Heavy duty gas turbine intelligent control method integrating error active disturbance rejection control and improved hyperbolic sine and cosine optimizer

Abstract

The invention discloses a heavy duty gas turbine intelligent control method integrating error active disturbance rejection control and an improved hyperbolic sine and cosine optimizer. The heavy duty gas turbine has the characteristics of numerous monitoring parameters, strong nonlinearity and high coupling degree. The method is used for solving the problem that the control effect is not ideal when the traditional PID control strategy faces the system. The method comprises the steps of collecting and preprocessing data of the heavy-duty gas turbine, establishing a Wiener model of the heavy-duty gas turbine based on variational Bayesian identification, designing an extended state observer and an error active-disturbance-rejection control strategy, and designing a multi-target operation optimization target function of the heavy-duty gas turbine; and then solving the optimal solution of the decision variable of the heavy duty gas turbine controller by using an improved hyperbolic sine and cosine optimizer. The invention provides a scheme and thought for the intelligent control method of the heavy duty gas turbine.

Description

Technical Field

[0001] The present invention belongs to the field of heavy-duty gas turbine intelligent control, and in particular relates to a heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and an improved hyperbolic sine-cosine optimizer. Background Art

[0002] Heavy-duty gas-fired generator sets are transitional energy and regulating power sources for achieving the "dual carbon" goals and building a new power system with new energy as the main body. They are one of the key energy conversion equipment for transforming high-carbon energy into low-carbon energy. They can absorb new energy and enhance the flexible operation capabilities of the power system. They have the advantages of excellent peak-shaving performance, high efficiency and low emissions.

[0003] Heavy-duty gas turbines are characterized by numerous monitoring parameters, strong nonlinearity, and a high degree of coupling, resulting in distinct operating mechanisms under different operating conditions. Traditional PID control strategies, commonly used in industry, have been shown to be ineffective and unable to meet performance expectations. Control strategy design and multi-objective optimization for heavy-duty gas turbines can help them better align with grid load requirements and maintain safe and stable unit operation.

[0004] The extended state observer of the Error Active Disturbance Rejection Control (EADRC) directly receives the tracking error signal, making it highly applicable in practical engineering applications. It can effectively address system uncertainties and external disturbances, improving system stability and robustness. Furthermore, the Improved Hyperbolic Sine-Cosine Optimizer (ISCHO), a novel meta-heuristic algorithm, boasts strong evolutionary capabilities, fast search speed, and robust optimization capabilities. It can continuously iterate to find the optimal solution for controller parameters, improving control quality and achieving multi-objective optimization.

[0005] How to combine the anti-interference capability of EADRC with the optimization capability of ISCHO and implement this complex control system optimization method in actual heavy-duty gas turbine operation to meet the expected system performance and adapt to flexible operation requirements remains a challenging task. Summary of the Invention

[0006] In view of the above technical problems, the purpose of the present invention is to provide an intelligent control method for a heavy-duty gas turbine that integrates error self-disturbance rejection control and an improved hyperbolic sine-cosine optimizer, so as to achieve high-quality control and multi-objective operation optimization of the heavy-duty gas turbine, ensure the safe and stable operation of the unit under different operating conditions, and at the same time achieve rapid tracking of load instructions and economic improvement, so as to realize flexible and efficient operation of the unit.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer comprises the following steps:

[0009] Acquire and pre-process heavy-duty gas turbine data.

[0010] Export the heavy-duty gas turbine operating history data from the historical station, including the startup process, operating condition change, and shutdown process data. Use linear interpolation to fill in missing values:

[0011]

[0012] Where x t is the interpolation result at time t, and the Z-Score method is used to detect and process outliers in the data:

[0013]

[0014] In the formula, μ is the mean of the data set of the selected time and type, σ is the standard deviation, and the threshold is set. If Z i If the threshold is exceeded, then x i If it is an outlier, it will be replaced with data from adjacent time points to ensure the smoothness of the data.

