A power plant intelligentization-oriented thermal power unit adaptive control method and system

Abstract

This application proposes an adaptive control method and system for thermal power units aimed at power plant intelligence. The method includes: establishing a thermal power unit type library and determining each controlled object corresponding to each thermal power unit in the type library; using a data-driven intelligent T-S fuzzy identification method to model the dynamic characteristics of each controlled object, obtaining a dynamic characteristic model of each controlled object; determining an adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic models of each controlled object; dynamically tuning the control parameters using a fuzzy tuning method or a swarm intelligence optimization algorithm; and performing adaptive control of the thermal power unit based on the dynamically tuned control parameters. The technical solution proposed in this application takes into account multiple objectives and constraints in the optimization of thermal power unit operation, such as safety, stability, economy, energy saving, and environmental protection, while flexibly meeting various user needs and possessing scalability.

Description

Technical Field

[0001] This application relates to the field of intelligent control of thermal power units, and in particular to an adaptive control method and system for thermal power units oriented towards power plant intelligence. Background Technology

[0002] Driven by the dual goals of "carbon peaking and carbon neutrality," the digital and intelligent development of thermal power generation is not only an effective means to fully tap its flexible peak-shaving potential and increase the proportion of new energy consumption in the power grid, but also a crucial link in achieving energy structure transformation. Based on this, the operation and control of thermal power units must simultaneously meet multiple objectives, including safety and stability, economic efficiency, energy conservation and emission reduction, and low-carbon environmental protection. The achievement of these objectives is closely related to the close cooperation and coordinated control of the main and auxiliary equipment of the unit; therefore, the construction of integrated intelligent power plants is imperative.

[0003] Currently, the successful trials of intelligent and digital construction in demonstration power plants have taken a solid step forward in the construction of intelligent power plants. However, with the increasing integration of power plant management and control, the complexity of the dynamic characteristics of the entire unit has intensified, and the difficulty of control strategy design has increased, posing a significant challenge to the safe and stable operation of thermal power units. Summary of the Invention

[0004] This application provides an adaptive control method and system for thermal power units oriented towards power plant intelligence, in order to at least address the technical problem that the increased complexity of the dynamic characteristics of the unit and the increased difficulty in the design of control strategies pose a great challenge to the safe and stable operation of thermal power units.

[0005] The first aspect of this application proposes an adaptive control method for thermal power units aimed at power plant intelligence, the method comprising:

[0006] Establish a thermal power unit type library and determine each controlled object corresponding to each thermal power unit in the type library;

[0007] The dynamic characteristics of each controlled object are modeled using a data-driven intelligent TS fuzzy identification method, resulting in a dynamic characteristic model of each controlled object.

[0008] Based on the dynamic characteristic models of each controlled object, the adaptive control strategy under multiple objectives and constraints is determined;

[0009] Dynamic tuning of control parameters is performed using fuzzy tuning methods or swarm intelligence optimization algorithms;

[0010] Adaptive control of thermal power units is performed based on dynamically tuned control parameters.

[0011] Preferably, the controlled object includes: a boiler-turbine coordinated control system, a combustion subsystem, a boiler subsystem, a steam turbine subsystem, and a gas turbine subsystem.

[0012] Furthermore, the dynamic characteristic modeling of each controlled object using the data-driven intelligent TS fuzzy identification method, to obtain the dynamic characteristic model of each controlled object, includes:

[0013] Determine the operating principle of the i-th controlled object and the importance of the variables' influence on the dynamic characteristics of the controlled object;

[0014] The input and output variables are determined based on the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object.

[0015] Based on the data types corresponding to the input and output variables, the running data of the controlled object during the historical period is obtained, and the running data is cleaned using a data cleaning algorithm.

[0016] The cleaned runtime data is divided into a training set and a validation set;

[0017] The dynamic characteristic model of the controlled object is determined based on the running data in the training set;

[0018] Where i∈[1~I], and I is the total number of controlled objects;

[0019] The data cleaning algorithms include: wavelet analysis algorithm, Wiener filtering algorithm, Kalman filtering algorithm, and neural network algorithm.

[0020] Furthermore, determining the dynamic characteristic model of the controlled object based on the runtime data in the training set includes:

[0021] The squirrel optimization algorithm is used to divide the running data in the training set, and the data division results of the running data in the training set under various working conditions are obtained.

[0022] The initial model of the controlled object is determined based on the data partitioning results of the training set's running data under various operating conditions.

[0023] The model parameters of the initial model are determined using the model parameter identification method, thereby obtaining the dynamic characteristic model of the controlled object.

[0024] Furthermore, the determination of the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic models of each controlled object includes:

[0025] The dynamic characteristic model of the controlled object is used as the prediction model;

[0026] The objective function and constraints of the predictive control rolling optimization process are determined based on the actual operation optimization requirements of the thermal power unit.

[0027] The optimal control law is obtained by optimizing the control law based on the objective function and constraints.

[0028] Furthermore, the dynamic tuning of control parameters using the fuzzy tuning method includes:

[0029] Determine the input and output variables of the fuzzy rule;

[0030] Construct fuzzy rules for parameter optimization based on the input and output variables;

[0031] The control parameters are tuned based on the constructed fuzzy rules for parameter optimization.

[0032] A second aspect of this application provides an adaptive control system for thermal power units aimed at power plant intelligence, the system comprising:

[0033] The first determining module is used to establish a thermal power unit type library and determine each controlled object corresponding to each thermal power unit in the type library;

[0034] The modeling module is used to model the dynamic characteristics of each controlled object using a data-driven intelligent TS fuzzy identification method, and obtain the dynamic characteristic model of each controlled object.

[0035] The second determining module is used to determine the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic model of each controlled object;

[0036] The tuning module is used to dynamically tune control parameters using fuzzy tuning methods or swarm intelligence optimization algorithms.

[0037] The control module is used for adaptive control of thermal power units based on dynamically tuned control parameters.

[0038] Preferably, the modeling module includes:

[0039] The first determining unit is used to determine the operating principle of the i-th controlled object and the importance of the variables on the dynamic characteristics of the controlled object;

[0040] The second determining unit is used to determine the input variables and output variables based on the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object.

[0041] The cleaning unit is used to obtain the running data of the controlled object in the historical period based on the data types corresponding to the input variables and output variables, and to clean the running data using a data cleaning algorithm.

