation measurement point data feature vector, adjust the components that exceed the allowable range of unit operation constraints to the allowable range, and obtain the local real-time operation measurement point data feature vector after constraint screening; Based on the feature vector of local real-time operating measurement point data after constraint screening, the reference response components corresponding to the eight types of frequency regulation mode identifiers in the shared frequency regulation response reference table are called in sequence, the matching degree with each reference response component is calculated, and an eight-dimensional unit frequency regulation mode representation is formed. Based on the frequency regulation mode characterization of the eight-dimensional unit, candidate frequency regulation mode identifiers are determined, and the frequency regulation mode characterization of the eight-dimensional unit and candidate frequency regulation mode identifiers are output.

4. The method for distributed frequency regulation optimization of thermal power plants based on variational inference according to claim 1, characterized in that, The process of fusing the shared frequency modulation response reference table correction information into a reference table correction quantity with consistent operating condition meaning using consensus-based distributed variational inference is as follows: Consensus-based distributed variational inference is used to fuse the correction information of the shared frequency modulation response reference table to obtain the correction amount of the candidate reference table. Correction amount for candidate reference table By performing boundary constraints and direction verification, a reference table correction amount with consistent working condition meaning is obtained. .

5. The method for distributed frequency regulation optimization of thermal power plants based on variational inference according to claim 4, characterized in that, Candidate Reference Table Correction Quantity for: in, The frequency modulation operating mode is identified as follows: The revised evidence set; This is a sixteen-dimensional reference deviation vector; A sixteen-dimensional vector that allows for correction range; The sixteen positive direction vectors are given, with each dimension taking the values ​​of 1, 0, and negative 1. This is the sixteen-dimensional evidence mean vector obtained under the constraints of the allowed correction range and correction direction; This is a sixteen-dimensional evidence precision vector constructed from the allowable correction range vector; For operating condition consistency threshold The resulting weighting coefficients; It is a sixteen-dimensional prior precision vector and is a fixed configuration parameter of the parameter server; A sixteen-dimensional fusion precision vector; This is the correction amount for the sixteen-dimensional candidate reference table; The parameter is a positive lower bound. This is an element-wise indicator function; it returns 1 if the condition is true and 0 otherwise. and For the operators of taking the smallest element and taking the largest element; This is element-wise multiplication; It is the element-wise reciprocal vector.

6. The method for distributed frequency regulation optimization of thermal power plants based on variational inference according to claim 4, characterized in that, Boundary constraints include a shared frequency modulation response reference table. Preset allowable correction range ,in, It is a sixteen-dimensional lower bound vector. It is a sixteen-dimensional upper bound vector.

7. The method for distributed frequency regulation optimization of thermal power plants based on variational inference according to claim 4, characterized in that, Frequency modulation power increment command for: in, This is a ten-dimensional characteristic vector representing the operational constraints of the generating unit. This represents the lower limit of the unit's output. This is the upper limit of the unit's output. The boundary of the rate of change of unit output. The boundary of the rate of change of unit output. The lower limit of the main valve opening. The upper limit of the main valve opening. The lower limit of the main steam pressure. The upper limit of the main steam pressure. This is the lower limit of the reheat temperature. This is the upper limit of the reheat temperature; This represents the real-time value of the unit's current active power. This is the real-time value of the main steam valve opening. This is the real-time value of the main steam pressure. This is the real-time value of the reheat temperature; For the constraint transformation coefficient vector, , , These are the coefficients that convert the main steam valve opening margin, main steam pressure margin, and reheat temperature margin into the usable power increment limit, respectively. To control cycle duration; For one-dimensional frequency modulation power increment prediction; This refers to the frequency modulation power increment command issued in the previous control cycle; and These are the upper and lower bounds of the allowed interval, respectively. and These are the lower and upper bounds of the executable instruction range after intersecting the allowed range and the rate of change boundary, respectively; This is a frequency modulation power increment command; To find the minimum value operator, This is the operator for finding the maximum value.

8. A distributed frequency regulation optimization system for thermal power plants based on variational inference, characterized in that, include: The acquisition module is used to acquire the set of units participating in frequency regulation in the thermal power plant, the unit operation constraints and operating condition consistency thresholds of each unit, initialize a unified shared frequency regulation response reference table across units in the parameter server, and send it to each unit. The matching module is used by each unit to acquire local real-time operating measurement data. Under the constraints of unit operation, it performs operating condition matching on the local real-time operating measurement data in conjunction with the shared frequency regulation response reference table to generate a characterization of the unit's frequency regulation operating condition mode. The calculation module is used by each unit to decompose the network based on the unit frequency regulation operating mode characterization, call the operating mode-shared frequency regulation response reference table, determine the frequency regulation operating mode identifier, and then decompose the local real-time operating measurement point data into the reference response component and unit difference parameter corresponding to the frequency regulation operating mode identifier under the constraints of the shared frequency regulation response reference table; calculate the frequency regulation power increment candidate based on the reference response component and the unit difference parameter, and generate shared frequency regulation response reference table correction information based on the deviation between the reference response component and the shared frequency regulation response reference table; The update module is used by the parameter server to receive the correction information of the shared frequency regulation response reference table of each unit, and to use consensus-based distributed variational inference to merge the correction information of the shared frequency regulation response reference table into a reference table correction amount with consistent operating conditions; based on the operating condition consistency threshold, the weighting coefficient of the reference table correction amount of each unit is determined and the shared frequency regulation response reference table is updated to obtain the updated shared frequency regulation response reference table, and the updated shared frequency regulation response reference table is sent to each unit; The determination module is used for each unit to obtain the predicted value of frequency regulation power increment based on the shared frequency regulation response reference table, unit difference parameters and frequency regulation power increment candidates, and to determine the predicted value of frequency regulation power increment under the unit operation constraints, and generate frequency regulation power increment command. The control module is used by the unit's control system to adjust the unit's output according to the frequency regulation power increment command.

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 distributed frequency regulation optimization method for thermal power plants based on variational inference 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 distributed frequency regulation optimization method for thermal power plants based on variational inference as described in any one of claims 1-7.

Images