[0015] To ensure the balance of the characteristics of each sensor, the Min-Max normalization method is used:

[0016]

[0017] A heavy-duty gas turbine Wiener model is established based on variational Bayesian identification.

[0018] Considering the characteristics of heavy-duty gas turbines, such as numerous monitoring parameters, strong nonlinearity, and high coupling, some data of continuous step disturbances in the fuel quantity during operation of heavy-duty gas turbines are selected, and the classic nonlinear Wiener model is used for approximation to obtain the transfer function Model between fuel quantity and speed. f-r .

[0019] The Wiener model is composed of a linear dynamic system and a static nonlinear element in series. The linear dynamic system uses a transfer function to describe the dynamic response of the heavy-duty gas turbine:

[0020]

[0021] G(q -1 ) is the transfer function model of the linear part of the heavy-duty gas turbine, which is given by:

[0022]

[0023] The transfer function model G(q -1 ) is converted into a finite impulse response model:

[0024]

[0025] Where L is the order of the FIR model. When L is large enough, the above formula can be converted to:

[0026]

[0027] For G(q -1 ) parameter identification can be converted into identification Θ=[θ0,θ1,θ2,…,θ L ] T .

[0028] Static nonlinear elements perform nonlinear transformation on the output signal of the linear system:

[0029] y(t)=f(x t )+e t

[0030] The nonlinear part can be written as a weighted sum of a set of known basis functions:

[0031]

[0032] f() is the basis function, and the above formula can be simplified to coefficient Λ=[λ0,λ1,…,λ M ] T Estimates.

[0033] Introducing the latent variable γ, we get the marginal likelihood function and the latent variable posterior distribution:

[0034] P(x;θ)=∫P(x,γ;θ)dy=∫P(x|γ;θ)P(γ;θ)dγ

[0035]

[0036] Using the variational Bayesian method, the joint distribution q(γ;θ) is introduced to approximate P(γ|x;θ):

[0037]

[0038] When the KL divergence reaches the minimum, the approximate solution q(γ;θ) is found, and the linear part parameter Θ of the heavy gas turbine is obtained. f-r and the nonlinear part parameter Λ f-r , complete the modeling.

[0039] Design an extended state observer and error active disturbance rejection control strategy.

[0040] Based on the heavy-duty gas turbine transfer function model, the corresponding error self-disturbance rejection control strategy is designed. Differentiating the output, the second-order dynamic system is obtained:

[0041]

[0042] In order to achieve reference tracking of the gas turbine rotor speed, the speed tracking error is defined as:

[0043] e n =n * -n

[0044] Its dynamic characteristics are expressed as follows:

[0045]

[0046] Where b0 represents the controller design parameter, f(t) represents the disturbance term, which is the input gain error (b0-b)u(t), internal dynamic uncertainty d(t), external disturbance Ideal nominal control input u * (t) and internal dynamics d * (t) function.

[0047] Add derivative feedback terms on both sides of the speed tracking error dynamic system:

[0048]

[0049] Where F(t) represents the total disturbance, and its value is estimated by the extended state observer ESO.

[0050] Assume that the total disturbance of the system is properly suppressed, that is Then the control law is proposed:

[0051]

[0052] Thus making

[0053] Define the state vector The dynamic model of speed tracking error can be rewritten as:

[0054]

[0055] The error-driven ESO can be constructed as follows:

[0056]

[0057] Where, represents the state vector of the ESO, is the estimated value of z, and L represents the gain vector of the ESO. The bandwidth of the extended state observer is defined as ω o , then det(sI-(A-LC))=(s+ω o ) 3 , thus we can infer For most industrial processes, ωo =(3~5)ω c ,ω c is the controller bandwidth. The controller is written in the form of a transfer function:

[0058]

[0059] Design multi-objective operation optimization objective function for heavy-duty gas turbines.