[0042] A partitioning unit is used to divide the cleaned running data into a training set and a validation set;

[0043] The third determining unit is used to determine the dynamic characteristic model of the controlled object based on the running data in the training set;

[0044] Where i∈[1~I], and I is the total number of controlled objects;

[0045] The data cleaning algorithms include: wavelet analysis algorithm, Wiener filtering algorithm, Kalman filtering algorithm, and neural network algorithm.

[0046] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method described in the first aspect embodiment.

[0047] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described in the first aspect.

[0048] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects.

[0049] This application proposes an adaptive control method and system for thermal power units oriented towards power plant intelligence. The method includes: establishing a thermal power unit type library and determining the controlled objects corresponding to each thermal power unit in the type library; using a data-driven intelligent TS fuzzy identification method to model the dynamic characteristics of each controlled object, obtaining a dynamic characteristic model for each controlled object; determining an adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic models of each controlled object; dynamically tuning the control parameters using a fuzzy tuning method or a swarm intelligence optimization algorithm; and performing adaptive control of the thermal power unit based on the dynamically tuned control parameters. The technical solution proposed in this application takes into account multiple objectives and constraints in the optimization of thermal power unit operation, such as safety, stability, economy, energy saving, and environmental protection, while also flexibly meeting various user needs and possessing scalability.

[0050] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0051] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0052] Figure 1 This is a flowchart of an adaptive control method for thermal power units aimed at power plant intelligence, according to an embodiment of this application;

[0053] Figure 2This is a schematic diagram illustrating the process of establishing a dynamic characteristic model of a controlled object according to an embodiment of this application;

[0054] Figure 3 This is a schematic diagram of a multi-objective adaptive predictive control strategy according to an embodiment of this application;

[0055] Figure 4 This is a structural diagram of an adaptive control system for thermal power units oriented towards power plant intelligence, provided according to an embodiment of this application.

[0056] Figure 5 This is a structural diagram of a modeling module provided according to an embodiment of this application;

[0057] Figure 6 This is a structural diagram of a third determining unit provided according to an embodiment of this application;

[0058] Figure 7 This is a structural diagram of a second determining module provided according to an embodiment of this application;

[0059] Figure 8 This is a structural diagram of a tuning module provided according to an embodiment of this application. Detailed Implementation

[0060] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0061] This application proposes an adaptive control method and system for thermal power units aimed at power plant intelligence. The method includes: establishing a thermal power unit type library and determining the controlled objects corresponding to each thermal power unit in the library; using a data-driven intelligent TS fuzzy identification method to model the dynamic characteristics of each controlled object, obtaining a dynamic characteristic model for each controlled object; determining an adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic models of each controlled object; dynamically tuning the control parameters using a fuzzy tuning method or a swarm intelligence optimization algorithm; and performing adaptive control of the thermal power unit based on the dynamically tuned control parameters. The technical solution proposed in this application takes into account multiple objectives and constraints in the optimization of thermal power unit operation, such as safety, stability, economy, energy saving, and environmental protection, while also flexibly meeting various user needs and possessing scalability.

[0062] The following description, with reference to the accompanying drawings, illustrates an adaptive control method and system for thermal power units aimed at power plant intelligence, based on an embodiment of this application.

[0063] Example 1

[0064] Figure 1 This is a flowchart of an adaptive control method for thermal power units aimed at power plant intelligence, according to an embodiment of this application. Figure 1 As shown, the method includes:

[0065] Step 1: Establish a thermal power unit type library and determine the controlled objects corresponding to each thermal power unit in the type library;

[0066] The controlled objects include: the boiler-turbine coordinated control system, the combustion subsystem, the boiler subsystem, the steam turbine subsystem, and the gas turbine subsystem.

[0067] It should be noted that thermal power generation has evolved to include various unit types. While similar design principles are followed in the research of control optimization strategies for these different types of units, specific details vary depending on the dynamic characteristics and control requirements of each unit. Currently, common typical unit types mainly include subcritical coal-fired units, ultra-supercritical coal-fired units, and combined cycle units. Furthermore, for these types of units, in actual operation, control optimization will be performed on the entire unit or only certain systems depending on different needs. For example, in the control of coal-fired units, either the entire boiler-turbine coordinated control system or specific subsystems such as steam temperature control, feedwater control, and combustion control can be selected. Therefore, step 1 can be specified as follows:

[0068] S1.1: Considering that the intelligent and digital transformation of power plants requires the popularization of various types of thermal power units, before designing the control strategy, a unit type library is first established with subcritical coal-fired units, ultra-critical (supercritical) coal-fired units and combined cycle units as components. New unit types can be added to this unit type library gradually as needed.

[0069] S1.2: In the operation and control of thermal power units, the ideal performance of the unit as a whole is based on the coordinated cooperation and mutual support of its various subsystems. The design of the control system should not only consider the control requirements of the entire unit but also ensure that each subsystem has excellent performance. Therefore, simply selecting the type of unit is insufficient to achieve the final control objective; the composition of the control system needs to be further refined to determine the controlled object under actual control requirements. Here, a typical type of unit is used as the parent node, and child nodes are created. These child nodes include the overall or partial systems of the boiler-turbine coordinated control system, combustion subsystem, boiler subsystem, steam turbine subsystem, and gas turbine subsystem.

[0070] Step 2: Use the data-driven intelligent TS fuzzy identification method to model the dynamic characteristics of each controlled object, and obtain the dynamic characteristic model of each controlled object;

[0071] In this embodiment of the disclosure, step 2 specifically includes:

[0072] Step 2-1: Determine the operating principle of the i-th controlled object and the importance of the variables to the dynamic characteristics of the controlled object;

[0073] Step 2-2: Determine the input and output variables based on the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object;

[0074] Steps 2-3: Based on the data types corresponding to the input and output variables, obtain the operation data of the controlled object during the historical period, and clean the operation data using a data cleaning algorithm;

[0075] Steps 2-4: Divide the cleaned running data into a training set and a validation set;

[0076] Steps 2-5: Determine the dynamic characteristic model of the controlled object based on the running data in the training set;

[0077] Where i∈[1~I], and I is the total number of controlled objects;

[0078] The data cleaning algorithms include: wavelet analysis algorithm, Wiener filtering algorithm, Kalman filtering algorithm, and neural network algorithm.