[0060] The design of the optimization objective function needs to fully consider the overall performance and operating characteristics of the gas turbine. The optimization objectives include the following key aspects: unit load tracking capability, unit operating stability, closed-loop regulation performance, comprehensive economic efficiency, and environmental friendliness.

[0061] Minimize the load following error and use its square integral as one of the optimization objectives:

[0062]

[0063] Introducing transition time T s , overshoot M p Etc., weight the response index of the closed-loop system:

[0064] J2=αT s +βM p

[0065] Define the fuel consumption rate as C f , the emission rate is E p , the economic and environmental optimization functions are:

[0066]

[0067] Where α, β and γ are weight coefficients.

[0068] By optimizing J1, J2 and J3, a set of non-dominated solutions is found to form the Pareto frontier, and the compromise point on the Pareto frontier is regarded as the optimal solution to the multi-objective problem. The solution is as follows:

[0069]

[0070] Where, represents the ideal optimal solution consisting of the optimal values ​​of the sub-objective functions, Represents a compromise solution, represents the optimal decision variable value.

[0071] The improved hyperbolic sine-cosine optimizer is used to solve the optimal solution of the decision variables of the heavy-duty gas turbine controller, and the Pareto frontier and Pareto optimal solution of the controller parameters are obtained, thereby achieving multi-objective operation optimization.

[0072] ISCHO starts by randomly initializing a set of feasible solutions:

[0073] X=rand(N,dim)×(ub-lb)+lb

[0074] Where X is the set of random feasible solutions, N is the number of feasible solutions, dim is the dimension of the problem, ub and lb are the upper and lower bounds of the variables, respectively.

[0075] First, the differential evolution algorithm is executed to perform mutation, crossover, and selection operations, which are described as follows:

[0076]

[0077] Where V i is the mutation vector of the i-th individual, and F is the scaling factor.

[0078]

[0079] Where U i is the trial vector of the ith individual, and CR is the crossover probability.

[0080]

[0081] Where, f(U i ) is the objective function value, f(X i ) is the objective function value of the current individual.

[0082] Then the elite non-dominated sorting method is used to estimate the Pareto optimal solution and calculate the crowding distance:

[0083]

[0084] Then enter the exploration phase, which is divided into two phases in the optimization process (iteration) and maintained in subsequent iterations to avoid local optimality. The value of switching between these two phases is determined by the following formula:

[0085]

[0086] Where Max-iteration is the maximum value of the iteration, ct is the coefficient for setting the two-phase switching point, and floor is the floor function.

[0087] The first phase exploration position update function is described as follows:

[0088]

[0089] Where t represents the current number of iterations, is the jth position of the optimal solution obtained so far, r1 / r2 is a random number in the interval [0,1], and W1 is the first exploration stage The weight coefficient of .

[0090] The second phase position update function is described as follows:

[0091]

[0092] In the formula, ε represents a small positive number, W2 is the The weight coefficient of .

[0093] Then it enters the development phase. The first development phase is described as follows:

[0094]

[0095] Where r7 / r8 are random numbers in the interval [0,1], and W3 is the weight coefficient of one-stage development.

[0096] The second phase position update function is described as follows:

[0097]

[0098] Where r 11 / r 12 is a random number in the interval [0,1], and W2 controls the degree of development in the second stage.

[0099] Then a bounded search strategy is executed to randomly initialize all feasible solutions in the new latent space. Each starting point of the strategy is calculated as follows:

[0100]

[0101] The above operations are performed in each iteration to determine the optimal individual in the population and output its position. The parameters u, m, n, p, q, Alpha, and Beta are dynamically adjusted and gradually reduced with the number of iterations t, allowing the algorithm to conduct a more aggressive search in the early stages and gradually focus on developing the optimal solution in the later stages.

[0102] If the current number of iterations t is equal to the maximum number of iterations Max-iteration, exit the optimization algorithm and output the Pareto frontier and the optimal solution. If the maximum number of iterations is not reached, update t and return to

[00082] .