[0079] Furthermore, steps 2-5 specifically include:

[0080] Step 2-5-1: Based on the squirrel optimization algorithm, divide the running data in the training set to obtain the data partitioning results of the running data in the training set under each working condition;

[0081] Step 2-5-2: Determine the initial model of the controlled object based on the data partitioning results of the training set under various operating conditions;

[0082] Step 2-5-3: Use the model parameter identification method to determine the model parameters of the initial model and obtain the dynamic characteristic model of the controlled object.

[0083] It should be noted that, in order to ensure the ideal operational optimization capability of the entire ultra-supercritical unit, this embodiment takes the boiler-turbine coordination system as the controlled object.

[0084] Based on the controlled object established in step 1, namely the ultra-supercritical unit boiler-turbine coordination system, the dynamic characteristics of the boiler-turbine unit are modeled in step 2 using the data-driven intelligent TS fuzzy identification method.

[0085] The TS fuzzy model identification method mainly consists of two parts: antecedent identification and consequent identification. As a data-driven modeling method, TS fuzzy modeling first clusters the sampled unit operation data in the antecedent identification process, which enables a simple division of data under different operating conditions. In the consequent identification, parameters of the sub-models corresponding to each data cluster are identified based on a certain form of dynamic characteristic model expression. The antecedent and consequent parts of TS fuzzy modeling together constitute the following fuzzy rules based on If-Then statements to approximate the nonlinear characteristics of complex systems:

[0086]

[0087] Among them, R i Let represent the fuzzy rule i; the generalized input variable x consists of all input and output variables, i.e., x = [u, y]. T c i and r i Let these represent the center and radius of the i-th cluster, respectively; θ is the output of the sub-model corresponding to the i-th cluster; i is the parameter matrix of this sub-model; n is the number of clusters.

[0088] For example, such as Figure 2 As shown, S2.1: Based on the operating principle of the furnace-machine coordination system and the importance of specific variables to its dynamic characteristics, the input and output variables are determined to obtain a simplified model.

[0089] In this embodiment, the ultra-supercritical unit's boiler-generator coordination system comprises a once-through boiler, a steam turbine, a generator and pulverizing system, an air supply system, and a steam-water circulation system, among other main and auxiliary systems. Raw coal is fed into the boiler furnace for combustion after passing through the pulverizing system. Feedwater, heated by the heat generated from boiler combustion, produces high-temperature, high-pressure gas that enters the steam turbine to drive the turbine blades, converting thermal energy into mechanical energy. This mechanical energy further drives the generator rotor to generate electricity. Simultaneously, the exhaust steam from the turbine enters the steam-water system for the next cycle. Based on the degree of influence of each variable on the overall operating status and performance of the boiler-generator coordination system in the above stages, the coal feed rate u is selected accordingly. B Water supply D fw and main steam valve opening μ T The input vector consists of u = [u1, u2, u3] = [u B D fw ,μ T Output power P, main steam pressure p st The output vector y = [y1, y2, y3] = [P, p] is composed of the outlet temperature T of the steam-water separator. st If we form a simplified model with three inputs and three outputs, then the number of input variables a and the number of output variables b are both 3.

[0090] S2.2: Sample 5000 sets of field operation data X within the unit's 30%-100% load range at 1-second intervals. 5000 = [x1', x2', ..., x 5000 For model training, N is set to 5000. Then, 800 sets of data from each of the 30%, 60%, and 90% rated load segments (the load segments can be freely selected) are selected as validation data, where m = 3 and M = 800.

[0091] S2.3: Considering the complexity of the operating environment of thermal power units and the frequent occurrence of internal and external disturbances, in order to ensure the speed and accuracy of model training results, it is necessary to first clean the sampled unit operating data to eliminate the influence of adverse factors such as high-frequency noise. Therefore, a data cleaning algorithm library is formed by introducing wavelet analysis, Wiener filtering, Kalman filtering, neural networks, and other methods. This library can not only adaptively select algorithms according to different filtering requirements, but also continuously enrich and update the algorithm library based on the development of data cleaning technology.

[0092] Given that wavelet analysis is a relatively mature method both theoretically and practically, this embodiment selects wavelet analysis from a data cleaning algorithm library for preprocessing training and validation data. The specific steps are readily available and will not be elaborated here. The filtered training data sequence is denoted as X. N .

[0093] S2.4: In TS fuzzy model identification, the first step is to process the cleaned training data X. N =[x1,x2,…,x N Spatial partitioning, or data clustering, is a crucial step in ensuring consistent modeling accuracy under various operating conditions. This paper proposes an intelligent automatic clustering method based on the Squirrel Optimization Algorithm for data partitioning. This method eliminates the need for manually setting the number of clusters; it automatically determines the number of clusters and the optimal cluster centers by combining a data similarity threshold constant, an upper limit for cluster size, and swarm intelligence iterative optimization.

[0094] Before starting clustering, initialize the number of clusters n=1, and then complete the automatic clustering of the data to be identified by following these steps:

[0095] S2.4.1: Algorithm initialization and data vector similarity calculation. The dimension of the data vector is D = a + b = 6. The first element x1 of all training data sequences to be clustered is taken as the current cluster center c. n Then, using equation (2), the data vectors to be clustered and c are obtained sequentially. n Similarity P between o :

[0096]

[0097] Where: the similarity coefficient ξ is a constant of 0.01; d(c n ,x i ) represents the cluster center c n and data vector x i The Euclidean distance between them.

[0098] The above formula is normalized as follows to eliminate the impact of differences in magnitude between data variables of different dimensions on the clustering results:

[0099]

[0100] In the formula: P max (c n ,x i ), P min (c n ,x i ) represent P respectively o (c n ,x i The maximum and minimum values ​​of ).

[0101] S2.4.2: Establishing the clustering scale based on a data similarity threshold. A similarity threshold λ = 0.65 is introduced to determine the clustering size of a given data vector x. i The correlation between (i = 1, 2, ..., N) and the current cluster, if P(c n ,x i If x ≥ λ, then x i If an element is assigned to the current cluster, then the number of elements in that cluster is N. n =N n +1.

[0102] To ensure the validity of the clustering results, the minimum size of each cluster is defined as N. L =200. If N n >N L Then consider c n The initial cluster centers are determined, and all elements in this cluster are removed from the data space to be clustered. At this point, the number of data vectors to be clustered is N = NN. n If the first element x1 of the data sequence is moved to the end, proceed to the next step; otherwise, return to step 4-1) to find the initial cluster center of the current cluster.

[0103] S2.4.3: Intelligent optimization of clustering centers based on the squirrel optimization algorithm. Similar to most swarm intelligence optimization algorithms in biological evolution, the squirrel algorithm is inspired by the population foraging process, in which individual squirrels gradually shorten the distance to the food through continuous position optimization. At this time, the food position can be regarded as the location of the optimal clustering center in the data space partitioning, and each squirrel individual is equivalent to an optimizable clustering center.

[0104] It should be noted that E1. Initialization of the parameters and population positions of the squirrel optimization algorithm. During the foraging process of squirrels, the S trees in the forest can be divided according to their priority of food preference. Among them, 1 hickory tree is the location of the optimal food position, S a oak trees and S - S a - 1 other common tree species. To simplify the search process, let the size of the squirrel population be equal to the number of trees in the forest, both being S (S < N), and there is a one-to-one correspondence between squirrel individuals and trees. In this embodiment, let S = 20, S a = 4. Then, the positions of the squirrel population can be initialized based on the current clustering center c n :

[0105]

[0106] Among them: represents the clustering center vector corresponding to the i-th squirrel under the current clustering, that is, the initial position vector of the squirrel population; c U and c L represent the upper and lower bounds of the clustering center search range. Here, c U = max{x1, x2,..., x N}. c L = min{x1, x2,..., x N}.

[0107] E2. Determination of the fitness function and sorting rules. In the clustering process based on the squirrel optimization algorithm, the effectiveness of the result can be judged by the average distance between each data vector and the clustering center in the current clustering. Therefore, this average distance can be set as the fitness function shown in Equation (5), which corresponds to the average distance d mean between the position vector of the squirrel individual and the data vector in the squirrel optimization algorithm.

[0108]

[0109] Among them: represents the average distance between the position of the l-th squirrel in the squirrel population and the data vectors included in this clustering during the optimization process of the current clustering center; N n represents the number of data vectors included in the current clustering; This is the j-th data vector in the current cluster n.

[0110] The fitness of individuals in the squirrel population was calculated according to equation (5) and sorted in ascending order, thus dividing them into 3 categories: the squirrel with the best fitness, namely the squirrel M closest to the hickory tree. h Located near an acorn tree, fitness is 2 to (S) a Squirrel M (+1) a The common squirrel M corresponding to the remaining elements of the fitness sequence n .

[0111] E3. Dynamic Updates of Squirrel Population Location. To continuously shorten the distance to the optimal food source, the hickory tree, individuals within a squirrel population need to optimize their location by moving from acorn trees to hickory trees, or from ordinary trees to hickory or acorn trees. The three dynamic update mechanisms of squirrel location in the above process are shown in the following equations:

[0112]

[0113]

[0114]

[0115] Where: k represents the current time; R1, R2, and R3 are random numbers between 0 and 1; P d This represents the probability of a squirrel being hunted, which is set to 0.45 here; G represents the position vector corresponding to the hickory tree, i.e., the optimal value in the fitness ranking at time k; c The movement coefficient in the squirrel's position update; the movement distance d g It can be calculated using the following formula:

[0116] d g =h g / tan(θ)

[0117] tan(θ)=D / L (9)

[0118] Where: coefficient h g Set to a constant of 8; θ represents the slip angle; the expressions for drag D and lift L are as follows:

[0119]

[0120]

[0121] Where: air density ρ = 1.204 kg / m³ -3 Air velocity V = 5.25 m / s -1 The surface area of ​​the object is A = 154 cm². 2 Drag coefficient CD =0.6; Lift coefficient C L =0.9.

[0122] E4. Seasonally Adaptive Intelligent Optimization. Since the ripening of fruits and vegetables is affected by seasonal factors such as temperature, light, and rainfall, squirrels are more active in foraging during autumn. At this time, the seasonally adaptive intelligent operation shown in the following formula is needed to avoid premature ripening convergence.

[0123]

[0124] With the arrival of winter, squirrels need to consume the food they stored up in the fall to survive the winter. After a winter, their food resources are almost depleted. Therefore, in the new year, squirrels need to reassess their food resources. First, we introduce the following minimum seasonal constant s. min As a threshold for determining whether winter has ended:

[0125]

[0126] In the formula: k represents the current iteration time; k max This represents the maximum number of iterations, with a value of 30.

[0127] like To signify the end of winter, the squirrel population is relocated in the search space using the following formula:

[0128]

[0129] in: and This indicates the upper and lower bounds of the current cluster center search range; levy() represents the Levy flight function with the following expression.

[0130]

[0131] In the formula: R a = rand[0,1]; R b = rand[0,1]; constant coefficient β = 1.5; the expression for α is:

[0132]

[0133] Where: Γ() is the factorial function, i.e., Γ(x) = (x-1)!.

[0134] The final cluster centers are obtained by continuously iterating through steps E1-E4 until the maximum number of iterations k is reached. max Alternatively, if the average distance between the cluster center and the data vectors in the cluster meets the requirement, the optimal squirrel position obtained at this point is considered the optimal choice for the current cluster center.

[0135] S2.4.4: Calculation of the current cluster radius. The cluster radius should be set as the farthest distance between each element in the current cluster and the cluster center. Therefore, the radius r of the current cluster is obtained by the following formula. n :

[0136]

[0137] S2.4.5: Determining the termination condition for clustering. After completing the current clustering, the number of remaining data vectors in the data space to be clustered is N = NN. n The value of N at this time is compared with the lower limit of the cluster size N. L For comparison, if N≤N L If the clustering process ends, the centers of the resulting clusters are determined to be... And radius r = {r1, r2, ..., r n If n is the optimal clustering result, then n = n + 1, and return to step 4-1) to continue clustering until the termination condition is met.

[0138] S2.5: The completion of the above data clustering process has also realized the division of different operating conditions of the unit. Next, a model parameter identification method needs to be introduced to establish the sub-models corresponding to each cluster. This process needs to fully consider the design requirements of the unit's subsequent control strategy. Different types of control strategies are often adapted to different model forms of the controlled object. This embodiment takes predictive control, one of the typical advanced control methods, as an example for explanation.