[0103] At this point, the Pareto frontier of the heavy-duty gas turbine control system after multi-objective operation optimization is obtained, and the compromise solution is calculated to obtain the optimal solution of the control strategy variables.

[0104] The present invention has the following beneficial effects:

[0105] This invention combines an error-based active disturbance rejection control strategy with an improved hyperbolic sine-cosine optimizer to make heavy-duty gas turbine control more intelligent and adaptive. The error-based active disturbance rejection control strategy allows the system to be robust to external disturbances and internal changes. A swarm intelligence optimization algorithm allows the system to optimize control strategy parameters based on changing environmental conditions.

[0106] These effects together achieve high-quality control of heavy-duty gas turbines, help improve system performance, and adapt to flexible operating requirements. Therefore, the present invention has broad application prospects and economic benefits in the intelligent control of heavy-duty gas turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 Flowchart of the intelligent control method for heavy-duty gas turbines integrating error active disturbance rejection control and improved hyperbolic sine-cosine optimizer

[0108] Figure 2 ISCHO architecture diagram

[0109] Figure 3 Schematic diagram of EADRC-ISCHO control strategy DETAILED DESCRIPTION

[0110] In order to be able to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and Examples. It should be noted that, in the case of no conflict, the features in the embodiments of the present application and the embodiments can be combined with each other. In the following description, a lot of specific details are set forth to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the limitation of the specific embodiments of the following disclosure.

[0111] The implementation process is as attached Figure 1 shown.

[0112] Step 1. Collect heavy-duty gas turbine data and preprocess it:

[0113] The heavy-duty gas turbine operation history data is exported from the history station, including the startup process, operating condition change, and shutdown process data. The sampling frequency is one second and the sampling time is one week.

[0114] The linear interpolation method is used to fill missing values ​​in the collected long-term data:

[0115]

[0116] Where x t is the interpolation result at time t, and the Z-Score method is used to detect and process outliers in the data:

[0117] Use the Z-Score method to detect and process outliers in the data:

[0118]

[0119] Where μ is the mean of the data set of the selected time and type, σ is the standard deviation, and the threshold is set to 3 or -3. If Z i If the threshold is exceeded, then x i are outliers and are replaced with data from adjacent time points.

[0120] To ensure the balance of the characteristics of each sensor, Min-Max normalization is used:

[0121]

[0122] Step 2. Modeling of heavy-duty gas turbine Wiener model based on variational Bayesian identification:

[0123] Select some data of continuous step disturbance of fuel quantity during operation of heavy-duty gas turbine, use classic nonlinear Wiener model to approximate, and obtain the transfer function Model of fuel quantity-speed f-r .

[0124] The Wiener model is composed of a linear dynamic system and a static nonlinear element in series. The linear dynamic system uses a transfer function to describe the dynamic response of the heavy-duty gas turbine system:

[0125]

[0126] G(q -1 ) is the transfer function model of the linear part of the heavy-duty gas turbine, which is given by:

[0127]

[0128] The transfer function model G(q -1 ) is converted into a finite impulse response model:

[0129]

[0130] Where L is the order of the FIR model. When L is large enough, the above formula can be converted to:

[0131]

[0132] For G(q -1 ) parameter identification is converted into identification Θ=[θ0,θ1,θ2,…,θ L T .

[0133] Static nonlinear elements perform nonlinear transformation on the output signal of the linear system:

[0134] y(t)=f(x t )+e t

[0135] The nonlinear part can be written as a weighted sum of a set of known basis functions:

[0136]

[0137] f() is the basis function, which is simplified to coefficients Λ=[λ0,λ1,…,λ M ] T Estimates.