[0139] Predictive control methods are mainly classified into three categories: dynamic matrix control, generalized predictive control (GPC), and model predictive control (MPC). Considering the complexity of the controlled object in industrial control processes, GPC and MPC are generally the primary methods used in theoretical and applied research. GPC is based on a controlled autoregressive moving average model of the dynamic characteristics of the controlled object, while MPC is suitable for the state-space representation of the controlled object model. The least squares method and subspace identification method are used respectively for these two model forms.

[0140] Here, we choose MPC and its corresponding subspace identification method. Since the subspace identification method is relatively mature, we will not go into the specific steps.

[0141] S2.6: After obtaining the sub-models corresponding to each cluster for the controlled object, based on the degree to which the input data at each time step belongs to different clusters, the outputs of the corresponding sub-models are fuzzily weighted using the following formula to obtain the global output of the model built at the current time step:

[0142]

[0143] Where: the membership degree μ in equation (19) iIt can be calculated using equation (19):

[0144]

[0145] Step 3: Determine the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic model of each controlled object;

[0146] In this embodiment of the disclosure, step 3 specifically includes:

[0147] Step 3-1: Use the dynamic characteristic model of the controlled object as the prediction model;

[0148] Step 3-2: Determine the objective function and constraints of the predictive control rolling optimization process based on the actual operation optimization requirements of the thermal power unit;

[0149] Step 3-3: Based on the objective function and constraints, optimize the solution of the control law to obtain the optimal control law.

[0150] It should be noted that, based on the dynamic characteristic model of the controlled object established in step 2, an adaptive control strategy under multiple objectives and constraints is designed in step 3.

[0151] To further promote the digital and intelligent transformation of the thermal power generation industry, unit control optimization must not only consider the deep and rapid peak-shaving requirements under dual carbon objectives, but also meet the requirements for safe and stable operation and energy conservation during operation. Therefore, the design of control strategies requires reasonable coordination of multiple control objectives and constraints, and comprehensive consideration of the unit's dynamic characteristics and the advantages of different control methods to achieve adaptive selection and diversified design of control algorithms, flexibly meeting the diverse needs of users.

[0152] For example, S3.1: Adaptively select the control method based on the characteristics of the actual control process. Currently, the most commonly used advanced control methods in the theoretical research and practical exploration of thermal power unit control optimization include predictive control, fuzzy control, and automatic disturbance rejection control. These control strategies can be combined to form a thermal power unit control method library, and then adaptively select the control method according to the dynamic characteristics of the controlled object and the control optimization requirements.

[0153] Given the significant advantages of predictive control algorithms in constraint handling and multi-objective optimization, this embodiment uses the MPC algorithm as an example to design the control strategy for the controlled object. Through the coordinated use of predictive models, rolling optimization, and feedback correction, it achieves safe, stable, economical, and environmentally friendly operation control of the unit. Figure 3 As shown.

[0154] Considering the safety, stability, energy saving, and low carbon requirements of power plant intelligence under the "dual carbon" background, a comprehensive objective function for the control process is established with setpoint tracking error, coal consumption rate, and carbon emissions during unit operation as sub-objectives:

[0155] J(Δu)=γ1J1(Δu)+γ2J2(Δu)+γ3J3(Δu) (20)

[0156] Where: Δu is the manipulated variable of the control process; J i (Δu)(i=1,2,3) represents the objective function of each sub-model; γ i (i = 1, 2, 3) represents the weighted sum of the objective functions of each sub-model.

[0157] To eliminate the impact of the order-of-magnitude differences in the performance indicators representing different sub-optimization objectives, equation (20) is normalized to obtain the following result:

[0158]

[0159] in: J i The normalized result of (Δu)(i=1,2,3) is expressed as:

[0160]

[0161] In the formula: and Representing J respectively i (Δu) represents the upper and lower bounds of its value.

[0162] The safety and stability of the control system are directly related to the magnitude and rate of change of the manipulated variable. Therefore, it is necessary to introduce magnitude constraints and rate constraints of the manipulated variable on the basis of the above multi-objective function. The rate of change of the manipulated variable is regarded as the increment of its magnitude per unit time. In this embodiment, the above constraints are set as -10≤u(t)≤10 and -1≤Δu(t)≤1, respectively.

[0163] In addition, corresponding constraints can be added based on factors related to operational flexibility such as peak shaving depth and speed of the unit, as well as energy saving, efficiency improvement, and environmental protection. These are not listed one by one in this embodiment.

[0164] S3.3: Based on the prediction model established in S2 and the multi-objective optimization problem constructed in S3.2, the control law is optimized and solved. Different types of predictive control methods often employ different solution methods. In this embodiment, the control law of the multi-objective intelligent MPC is obtained by relying on matrix operations and quadratic programming.

[0165] In addition to the classical solution methods mentioned above, intelligent methods such as swarm intelligence optimization algorithms can also be used. These methods do not rely on the prediction model of the controlled object or the specific expression of the control objective function and constraints. They can directly iteratively optimize the manipulated variables to obtain the optimal control law, and the algorithm settings can be flexibly improved according to actual solution requirements. They are simple to implement, highly applicable, and easy to promote. However, compared to classical solution methods, due to a lack of extensive application exploration, the stability and robustness of these methods in practical engineering applications have not been rigorously verified. Therefore, the appropriate solution method can be adaptively selected according to user needs.

[0166] For ease of description, this embodiment solves the control law using the traditional quadratic programming method. Let the prediction time domain and control time domain of the constructed predictive control strategy be T, respectively. P and T C (T P >T C The sub-objective weighting coefficients in the initial control objective function are γ1 = 0.4, γ2 = 0.3 and γ3 = 0.3. The control parameters obtained above determine the model as the prediction model. After matrix operations and iterative optimization, equation (21) is transformed into the following quadratic programming form:

[0167]

[0168] Where H and G represent parameter matrices, respectively;

[0169]

[0170] Then, the control law sequence ΔU(t) at the current time t is obtained by solving the standard quadratic programming problem, and the first element of the sequence is fed into the control system as the current control law Δu(t).

[0171] Step 4: Dynamically tune the control parameters using fuzzy tuning methods or swarm intelligence optimization algorithms;

[0172] In this embodiment of the disclosure, step 4 specifically includes:

[0173] Step 4-1: Determine the input and output variables of the fuzzy rule;

[0174] Step 4-2: Construct fuzzy rules for parameter optimization based on the input and output variables;

[0175] Step 4-3: Tune the control parameters based on the constructed parameter-oriented optimization fuzzy rules.

[0176] It should be noted that, considering the variable operating conditions and complex environment of thermal power units during operation, fixed control algorithm parameters often cannot guarantee that the control system maintains ideal performance at all times. In multi-objective optimization problems, there are often conflicts between the sub-objectives. If their weight coefficients are set to constants, as the control process progresses, the sub-objectives will be unable to coordinate their relationships, significantly sacrificing one or more performance indicators. Therefore, intelligent methods can be used to dynamically tune the control parameters in real time.

[0177] The fuzzy tuning method does not require precise quantification of each variable or indicator. It provides parameter results by taking values ​​in a fuzzy range, which is simple and easy to expand flexibly. Therefore, this embodiment relies on this method to perform real-time dynamic tuning of parameters.

[0178] For example, S4.1: Determination of input and output variables of fuzzy rules. The fuzzy rules constructed here are used to obtain the weighting coefficients of each sub-objective function of the comprehensive optimization objective shown in equation (21). The output variables of the fuzzy rules can be set as {γ1γ2γ3}. Since the weight coefficients are between 0 and 1, the fuzzy universe of discourse of each output variable is [0,1].

[0179] The weighting coefficients of sub-objectives need to be dynamically adjusted in real time based on their corresponding performance indicators, such as setpoint tracking accuracy, power generation coal consumption, and carbon emissions. The feedback results of each normalized sub-objective at the current moment can be displayed. As the input vector for fuzzy rules, since each index has been normalized, its fuzzy universe is also [0,1].

[0180] S4.2: Construction of Fuzzy Rules for Parameter Optimization. As the core of real-time dynamic adjustment of algorithm parameters, the rationality of fuzzy rules directly determines the effectiveness of algorithm adjustment.

[0181] S4.2.1: Set fuzzy set labels to divide the value range of input and output variables. To make the division results more detailed, seven fuzzy set labels, namely negative large (NB), negative medium (NM), negative small (NS), median (ZO), positive small (PS), positive medium (PM), and positive large (PB), are used to record the magnitude of variable values ​​in ascending order.

[0182] S4.2.2: Before constructing fuzzy rules, the form of the membership function for each variable needs to be set. Generally, Gaussian function, trigonometric function, square wave function, etc., can be selected according to the function characteristics and actual needs. Given the ideal statistical performance of Gaussian function, the membership function in this embodiment adopts the form of Gaussian function.

[0183] S4.2.3: Determining the adjustment principle of the weight coefficients serves as the fundamental rule for constructing fuzzy rules. In the process of optimizing the control law, the smaller the objective function value, the more ideal the control effect. Therefore, if the sub-optimization objective... As the corresponding performance index value increases, its corresponding weighting coefficient γ needs to be increased. i The values ​​of (i = 1, 2, 3) are determined, and vice versa; if there is a significant difference in the values ​​of two different sub-targets at a certain moment, such as and Taking NS and PM respectively, it means The corresponding performance degradation necessitates increasing the weighting coefficient γ. i The value of . Based on the above adjustment principles and combined with expert experience, the fuzzy rules shown in equation (24) can be obtained. The number of complete fuzzy rules is 7×7×7=343, which will not be listed here.

[0184]

[0185] In addition to the fuzzy tuning method mentioned above, the controller parameters can also be selected through swarm intelligence optimization algorithms. This is not limited to the sub-objective weighting coefficients, but also applies to other parameters in the algorithm.

[0186] Step 5: Perform adaptive control of the thermal power unit based on the dynamically tuned control parameters.

[0187] In this embodiment, the effectiveness of the proposed method is verified through simulation testing and statistical analysis, laying the foundation for its practical application in the intelligent construction of power plants. Specifically, this includes:

[0188] This embodiment takes the boiler-turbine coordination system of a 1000MW power plant in northern China as the controlled object. First, the dynamic characteristics of the boiler-turbine coordination system are modeled by the data-driven intelligent TS fuzzy identification method.

[0189] 1) Based on the simplified three-input three-output model, 5,000 sets of field operation data of a 1,000MW ultra-supercritical unit in a power plant in northern China were sampled at 1s intervals within the 30%-100% load range for model training. Then, 800 sets of data from the 30%, 60%, and 90% rated load ranges were selected as validation data.

[0190] 2) After the data cleaning, data clustering, parameter identification, and sub-model output weighting operations described above, the final model output of the unit is obtained. To evaluate the accuracy of the modeling results, the mean absolute error (MAE), root mean square error (MSE), and fitting accuracy (R²) of each output in the modeling process are used as performance indicators, and the corresponding fitting curves between the model output and the actual operating output are plotted.

[0191] 3) The input variable portions of 800 sets of validation data vectors for each of the 30%, 60%, and 90% rated load ranges were used as inputs to the established model. The corresponding model outputs were then compared with the actual outputs in the validation data, and the model output fitting curves for the validation phase were plotted. MAE, MSE, and R2 were used as performance indicators. Since the model validation process involves multiple operating conditions, the performance indicators were taken as the average values ​​of several validation operating conditions.

[0192] 3) List and record the performance indicators during the model identification and verification process and perform statistical analysis.

[0193] Table 1 Statistical Results of Performance Indicators in the Identification and Verification Process of the 1000MW Unit Model

[0194]

[0195] 5) As can be seen from the results in Table 1, after adopting the modeling method used in this embodiment, the output of the model can approximate the actual operating data of the unit with a very small average error and root mean square error. Moreover, the accuracy of each output variable in the model identification and verification process is higher than 95%, and the results are ideal.

[0196] 6) Based on the control parameters of the boiler-machine coordination system obtained above, determine the model, verify the feasibility and effectiveness of the designed multi-objective intelligent predictive control strategy through setpoint tracking experiment, anti-disturbance test and robustness test, and define the mean absolute tracking error (MAE), mean coal consumption rate (ME) and mean carbon emission (MC) as performance indicators by comparison with the traditional MPC strategy.

[0197] 7) In the tracking performance test, the output vector under 30% load condition is used as the initial set value. At the 20th second of the simulation process, the set value is changed to the output vector under 40% load condition. The variable response curve of the control process is plotted, and the performance index values ​​are recorded in a list.