[0138] Introducing the latent variable γ, we get the marginal likelihood function and the latent variable posterior distribution:

[0139] P(x;θ)=∫P(x,γ;θ)dy=∫P(x|γ;θ)P(γ;θ)dγ

[0140]

[0141] Using the variational Bayesian method, the joint distribution q(γ;θ) is introduced to approximate P(γ|x;θ):

[0142]

[0143] When the KL divergence is minimized, the approximate solution q(γ;θ) is found, and the linear part parameter Θ of the heavy gas turbine is obtained. f-r and the nonlinear part parameter Λ f-r , complete the modeling.

[0144] Step 3. Design the extended state observer and error active disturbance rejection control strategy:

[0145] Based on the heavy-duty gas turbine transfer function model in step 2, the corresponding error self-disturbance rejection control strategy is designed. Differentiate the output to obtain the second-order dynamic system:

[0146]

[0147] In order to achieve reference tracking of the gas turbine rotor speed, the speed tracking error is defined as:

[0148] e n =n * -n

[0149] Its dynamic characteristics are expressed as follows:

[0150]

[0151] Where b0 represents the controller design parameter, f(t) represents the disturbance term, which is the input gain error (b0-b)u(t), internal dynamic uncertainty d(t), external disturbance Ideal nominal control input u * (t) and internal dynamics d * (t) function.

[0152] Add derivative feedback terms on both sides of the speed tracking error dynamic system:

[0153]

[0154] Where F(t) represents the total disturbance, and its value is estimated by ESO. Then the control law is proposed:

[0155]

[0156] Thus making

[0157] Define the state vector The dynamic model of speed tracking error can be rewritten as:

[0158]

[0159] The error-driven ESO can be constructed as follows:

[0160]

[0161] Where, represents the state vector of the ESO, is the estimated value of z, and L represents the gain vector of the ESO. The bandwidth of the extended state observer is defined as ω o , then det(sI-(A-LC))=(s+ω o ) 3 , thus we can deduce l1=3ω o ,l2=3ω o 2 -l1k1,l3=ω o 3 For most industrial processes, ω o =(3~5)ω c ,ω c is the controller bandwidth. The controller is written in the form of a transfer function:

[0162]

[0163] Step 4. Design the multi-objective operation optimization objective function of the heavy-duty gas turbine:

[0164] The optimization objectives include the following key aspects: unit load tracking capability, unit operating condition stability, closed-loop regulation performance, comprehensive economy and environmental friendliness.

[0165] Minimize the load tracking error and take the ITAE of the error as one of the optimization objectives:

[0166]

[0167] Introducing transition time T s , overshoot M p Etc., weight the response index of the closed-loop system:

[0168] J2=αT s +βM p

[0169] Define the fuel consumption rate as C f , the emission rate is E p , the optimization function is:

[0170]

[0171] Where α, β and γ are weight coefficients.

[0172] Optimize J1, J2 and J3 to find a set of non-dominated solutions to form the Pareto frontier, and regard the compromise point on the Pareto frontier as the optimal solution to the multi-objective problem

[0173]

[0174] Where, represents the ideal optimal solution consisting of the optimal values ​​of the sub-objective functions, Represents a compromise solution, represents the optimal decision variable value.

[0175] Step 5. Use the improved hyperbolic sine-cosine optimizer to find the optimal solution for the heavy-duty gas turbine controller decision variables, as shown in the attached figure. Figure 3 As shown, the Pareto frontier and Pareto optimal solution of the controller parameters are obtained, thereby achieving multi-objective operation optimization:

[0176] ISCHO starts by randomly initializing a set of feasible solutions, as shown in the following figure. Figure 2 As shown:

[0177] X=rand(N,dim)×(ub-lb)+lb

[0178] Where X is the set of random feasible solutions, N is the number of feasible solutions, dim is the dimension of the problem, ub and lb are the upper and lower bounds of the variables, respectively.