[0198] 8) In the disturbance suppression capability test, the output vector under 30% load condition is used as the initial set value. At the 20th second of the simulation process, a disturbance signal of 10% is applied to the output power. The response curves of each variable during the disturbance suppression process are plotted and the performance index values ​​are recorded in a table.

[0199] 9) In the robustness test based on the Monte Carlo test, the number of test samples was set to 100. The boiler-turbine coordination system model under 50% load was used as the nominal model. At the 20th second of the simulation process, a disturbance signal of ±10% was applied to any parameter of the nominal model. The response curves of each variable during the robustness test were plotted, the dispersion of the curves was observed, and the performance index values ​​were recorded in a list.

[0200] 10) List and analyze the performance indicators compared during the feasibility and effectiveness verification process of the above control algorithm.

[0201] Table 2. Statistical results of simulation test performance indicators for multi-objective intelligent predictive control strategies.

[0202]

[0203] 11) The results in the table show that, in the tracking test, disturbance suppression test and robustness test, compared with the traditional classic MPC strategy, under the action of the designed control strategy, each output variable can track the change of the set value with higher accuracy, and has superior performance in reducing coal consumption and carbon emissions. It has significant advantages in achieving precise and rapid load regulation, disturbance suppression and energy saving and emission reduction of ultra-supercritical units.

[0204] In summary, the adaptive control method for thermal power units oriented towards power plant intelligence proposed in this embodiment has the following advantages: 1. This invention is based on the intelligent and low-carbon transformation needs of thermal power units under the "dual-carbon" strategic goal. It constructs a complete adaptive control optimization strategy that includes key aspects such as unit selection, controlled object selection, dynamic characteristic modeling, intelligent control strategy design, and parameter dynamic tuning. The strategy has a clear roadmap and rich functionality, providing an effective reference for the design of intelligent control strategies for thermal power units under my country's "dual-carbon" background. 2. In the design of the control method of this invention, multiple objectives and constraints such as safety, stability, economy, energy saving, and environmental protection in the operation optimization of thermal power units can be simultaneously considered. Furthermore, advanced intelligent methods such as fuzzy adjustment and swarm intelligence optimization are introduced into the algorithm, comprehensively empowering the intelligent construction of thermal power units and the digital low-carbon transformation of the entire power generation industry. 3. This invention widely applies advanced intelligent algorithms such as swarm intelligence optimization algorithms to multiple aspects such as data cleaning, data clustering, modeling, and control parameter optimization, fully expanding the application scope of such algorithms and maximizing their functional advantages and flexible application value. 4. The control method proposed in this invention fully considers the differences in characteristics of different controlled objects and the changes in control requirements during actual production and operation. The strategy selection in each stage of modeling, control, and optimization is diversified and intelligent, flexibly meeting various user needs. At the same time, new elements can be added to the unit type library or various algorithm libraries, making it valuable for reference in the field of new energy power generation and the construction of new power systems, demonstrating good scalability.

[0205] Example 2

[0206] Figure 4 This is a structural diagram of an adaptive control system for thermal power units aimed at power plant intelligence, provided according to an embodiment of this application. Figure 4 As shown, the system includes:

[0207] The first determining module 100 is used to establish a thermal power unit type library and determine each controlled object corresponding to each thermal power unit in the type library;

[0208] Modeling module 200 is used to model the dynamic characteristics of each controlled object using a data-driven intelligent TS fuzzy identification method, and obtain the dynamic characteristic model of each controlled object.

[0209] The second determining module 300 is used to determine the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic model of each controlled object.

[0210] Tuning module 400 is used to dynamically tune control parameters using fuzzy tuning methods or swarm intelligence optimization algorithms;

[0211] The control module 500 is used for adaptive control of thermal power units based on dynamically tuned control parameters.

[0212] In this embodiment of the disclosure, the controlled object includes: a boiler-turbine coordinated control system, a combustion subsystem, a boiler subsystem, a steam turbine subsystem, and a gas turbine subsystem.

[0213] In the embodiments disclosed herein, such as Figure 5 As shown, the modeling module 200 includes:

[0214] The first determining unit 201 is used to determine the operating principle of the i-th controlled object and the importance of the variables on the dynamic characteristics of the controlled object;

[0215] The second determining unit 202 is used to determine the input variables and output variables based on the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object.

[0216] The cleaning unit 203 is used to obtain the running data of the controlled object in the historical period based on the data types corresponding to the input variables and output variables, and to clean the running data using a data cleaning algorithm.

[0217] The partitioning unit 204 is used to divide the cleaned running data into a training set and a validation set;

[0218] The third determining unit 205 is used to determine the dynamic characteristic model of the controlled object based on the running data in the training set;

[0219] Where i∈[1~I], and I is the total number of controlled objects.

[0220] Furthermore, such as Figure 6 As shown, the third determining unit 205 includes:

[0221] Sub-unit 2051 is used to divide the running data in the training set based on the squirrel optimization algorithm to obtain the data division results of the running data in the training set under various working conditions;

[0222] The first determining subunit 2052 is used to determine the initial model of the controlled object based on the data partitioning results of the running data in the training set under various working conditions.

[0223] The second determining subunit 2053 is used to determine the model parameters of the initial model using the model parameter identification method, so as to obtain the dynamic characteristic model of the controlled object.

[0224] In the embodiments disclosed herein, such as Figure 7 As shown, the second determining module 300 includes:

[0225] The fourth determining unit 301 is used as the dynamic characteristic model of the controlled object as the prediction model;

[0226] The fifth determining unit 302 is used to determine the objective function and constraints of the predictive control rolling optimization process based on the actual operation optimization requirements of the thermal power unit;

[0227] The optimization solution unit 303 is used to optimize the solution of the control law based on the objective function and constraints to obtain the optimal control law.

[0228] In the embodiments disclosed herein, such as Figure 8 As shown, the tuning module 400 includes:

[0229] The sixth determining unit 401 is used to determine the input and output variables of the fuzzy rule;

[0230] Construction unit 402 is used to construct fuzzy rules for parameter optimization based on the input and output variables;

[0231] Tuning unit 403 is used to tune control parameters based on constructed parameter-oriented optimization fuzzy rules.

[0232] In summary, the adaptive control system for thermal power units provided by this disclosure takes into account multiple objectives and constraints in the optimization of thermal power unit operation, such as safety, stability, economy, energy saving and environmental protection, while flexibly meeting various user needs and having scalability.

[0233] Example 3

[0234] To implement the above embodiments, this disclosure also proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method described in Embodiment 1.

[0235] Example 4

[0236] To implement the above embodiments, this disclosure also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in Embodiment 1.

[0237] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0238] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0239] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A power plant intelligent-oriented thermal power unit adaptive control method, characterized in that, The method includes: Establish a thermal power unit type library and determine the controlled object corresponding to each thermal power unit in the type library, wherein the controlled object is the ultra-supercritical unit boiler-turbine coordination system; A data-driven intelligent TS fuzzy identification method is used to model the dynamic characteristics of the controlled object, resulting in a dynamic characteristic model of the controlled object, including: Determine the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object; The input and output variables are determined based on the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object. Among them, the coal feed rate, water feed rate and main steam valve opening are selected as input variables, and the output power, main steam pressure and steam-water separator outlet temperature are selected as output vectors. Based on the data types corresponding to the input and output variables, the operation data of the controlled object during the historical period is obtained, and the operation data is cleaned using a data cleaning algorithm, wherein the data cleaning algorithm includes: wavelet analysis algorithm, Wiener filtering algorithm, Kalman filtering algorithm and neural network algorithm; The cleaned runtime data is divided into a training set and a validation set; The dynamic characteristic model of the controlled object is determined based on the running data in the training set; Based on the dynamic characteristic models of each controlled object, the adaptive control strategy under multiple objectives and constraints is determined; Dynamic tuning of control parameters is performed using fuzzy tuning methods or swarm intelligence optimization algorithms; Adaptive control of thermal power units based on dynamically tuned control parameters; The step of determining the dynamic characteristic model of the controlled object based on the running data in the training set includes: The squirrel optimization algorithm is used to divide the running data in the training set, and the data division results of the running data in the training set under various working conditions are obtained. A parameter identification method is introduced to establish sub-models of the controlled object corresponding to each cluster; After obtaining the sub-models of each cluster corresponding to the controlled object, the outputs of the corresponding sub-models are fuzzily weighted according to the degree to which the input data at each time belongs to different clusters, so as to obtain the global output of the model built at the current time. The determination of the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic models of each controlled object includes: The dynamic characteristic model of the controlled object is used as the prediction model; A comprehensive objective function for the control process is established, with setpoint tracking error, coal consumption rate, and carbon emissions during unit operation as sub-objectives, and is expressed as: J (Δu) = γ1J1 (Δu) + γ2J2 (Δu) + γ3J3 (Δu) Among them, J i (Δu) represents J i The normalized result of (Δu), i=1,2,3, J i (Δu) represents the objective function of each sub-model, γ i This represents the weighted sum of the objective functions of each sub-model; Based on the comprehensive objective function, amplitude and rate constraints of the manipulated variable are introduced, and the rate of change of the manipulated variable is regarded as the increment of its amplitude per unit time. Based on the prediction model, the comprehensive objective function, and constraints, the control law is optimized and solved to obtain the optimal control law.

2. The method as described in claim 1, characterized in that, The dynamic tuning of control parameters using the fuzzy tuning method includes: Determine the input and output variables of the fuzzy rule; Construct fuzzy rules for parameter optimization based on the input and output variables; The control parameters are tuned based on the constructed fuzzy rules for parameter optimization.

3. An adaptive control system for thermal power units aimed at intelligent power plants, characterized in that, The system includes: The first determining module is used to establish a thermal power unit type library and determine the controlled object corresponding to each thermal power unit in the type library, wherein the controlled object is the ultra-supercritical unit boiler-turbine coordination system. The modeling module is used to model the dynamic characteristics of the controlled object using a data-driven intelligent TS fuzzy identification method, resulting in a dynamic characteristic model of the controlled object, including: Determine the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object; The input and output variables are determined based on the operating principle of the controlled object and the importance of the variables to the dynamic characteristics of the controlled object. Among them, the coal feed rate, water feed rate and main steam valve opening are selected as input variables, and the output power, main steam pressure and steam-water separator outlet temperature are selected as output vectors. Based on the data types corresponding to the input and output variables, the operation data of the controlled object during the historical period is obtained, and the operation data is cleaned using a data cleaning algorithm, wherein the data cleaning algorithm includes: wavelet analysis algorithm, Wiener filtering algorithm, Kalman filtering algorithm and neural network algorithm; The cleaned runtime data is divided into a training set and a validation set; The dynamic characteristic model of the controlled object is determined based on the running data in the training set; The second determining module is used to determine the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic model of each controlled object; The tuning module is used to dynamically tune control parameters using fuzzy tuning methods or swarm intelligence optimization. The control module is used for adaptive control of thermal power units based on dynamically tuned control parameters; The step of determining the dynamic characteristic model of the controlled object based on the running data in the training set includes: The squirrel optimization algorithm is used to divide the running data in the training set, and the data division results of the running data in the training set under various working conditions are obtained. A parameter identification method is introduced to establish sub-models of the controlled object corresponding to each cluster; After obtaining the sub-models of each cluster corresponding to the controlled object, the outputs of the corresponding sub-models are fuzzily weighted according to the degree to which the input data at each time belongs to different clusters, so as to obtain the global output of the model built at the current time. The determination of the adaptive control strategy under multiple objectives and constraints based on the dynamic characteristic models of each controlled object includes: The dynamic characteristic model of the controlled object is used as the prediction model; A comprehensive objective function for the control process is established, with setpoint tracking error, coal consumption rate, and carbon emissions during unit operation as sub-objectives, and is expressed as: J (Δu)=γ1J1 (Δu)+γ2J2 (Δu)+γ3J3 (No) Among them, J i (Δu) represents J i The normalized result of (Δu), i=1,2,3, J i (Δu) represents the objective function of each sub-model, γ i This represents the weighted sum of the objective functions of each sub-model; Based on the comprehensive objective function, amplitude and rate constraints of the manipulated variable are introduced, and the rate of change of the manipulated variable is regarded as the increment of its amplitude per unit time. Based on the prediction model, the comprehensive objective function, and constraints, the control law is optimized and solved to obtain the optimal control law.

4. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1-2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-2.

Images