[0179] First, the differential evolution algorithm is executed to perform crossover, mutation, and selection operations, and then the hyperbolic sine and cosine algorithm is searched and developed, as described below:

[0180]

[0181] Where V i is the mutation vector of the i-th individual, and F is the scaling factor.

[0182]

[0183] Where U i is the trial vector of the ith individual, and CR is the crossover probability.

[0184]

[0185] Where, f(U i ) is the objective function value, f(X i ) is the objective function value of the current individual.

[0186] The elite non-dominated sorting method is used to estimate the Pareto optimal solution and calculate the crowding distance:

[0187]

[0188] Then enter the exploration phase, which is divided into two phases in the optimization process (iteration) and maintained in subsequent iterations to avoid local optimality. The value of switching between these two phases is determined by the following formula:

[0189]

[0190] Where Max-iteration is the maximum value of the iteration, and ct is the coefficient for setting the two-phase switching point.

[0191] The first phase exploration position update function is described as follows:

[0192]

[0193] Where t represents the current iteration number. is the jth position of the optimal solution obtained so far, r1 / r2 is a random number in the interval [0,1]. W1 is the first exploration stage The weight coefficient of .

[0194] The second phase position update function is described as follows:

[0195]

[0196] In the formula, ε represents a small positive number, W2 is the The weight coefficient of .

[0197] In the first development phase, the adjacent space is developed as follows:

[0198]

[0199] Where r7 / r8 are random numbers in the interval [0,1], and W3 is the weight coefficient of one-stage development.

[0200] The position update function in the second stage of development is described as follows:

[0201]

[0202] Where r 11 / r 12 is a random number in the interval [0,1], and W2 controls the degree of development in the second stage.

[0203] Then, a bounded search strategy is executed. In order to fully utilize the potential search space, ISCHO randomly initializes all feasible solutions in the potential space. Each starting point of the strategy is calculated as follows:

[0204]

[0205] The above operations are performed in each iteration to determine the optimal individual in the population and output its position. The parameters u, m, n, p, q, Alpha, and Beta are dynamically adjusted and gradually reduced with the number of iterations t, allowing the algorithm to conduct a more aggressive search in the early stages and gradually focus on developing the optimal solution in the later stages.

[0206] If the current number of iterations t is equal to the maximum number of iterations Max-iteration, exit the optimization algorithm and output the Pareto frontier and the optimal solution. If the maximum number of iterations is not reached, update t and return to

[00186] .

[0207] At this point, the Pareto frontier of the heavy-duty gas turbine speed control system after multi-objective operation optimization is obtained, and the compromise solution is calculated to obtain the optimal solution of the control strategy variables.

[0208] It should be understood that the exemplary embodiments described herein are illustrative and not restrictive. Although one or more embodiments of the present invention have been described in conjunction with the accompanying drawings, it should be understood by those skilled in the art that various changes in form and details may be made without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims

1. A heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer, characterized in that: Follow the steps below to implement it: Step 1: Collect heavy-duty gas turbine data and preprocess them; Step 2: Establish a heavy-duty gas turbine Wiener model based on variational Bayesian identification; Step 3: Design the extended state observer and error self-disturbance rejection control strategy; Step 4: Design the multi-objective operation optimization objective function of the heavy-duty gas turbine; Step 5: Use the improved hyperbolic sine-cosine optimizer to solve the optimal solution of the decision variables of the heavy-duty gas turbine controller.

2. According to claim 1, a heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer is characterized in that: The step 1 specifically involves collecting heavy-duty gas turbine data and preprocessing it, including the following steps: Step 1: Export the heavy-duty gas turbine operation history data from the history station, including the startup process, variable operating conditions, and shutdown process data. Step 2: Preprocess the data and use linear interpolation to fill missing values: In the formula, x t is the interpolation result at time t. Use the Z-Score method to detect and process outliers in the data: In the formula, μ is the mean of the data set of the selected time and type, σ is the standard deviation, and the threshold is set. If Z i If the threshold is exceeded, then x i is an outlier, and it is replaced with data from adjacent time points to ensure the smoothness of the data. To ensure the balance of the characteristics of each sensor, the Min-Max normalization method is used:

3. A heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer according to claims 1 to 2, characterized in that: The step 2 specifically establishes a heavy-duty gas turbine Wiener model based on variational Bayesian identification, including the following steps: Step 1: Considering the characteristics of heavy-duty gas turbines, which have many monitoring parameters, strong nonlinearity, and high coupling, some data of continuous step disturbances in the fuel quantity during operation of heavy-duty gas turbines are selected, and the classic nonlinear Wiener model is used for approximation to obtain the transfer function Model between fuel quantity and speed. f-r . The Wiener model is composed of a linear dynamic system and a static nonlinear element in series. The linear dynamic system uses a transfer function to describe the dynamic response of the heavy-duty gas turbine: G(q -1 ) is the transfer function model of the linear part of the heavy-duty gas turbine, which is given by: The transfer function model G(q -1 ) is transformed into a finite impulse response model: Where L is the order of the FIR model. When L is large enough, the above formula can be converted to: For G(q -1 ) can be converted into identifying Θ=[θ0,θ1,θ2,…,θ L ] T . Static nonlinear elements perform nonlinear transformation on the output signal of the linear system: y(t)=f(x t )+e t The nonlinear part can be written as a weighted sum of a set of known basis functions: f() is the basis function, and the above formula can be simplified to coefficient Λ=[λ0,λ1,…,λ M ] T Estimates. Step 2: Introduce the latent variable γ to obtain the marginal likelihood function and the latent variable posterior distribution: P(x;θ)=∫P(x,γ;θ)dy=∫P(x|γ;θ)P(γ;θ)dγ Using the variational Bayesian method, the joint distribution q(γ; θ) is introduced to approximate P(γ|x; θ): When the KL divergence reaches the minimum, the approximate solution q(γ;θ) is found, and the linear part parameter θ of the heavy gas turbine is obtained. f-r and the nonlinear part parameter Λ f-r , the modeling is completed.

4. A heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer according to claims 1 to 3, characterized in that: The step 3 specifically designs an extended state observer and an error auto-disturbance rejection control strategy, including the following steps: Step 1: Based on the heavy-duty gas turbine transfer function model, design the corresponding error self-disturbance rejection control strategy. Differentiate the output to obtain the second-order dynamic system: In order to achieve reference tracking of the gas turbine rotor speed, the speed tracking error is defined as: e n =n * -n Its dynamic characteristics are expressed as follows: Where b0 represents the controller design parameter, f(t) represents the disturbance term, which is the input gain error (b0-b)u(t), internal dynamic uncertainty d(t), external disturbance Ideal nominal control input u * (t) and internal dynamics d * (t). Add derivative feedback terms on both sides of the speed tracking error dynamic system: Where F(t) represents the total disturbance, and its value is estimated by the extended state observer ESO. Step 2: Assume that the total disturbance of the system is properly suppressed, that is, Then the control law is proposed: So that Define the state vector The dynamic model of speed tracking error can be rewritten as: The error-driven ESO can be constructed as follows: In the formula, represents the state vector of ESO, is the estimated value of z, and L represents the gain vector of ESO. The bandwidth of the extended state observer is defined as ω o , then det(sI-(A-LC))=(s+ω o ) 3 , so we can deduce l1=3ω o , For most industrial processes, ω o =(3~5)ω c ,ω c is the controller bandwidth. The controller is written in the form of transfer function:

5. A heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer according to claims 1 to 4, characterized in that: The step 4 specifically designs a multi-objective operation optimization objective function of a heavy-duty gas turbine, including the following steps: Step 1: The design of the optimization objective function needs to fully consider the overall performance and operating characteristics of the gas turbine. The optimization objectives include the following key aspects: unit load tracking capability, unit operating condition stability, closed-loop regulation performance, comprehensive economy and environmental friendliness. Minimize the load following error and take its square integral as one of the optimization objectives: Introduce transition time T s , overshoot M p Etc., weight the response index of the closed-loop system: J2=αT s +βM p Define the fuel consumption rate as C f , the emission rate is E p , the economic and environmental optimization functions are: Where α, β and γ are weight coefficients. Step 2: By optimizing J1, J2 and J3, a set of non-dominated solutions is found to form the Pareto frontier, and the compromise point on the Pareto frontier is regarded as the optimal solution to the multi-objective problem. The solution is as follows: In the formula, represents the ideal optimal solution consisting of the optimal values ​​of each sub-objective function, Represents a compromise solution, represents the optimal decision variable value.

6. A heavy-duty gas turbine intelligent control method integrating error auto-disturbance rejection control and improved hyperbolic sine-cosine optimizer according to claims 1 to 5, characterized in that: The step 5 specifically uses an improved hyperbolic sine-cosine optimizer to solve the optimal solution of the decision variables of the heavy-duty gas turbine controller, including the following steps: Step 1: Use the improved hyperbolic sine-cosine optimizer to solve the optimal solution of the decision variables of the heavy-duty gas turbine controller, obtain the Pareto frontier surface and Pareto optimal solution of the controller parameters, and thus achieve multi-objective operation optimization. ISCHO starts by randomly initializing a set of feasible solutions: X=rand(N,dim)×(ub-lb)+lb Where X is the set of random feasible solutions, N is the number of feasible solutions, dim is the dimension of the problem, ub and lb are the upper and lower bounds of the variables, respectively. Step 2: First, execute the differential evolution algorithm to perform mutation, crossover and selection operations, as described below: Where V i is the mutation vector of the ith individual, and F is the scaling factor. Where U i is the trial vector of the ith individual, and CR is the crossover probability. In the formula, f(U i ) is the objective function value, f(X i ) is the objective function value of the current individual. Step 3: Use the elite non-dominated sorting method to estimate the Pareto optimal solution and calculate the crowding distance: Step 4: Enter the exploration phase, which is divided into two phases in the optimization process (iteration) and maintained in subsequent iterations to avoid local optimality. The value of switching between these two phases is determined by the following formula: Where Max-iteration is the maximum value of the iteration, ct is the coefficient for setting the two-phase switching point, and floor is the rounding function. The first phase exploration position update function is described as follows: In the formula, t represents the current iteration number, is the jth position of the optimal solution obtained so far, r1 / r2 is a random number in the interval [0,1], and W1 is the first exploration stage The weight coefficient of . The second phase position update function is described as follows: In the formula, ε represents a small positive number, W2 is the The weight coefficient of . Step 5: Enter the development phase. The first development phase is described as follows: Where r7 / r8 is a random number in the interval [0,1], and W3 is the weight coefficient of the first-stage development. The second phase position update function is described as follows: In the formula, r 11 / r 12 is a random number in the interval [0,1], and W2 controls the degree of development in the second stage. Step 6: Execute a bounded search strategy to randomly initialize all feasible solutions in the new latent space. Each starting point of the strategy is calculated as follows: The above operations are performed in each iteration to determine the best individual in the population and output its position. At the same time, the parameters u, m, n, p, q, Alpha and Beta are dynamically adjusted with the number of iterations t, gradually decreasing, so that the algorithm can perform a more aggressive search in the early stage and gradually focus on developing the optimal solution in the later stage. Step 7: If the current number of iterations t is equal to the maximum number of iterations Max-iteration, exit the optimization algorithm, output the Pareto frontier and the optimal solution, if the maximum number of iterations is not reached, update t and return to step 3. At this point, the Pareto frontier of the heavy-duty gas turbine intelligent control system after multi-objective operation optimization is obtained, and the compromise solution is calculated to obtain the optimal solution of the control strategy variables.

Images